e m Docu -- SJ W a i v s nt UNITED STATES OF AMERICA FEDERAL TRADE COMMISSION WASHINGTON, D.C. 20580 Project Code: DXI MEMORANDUM TO: FROM: DATE: August 8, 2012 SUBJECT: Googlc Inc. File No. 111-0163 1 RECOMMENDATION: That the Commission Issue the Attached Complai t SJ W a i v s nt me u c o D natural and probable effect ofGoogle's conduct is to diminish the incentives of vertical websites to invest in, and to develop, new and innovative content. Jn the alternative, Googlc's conduct may be condemned as a stand-alone violation of Section 5. Google has presented no efficiency justifi cation for its conduct. Third, Staff has investigated whether Google has employed ru1ticompetifve contractual restrictions on the a· utomated cross-management of advertising camp,laigns. . Google's main rival (Microsoft) has alleged that Googlc is denying Microsoft c ·tical scale by employing these restrictions, and thus impairing Microsoft's ability to compe e effectively in the markets for general search and search advertising. We conclude that thes restrictions should be condemned under Section 2 because they limit the ability of advertisers to make use of their own data, and as such, have reduced innovation and increased transaction costs among advertisers and third-party businesses, and also degraded the quality ofGoogle's rivals in search and search advertising. Google 's proffered efficiency justification for these restrictions appears to be pretextual. Fourth, Staff has investigated whether Google has entered into anticomprtive, exclusionary agreements with websites for syndicated search and search advcrtisrng services. We conclude that Google's agreements should be condemned under Section 2 because they foreclose some p01tion of the market, and, although the agreements result in only modest anticompetitive effects on publishers, the impact of the agreements in denying scale to competitors is both competitively significant to its main rival (Microsoft) today, as well as a significant barrier to entry for potential entrants in the longer term. While Googe presents efficiency justifications for these agreements, on balance, Staff finds them to be nonpersuasive. Il l SJ W a i v s nt e m Docu TABLE OF CONTENTS I. II. HISTORY OF THE ~VESTIGATIO~ AND RELATED PROCEED~GS ... -1- A. FTC Investigation ...•...............................................................-1- B. European Commission Investigation ...........................................-2- C. Multi-State Investigation ........................... .............................. -3- 0. Private Litigation .... , ...............................................................-3- STATEMENT OF FACTS ................................................................-4A. The Parties .............................................................. -4- 1. Coogle ........................................................... , .............-4- 2. General Search Competitors .............................................-5- 3. 4. B. ........ . ... 1 a. Microsoft ............................................................-5- b. Yahoo! ...............................................................-5- Major Vertical Complainants ...........................................-6- a. Amazon .................................................... .........-6- b. eBay ..................................................................-.6- c. NexTag .............................................................-6- d. Foundem ...........................................................-6- e. Expedia ................................... .........................-7- f. TripAdvisor ........................................................-7- g. Yelp . .. . .. . ............. . ....... . . .....................................-7- Facebook ........................................................ ............-7- Industry Background ............................. · · · .. · .. · · · · · · .. · .. · 1. T.... ·.......-8 General Search ....................................... ........ ·~· ...........-8- v SJ W a i v s nt e m Docu a. Overview of the AdWords Platform .........................-41- b. The Restrictive Conditions ............................•....... -43- c. Effects of the Restrictive Conditions ............... .........-44i. Effects on Advertisers and Search Engi e Marketers ("SEMs") .................................-44- ii. Effects on Competitors .............. .................-46- d. 4. Ill. Internal Googlc Discussions Regarding the Restrictions .......................................................-48- Google's Exclusive and Restrictive Syndication Agreements ....................................................................................-51a. Publishers and Market Structure ............................-52- b. Development of the .Market for Search Syndication ......-53- c. Specifics of Coogle's Syndication Agreements ............-53- d. Effects of Exclusivity and Preferred Placement ...........-55i. Common Publisher Responses .......................-55- ii. Publishers' Views of Exclusivity Provisi ns ...... -58- iii. Effects on Competitors .................................-63- LEGAL ANALYSIS .................. ... ........................ ..........................-63A. Google Has Monopoly Power in Relevl:lnt Markets ........................-64L. 2. Relevant Markets and Market Shares ........................... ....-64a. Horizontal Search ...............................................-64- b. Search Advertising ..............................................-69- c. Syndicated Search and Search Advertising ("Search Intermediation'') ......................... r... . ··. -73- Substantial Barriers to Entry Exist ...................................-75- vii SJ W a i v s nt e m Docu 4. IV. Coogle's Exclusive and Restrictive Syndication Agreements ..................................................................................-1 02- I a. Google's Agreements Foreclose a Substantial Portion of tbe Relevant Market .............................-1 02- b. Goog1e's Agreements Have Resulted In Anticompetitive Effects ...................................... -105- c. Coogle's Agreements Are Not .J usti fied B)' E fficienci cs .... ............ , . . . . . . . . . •. . •. •. . . . . . •. . .•... .•..-11 0- POTENTIAL REMEDIES ............................................. ....... . ........-112A. Scraping ................................................................. , ..........-112- B. API Restrictions ..................................................................-113- C. Exclusive and Restrictive Syndication Agreements .......................-113- V. LITIGATION RISKS ....................................................................-114- VI. CONCLUSION ........................ .................... .... ... ............. ............-116APPENDICES APPENDIX 1: Table of Exclusive Agreements ..................................................-1APPENDIX 2: Table of Preferred Placement Agreements ...................................-!AJ>PENDIX 3: Explanation of Data Sources ...................................................-1APPENDIX 4: Glossary of Terms ................................................................-1- IX SJ W a i v s nt e m Docu B. EUROPEAN COMMISSION INVESTIGATION The European Commission ("EC") has been conducting a parallel investigation of Google since November 2010. On May 21, 2012, Commissioner Joaquin Aim ·a issued Googlc a letter, signaling the EC's possible intent to issue a Statement ofObjec ions ("SO") against Google for abuse of dominance in violation of Article I 02 of the EC Tr aty. The letter set out the EC's concern in four areas: (I) Google's "favourable treatmen of its own vettical search services as compared to those of its competitors in its nah1ral sea ch results"; (2) Google's "practice of copying third party content" to supplement its own ve ical ofTcrings; (3) Google's "exclusivity agreements with publishers for the provisio of search advertising intem1ediation services"; and (4) Google' s " restrictions with regard to the portability and cross-platform management of online advertising campaigns.'.4 In his letter, Commissioner Almunia offered Googlc the opportunity to resolve the concerns prior to the issuance of an SO by coming forward "with a written description of possible solutions" to the EC's eoncems. 5 On June 30, 2012, Google submitted a settlement proposal to the EC. Although Googlc denied any infringement of European Union ("EU") competition law, Google proposed to enter into several "commitments," designed to address the EC's sta 'e d concerns.6 FTC staff has coordinated closely with EC staff throughout the course o our parallel investigations. Staffhas received waivers from Googlc, Microsoft, Yahoo! , an a handful of other parties to discuss and exchange infonnation with the EC. Staff has had regular telephone calls with EC staff, where we have updated one another on theories We have also exchanged documents of mutual interest. 2 d evidence. SJ W a i v s nt me u c o D Google to tott liability. 13 The Kinderstart court also dismissed the plaintift"s complaint, rejecting the claim that Google's search results were an essential facility for vertical websites, because Kindcrstart had not been e liminated from the downstream mlket and continued to get high rankings from other search cngincs. 14 The AdWords cases address a common fact pattern, but are decided on grounds. Plaintiffs in these cases argued that Google increased the minimum bi s for the keywords the website had purchased, which made those keywords effectively u available, thus depriving the plaintiff website oftrat1ic. The complaint in TradeComet.co , LLC v. Coogle, inc. 15 was dismissed for improper venue, while the allegations in Goog e, Inc. v. myTriggers. com, inc. 16 were dismissed on grounds that they failed to describe harm to competition as a whole. Both cases were dismissed with little discussion of the merits. In Person v. Coogle, Inc., 11 Judge Fogel of the Northern District of California criticized the plaintiffs market definition, finding no basis for distinguishing th alleged "search advertising market" fro m the larger market for Internet advcrtising. 18 The Ninth Circuit affirm ed the conclusion that the plaintiff failed to plead facts sufficient to raise the allegations in its complaint beyond a speculative level, but did not address mar t dcfinition. 19 II. STATEMENT OF FACTS A. THE P ARTTES 1. Google Googlc is an Internet search technology company, founded in 1998 and eadquartered in Mountain View, California. Googlc's products and services include a genera "horizontal" search engine, as well as numerous integrated "vertical'' websites that focus on specific 4 SJ W a i v s nt e m Docu 3. Major Vertical Complainants Staff has met with, interviewed, and subpoenaed numerous vertical websites offering shopping, travel, local, and financial services. We identify here some of the rna complainants. ln general, these companies complain tbat Google's practice of referencing its own vertical results over the complainants' websites on Google 's search pag has negatively impacted the complainants' ability to compete for users and advertis rs. a. Amazon Amazon is the world's Largest online retailer, and also produces consum r electronics, notably the Amazon Kindle e-book reader and the Kindle Fire tablet. Amazon's product search feature competes with Google Product Search. b. eBay eBay operates an online auction and shopping website in which people and businesses buy and sell a broad variety of goods and services worldwide. eBay has expan*d from its origina l "set-time" auction format to include "Buy It Now" standard shopping, nd a variety of other services. eBay's product search feature competes with Google Product Search. c. NexTag NexTag is a shopping comparison website in the U.S. that competes wi Google Product Search. d. Foundcm Foundem is a shopping comparison website in the United Kingdom that competes with Googlc Product Search. We understand that Foundem was the first vertica website to publicly accuse Google ofpreferencing its own vertical content over that of co petitors on 6 SJ W a i v s nt me u c o D recently introduced social networking site, Google Plus. Facebook has complained, among other things, that Google's prcfercncing of Google Plus results over Facebook results on Google's search page is negatively impacting its ability to compete for users. B. INDUSTRY BACKGROUND 1. General Search The Internet is a vast, largely unorganized collection of constantly chan ing information. If the Internet can be roughly analogized to a huge and highly dyn mic library, then algorithmic search engines are the card cata log. Unlike a traditional library, the Internet is too large and changes too rapidly for traditional cataloging. 32 Instead, search engines (like Google) deploy computer programs that constantly "crawl" the web, building and updating automated indexes of web content. Similarly, the process of finding relevant information inside these web indexes is automated. Sophisticated algorithms evaluate the content of the end user's request for information to determine which parts of the web index may contain relevant responses. The identified potential responses arc then ranked by additional algorithms based on the predicted likelihood of their relevance, and displayed to the end user in response to his or her query. Critically, all of this complex activity occurs rapidly and automatically, without any direct human intervention. As users search for information on the Internet, they necessarily provide the search engine with valuable information - the precise topic users arc interested in at that moment. Although a user does not pay for the web search service, the user's focused inte est - or intent - is very valuable to advertisers, because users arc e ffectively identifying hcmselves as potential customers through the content of their queries. For example, a business selling 8 SJ W a i v s nt me u c o D advcrtising:l 6 The growth of the Internet has created entirely new business models that can take advantage of ways, unique to the Internet, to identify and reach potential customers with advertising. 37 Among the reasons advertisers have shifted budget online is the l.gh degree of tracking possible and the quantifiable, superior return on investment. 38 Online advertising is primarily made up of display and search advertisin , although some other types of advertising (e.g., contextual , re-targeted display, and social edia advertising) also have some presence. Display advertising typically consists of containing graphics and other rich media appearing on white space on a web pa e. Search advertising consists of text ads (displayed on the right-hand side of the search rasults page, at the top of the page above the search results, and below the search results) matched to specific keyword queries entered into the search engine by the user. Search advertising makes up the bulk of online advertiser spend, primar' ly because advertisers believe that search advertising provides unprecedented precision in i entifying potential customers, measurabi lity, and the highest return on investment. 39 Simr ly put, "it is the most effective marketing ever."40 Search advertising is highly valued by ad ertisers because they learn crucial information about the user from the query alone: the] learn that the user is interested in a particular subject, right now.41 Thus, search adve11isi g is a highly effective method of reaching users who arc interested in learning about or purc1asing products. Search advertising is often called "direct response•· advertising, as it ' is intended to elicit a response from a consumer, such as the purchase of a product or signing up for a service.'A2 With pure djsplay advertising, all the advertiser knows about the user is hat he or she is viewing a particular web page (similar to the information an advertiser may h ve about a 10 SJ W a i v s nt me u c o D the user leave to go to a dedicated search website like Google or Bing; the search provider picks up incremental search volume, as some users will not bother to run a search if they I bave to leave the publisher's website to do it; and, most importantly, the resulti g search traffic can be monetized through search advertising in the same way as a search run on Google or Bing. The process works very similarly to a web search conducted on Google. Google receives queries from the third-party website, evaluates them against a subset o its web index, and then de livers web search results to the user on the third-party publisl er's website. 49 As with web search on Google.com, the consumer pays for none oft ese services. Instead, pub I is hers pay Google for syndicated search either on a cost-per-user-query basis (for example, $.95 per 1,000 queries), or by accepting search advertisements from Google and splitting the revenues from the se-arch advertisements run on the publisher's website. The resulting revenue sharing arrangement is often referred to as the "traffic ac~uisition cost" (or ''TAC"). Publishers arc generally able to select the web search and search advertisement syndication services separately or together. Thus, publishers that do not wish to offer web search generally (or Google's web search, specifically) can- and do- participa e in Google's AdScnsc program to receive search advertisements without the corres onding web search functionality.50 4. Mobile Searcb 51 In recent years, the focus of search (and related advertising) has begun s jfting from the traditional desktop model to the rapidly emerging- and lucrative- frontier f mobile (or "smartphone") devices. At the forefront of this shift is Google's mobile operating system, 12 SJ W a i v s nt e m Docu In addition, click data (the website links on which a user actually clicks) is important for evaluating the quality of the search results page. As Google' s former chief of search quality Udi Manber testified: The ranking itself is affected by the click data. If we discover th t, for a particular query, hypothetically, 80 percent or people click on R ult No. 2 and only l 0 percent click on Result No. I, after a while we fi~ur probably Result 2 is the one people want. So we'll switch it. Testimony from Scrgcy Brin and Eric Schmidt confirms that click data is impo nt for many purposes, including, most importantly, providing ''feedback" on whether Googl 's search algorithms are offering its users high quality results. 58 Finally, search providers run experiments on large volumes of users. Search engines conduct experiments on everything from ranking of search results to user interface and design decisions. 59 As Larry Page and Sergey Brin stated in their 2005 annual letter to shareholders: Our teams are more productive once they get real users and feed ack. We have learned that the best way to make something great is to actu lly launch it to the public. T hat's why we have the Google Labs and 'beta' la els- these arc our experiments. 60 Multiple experiments are conducted simullaneously. 61 The more search users l~ere are at any given time, the more experiments can be run, the faster they can be completed, td the more improvements that can be made to the search algorithms. 62 According to Micro oft chief economist (and Harvard professor) Susan Athey, Microsoft's search quality tea 1 is greatly hampered by having insufficient search volume to conduct expcrimcnts.63 With improved search quality, particularly for ·'tail" queries, Bing assert that it will be better positioned to compete with Google for users (and, thus, for advertisers , and so to 14 SJ W a i v s nt me u c o D and - importantly - also serves to attract more advertisers, that, generally spe the ir advertisements to reach as broad an audience as possible. 74 Jn sum, Bing asserts that a larger volume of advertisements- and the im roved coverage, quality, conversion rates, and revenues that come from such an increa 'ed volumewi ll allow it to better compete with Google for both advertisers and website pub ishers, and so to constrain the exercise by Google of monopoly power. 3. The Scale C urve Googlc acknowledges the importance of scale in the abstract. Google documents are replete with references to the "virtuous cycle" among users, advertisers, and publishers; 75 and testimony from Google executives confirms the continuing viability of the "cycle." 76 However, Google argues that, while scale matters, it only matters up to a point, beyond which there arc substantially "diminishing returns" to increasing volumes of both queries and advertisements. 77 For example, Sergey Brio testified that a ''rough rule of thumb" might be, as query volume doubles, a search engine might expect to see a one percent incr ase in quality.7l! Googl.e argues that Bing's query and advertiser volume have passed the oint at which scale should - or would - matter significantly to Microsofi, and that any volume gains I made by Bing would yield minimal improvements in either Bing's search quality or its monetization ability. 79 Microsoft does not dispute the notion that there are generally diminishing returns to scale. 80 The main bone of contention between Google and Microsoft is where on this scale curve Microsoft currently operates. This is an important question, but I one which evades easy answers. This is, in part, because neither party can identify a fixed number of queries or ads that constitutes the " minimum efficient" point of 16 on. SJ W a i v s nt me u c o D platform, it calculated that Bing would receive a 20 percent boost in revenue per search (''R PS") on the basis ofYahoo!'s additional volume.89 Although Bing's RPS (and consequently, RPM) has improved with the addition of Yahoo! query and ads v lume, it has not improved as substantially as Microsoft initially forccast.90 In this investigation, the question of how and why scale matters has tak n a prominent position in several allegations advanced by complainants: speciticall , whether the conduct under review denies Googlc's main competitor - Microsoft- the scale t needs to successfully constrain Googlc's monopoly over search and search advertising. hese allegations arc discussed in detail in the following section. D. GOOGLE'S SUSPECT CONDUCT Staff has conducted a comprehensive investigation into several areas of alleged anticompctitive conduct. Below, we lay out four of the five main areas of Staff's investigation. 91 1. Google's Preferencing of Googlc Vertical Properties 'fitbin Its Search Engine Results Page ("SERP") 1 Staff has investigated whether Google is unlawfully preferencing its own vertical properties, wh ile demoting rival vertical properties. in order to maintain, preser e, or enhance Googlc's monopoly power in the markets for search and search advertising. Co plainants allege that Google's conduct is anticompetitivc because it forecloses alternative search platforms that might operate to constrain Google's dominance in search and sea ch advertising. Although it is a close call, we do not recommend that the Commis ·an issue a complaint against Googlc for this conduct. 18 SJ W a i v s nt me u c o D search capabilities within specific commercial categories, and thus might cause users to shift their searches in those categories away from Google's general web search platfi rm. As users moved to vertical search websites, those websites could, in tum, become more ttractive vehicles for advertisers, thus resulting in potentially significant revenue losses t Google. In sh011: Vertical search is of tremendous strategic importance to Google. the Jisk is that Google is the go-to place for finding information cases where there is sufficiently low monetization potential that vertical search competitor has fi lled the space with a better alter Otherwise nly in the o niche 101 ative. A 2008 presentation, entitled "Online Advertising Challenges: Rise oft e Aggregators," further highlights the problems faced by Google with regard to the leading UK-based finance vertical website, MoneySupem1arkct: Issue 1. Consumers migrating to MoneySupermarket. Driver: General search engines not solving consumer queries as well as specialized vertical search.... Consequence: Increasing proportion of visitors going directly to MoneySupermarket. . . . Google Implication: Loss of query volumes. I Issue 2: MoneySupermarket has better advertiser proposition. Dp ver: MoneySupermarkct offers cheaper, lower risk (CPA-based) lead to advertisers. Google Implication: Advertiser pull: Direct advertis switch spend to MoneySupcrmarket/othcr channels. 102 Pa1tly in response to this new competitive threat - the "rise of aggrcgato " - Google decided to hone in on certain "key" ve1tical search areas (shopping, local, finan e, and travel) and invest in developing existing - or creating new- vertical properties. 103 In e rtain areas where Google already bad existing vertical properties, such as shopping and local, 104 Googlc saw a critical need to invest further and take measures to increase user traffic to properties. 105 In potentially lucrative areas where strong verticals already exi 20 SJ W a i v s nt me u c o D and against its natural search results, because the web index and other indices all bad their own ranking algorithms and scores. 114 Google referred to the difficulty of comparing these ranking scores as an "apples to oranges" problem. 115 Google did, however, frei uently compare the quality of its vertical results to that of its competitors using others oring methods. 116 The verticals were initially placed in one of three locations: if Googled emed the vertical content to be highly relevant, it would go into position one, above then tural search results; ifGoogle deemed the content somewhat relevant, it would go into posit on four (or midway down the frrst page of natural search results); and ifGoogle deemed the content only marginally relevant, it would go into position I 0 (or at the bottom of the first page of natural search results). 117 In 2012, Google claims that it changed its algorithms to display Universal Search results in any position on the SERP, depending on the same initial relevancy screen. 118 next page. 22 SJ W a i v s nt me u c o D maximize the percentage of queries for which it displayed Universal Search results. 119 Evidence shows that Googlc sought to increase such "triggering" of Universal Search results not only to provide users with the "right" answer to their queries, but also to dri e traffic to Google properties. 120 Google recognized that the frequen t display of its vertica properties on the SERP was necessary to drive traffic to its properties, and thus, grow user sh rc in highly commercial areas such as shopping and local. 121 Google continued to trigger U iversal Search results frequently- and prominently - even when it detem1ined that sho 'ing such results in the top position would "cannibalize" revenue from the top ads, as the ompany was willing to lose short-term revenue with the long-term goal of retaining and growing vertical search query share. 122 Second, Google embellished its Universal Search results with photos and other eyecatching interfaces, recognizing that these design choices would help steer users to Googlc's vertical properties. 123 Third party studies show the substantial difference in traffic with prominent, !:,TTaphical user interfaces. 124 These "rich" user interfaces are not available to competing vertical webs ites. 125 Moreover, Google 's Universal Search results often were not labeled as being provided by Google affiliated services, but were integrated dire tly into the search results. Third, Googlc displayed its Universal Search results at or ncar the top o This desirable positioning ofGoogle's Universal Search results pushes all other eb search results down, which significantly decreases click-through to the websitcs displa ed in Google 's natural search results. 127 Google displays its Universal Search results n these prominent positions without comparing the quality ofGoogle's vertical content o that of its vertical competitors, 128 or evaluating whether users would prefer to see Google's content or 24 SJ W a i v s nt e m Docu Although Google tracks user click-through rates (and relies on such click-through data to improve its web search results in a number of ways, see supra p. 14), Google has not relied on click-through data to rank its Universal Search results against other web search results. 134 According to Marissa Mayer, Google did not use click-through rates to determine the position of the Universal Search properties because it would take too long to move up on tbc SERP on the basis of user click-through ratc. 135 Rather than comparing its content with that of competitors, Google used lhe occurrence of competing vertical websites in its natural search results to automa~ically boost the ranking of its own vertical properties above that of competitors. 136 For exam~lc, where Google's algorithms deemed a comparison shopping website relevant to a user's query, Google automatically returned Google Product Search - above any rival comparison shopping websites. 137 Similarly, when Google's algorithms deemed local websites, such as Yelp or CityScarcb, relevant to a user's query, Google automatically returned Google Local at the top ofthe SERP. 138 Google also dedicates space at the top of its SERP to its social network rtical, Google Plus. Google provides links to Googlc Plus pages that might be rclevan to a query on the right-hand side of the SERP, and "auto suggests" Google Plus pages for user queries, regardless of which social media sites are the most relevant, comprehensive, or ave the freshest results in response to any given user query. 139 Google also displays prolincnt links to Google Plus pages when users make navigational queries to many companies' wcbsitcs: For example, in response to the navigational query "Dell" a user is presented with the SERP shown on the next page. 26 SJ W a i v s nt me u c o D (offers),t 45 or "Sponsored" (the new paid Google Shopping ads), 146 and other tijes provided no label (flight search). 147 In May 20 12, Google announced that its shopping pr pcrty, Google Product Search- which will now be known as Google Shopping - will be transitioned to a paid listing model in the fall of2012. 148 Under the paid model, Lcrchants will pay Google directly to appear in Google Shopping, and Google will no long r include product listings for merchants who do not pay for placement. 149 Google's dedicated ads do not compete with other ads through Google's dWords auction for placement on Google's SERP. Instead, they enjoy automatic placemjnt in the most effective advertising places on the SERP, usually above the natural search ~esults. 150 Google also does not compare the quality of itc.; own ads to the quality of competitors' ads that provide the same vertical service. For example, although it displays its flight search above any nat un-~1 search results for flight-booking sites, Google does not provide tbe most flight options for travclcrs. 151 As with Google's Universal Search results, Google's rich user interfaces for its ads-based vertical offerings, which are unavailable to competitors, lead to higher clicks for Google 's ads. 152 e. . Google's Demotion of Competing Vertical Web ites While Google embarked on a multi-year strategy of developing and shoWI asing its own vertical properties, Google simultaneously adopted a strdtcgy of demoting, or refusing to display, links to cettain vertical websites in highly commercial categories. AccoLing to Googlc, the company has targeted for demotion vertical websites that have "little or no original content," or that contain "duplicative'' content. 153 Similarly, Google has identified comparison shopping websites as undesirable to users, and has developed several algorithms to demote these websites on its 28 SERr Through SJ W a i v s nt me u c o D algorithms that demote sites that "scrape a large percentage of their content from other sites." 163 These algorithms are not applied to Google vertical sites. Google's vertical properties would rank poorly if they were crawled and rdexed by Google because they have never been "engineered" for ranking by the search cntine. 164 I Unlike Google's vertical competitors, who expend considerable resources on op imizing their websites in order to rank highly on Googlc' s SERP, Google docs not expend the time and resources to optimize its own vertical properties; it simply places them on the SERP. f. Effects of Google's SERP Changes on Vertical tvals Vertical websites, such as comparison shopping and local websitcs, are h avily dependent on Google's web search results to reach users. 165 Thus, Google is in the unique position of bei ng able to " make or break any web-based busincss." 166 Google's prominent placement and display o[ its Universal Search properties, combined with the demotion of certain vertical competitors in Google's natural search results, has resulted in s ignificant loss of traffic to many competing vertical webjites. Data from various comparison shopping and other competing websitcs shows drops in traffic that correlate to changes implemented by Googlc to its SERP. 167 Google's internal d ta eonfinns the impact, showing that Google anticipated significant traffic loss to certain cate ories of ve11ieal wcbsites when it implemented many of the algorithmic changes described above. 168 While Coogle's changes to its SERP led to a significant decrease in traffi for the websites of many vertical competitors, Googlc's prominent showcasing of its vertica l propctties led to gains in user share for its own properties. 169 For example, Google's inclusion ofGoogle Product Search as a Universal Search result took Google Product Search from a rank of seventh in page views in July 2007 to the number one rank by Jul 2008. 170 30 SJ W a i v s nt e m Docu Google considered several options for obtaining information for display n its own vertical prope1ties: developing its own content; obtaining licenses from other co tent creators; and obtaining content by crawling the world wide web (in the same wal>' that Googlc crawls the world wide web for its general web index). Ultimately, Goo lc settled on a combination of all three of these alternatives. Much ofGoogle's vertical content is currently obtained through feeds fr m various wcbsitcs, pursuant to free licenses from those sites for Google to use that data. oogle's standard license agreement allows Google to usc third parties' data feeds for any purpose. 178 Many website publishers, such as Shopzilla, have agreed to these terms because they believe they do not have the levemge to negotiate with Googlc regarding tbc terms of their licenses, because they want the benefits of appearing in Google's vertica1. 179 l~on to the feeds it receives, Google 's use of crawled content is .Ervasive. Indeed, the content of any website that Google crawls for indexing purposes (for Google's web search) may be used by Google for any of its vertical search properties in a umber of different ways. For example, Google has often included "snippets" (or excerpts) of user l also uses reviews from local or shopping properties on its own vertical properties. Googlt the rankings of various businesses or products to aid its own determination rega ing the order in which those businesses or products should be ranked within its own ver ·cal properties. For example, Google calculates the popularity of a product for the purpose of ranking it in Google Product Search based on three factors: (I) Amazon Sales Rank; (2) the number of merchants offering the product for sale; and (3) the quality of those merchants. 180 I Because Amazon did not provide competitively sensitive information such as A azon Sales 32 SJ W a i v s nt e m Docu For Googlc Local, Google needed photos, addresses, hours, and reviews. Google originally obtained this content through licenses with the.<>e websites. In late 2006, Google decided that it wanted more control over its local content. 186 Google recognized that review content, in particular, was "critical to w inning in local search," but that Google had an "unhealthy dependency" on Yelp for much of its review content. 187 Google feared that its heavy reliance on Yelp content, along with Yelp's success in certain categories 1nd geographies, could lead Yelp and other local information websites to siphon use s' local queries away from Google. 188 I In order to acquire direct access to a large storehouse of user content, maragcrs """ working on Goo le Local attem ted to convince Google executives to purchase Yelp, but they were rebuffed. ..:.,___---- 189 Instead, Google decided to launch a redesigned version of Google Maps, in which users could submit reviews directly to Google. 190 Googlc understood that the existence of a critical mass of user reviews (like those users bad already submitted to websites like Yclp and Trip Advisor) was import nt in attracting additional user reviews. 191 Google also knew that its partners- such a, Yelp and Trip Advisor - wou ld be unhappy about Googlc 's osc of their content to collect f oogle 's own conl'ent. 192 Indeed, upon learning of Google's intent to collect its own revi ws and to develop this now-directly competing properly, Yelp discontinued its data feed tol Google, and asked Googlc to remove all Yelp content that Google fea tured on Google Loca1. ]93 move - and did remove - Yelp's content. However, after offering its own review site for more than two years, Google recognized that it had failed to develop a community of users- and thus, the critical mass of user reviJ s- that it needed to sustain its local produet. 194 In an attempt to gain quick access to a lar~e storehouse I 34 SJ W a i v s nt me u c o D included in Google Places or not have their property appear in Google web search results at al1? 03 Critically, for Google, this meant that it could now force local websites- that needed access to Google's web search to reach users - to accede to Google's use ofthe large storehouse of reviews that Google's rivals had built in order to develop its own user base.204 Indeed, G9 o most stmu taneous y aunched a new rcvicws-collectiof product- Hotpot - to (again) try to solicit original user reviews, this time seeding it w ith ~eviews from thi rd-party websites with no attribution. 205 Yelp, TripAdvisor, and CitySearch a 1 complained to Google? 06 All of these parties sought removal of their user review content from Google Placcs/Hotpot, as well as the removal oftheir reviews from Coogle's aggregated review count on the main SERP? 07 This time, however, Googlc told each company that ifYelp, TripAdvisor, and CityScarch wanted to have their content removed from Google PlacesfHotpot, they would have to exclude their websitcs from being crawled by Google altogether, which meant complete exclusion from Google's SERP. 208 This was rot ccssary - tt was JUSt a po tcy decision by Google. 209 Like many other vertical websitcs, Yelp, TripAdvisor, and Citysearch reJjed heavily I on Google 's web search results to reach users, and thus could not risk removal lom Google's web search index? 10 Instead, they each attempted to negotiate with Google, seling removal from Coogle Local (without simultaneous removal from Google's web search r suits), or at least a user interface that provided sufficient attribution o f their content.211 Facing what seemed to be an ali-or-nothing choice, Yelp also began widely . publicizing Google's refusal to remove Yelp content fro m Google Local (including filing a complaint with the Commission), and ultimately, in July 2011, sent Googte a Cease and Desist letter. 212 In its letter, Yelp clearly indicated that it expected to remain in Google web 36 SJ W a i v s nt me u c o D Google had already collected sufficient reviews by bootstrapping its review collection on the display of other websites' reviews. It no longer needed to display third-party reviews, particularly while under investigation for this precise conduct. b. The "Shopping" Story Much ofGoogle Product Search content is obtained through feeds from arious websitcs with corresponding license agreements, from crawls, and to a lesser ex ent, by generating its own contcnt.220 As Google sought to develop a stronger shopping offering beginn ing around 2006, Google recognized the need to improve its data in sever l areas. Googlc decided to supplement its feeds with additional merchant review , product reviews, and product listings it could get from crawls, particularly from Amazon. 221 Amazon had a license agreement with Google starting in June 2009. Pursuant to this agreement, Amazon provided Googlc with only a limited data feed of information about its products, and sought to limit how Google used the data, because Amazon has always feared that Google would usc Amazon's comprehensive product catalogue and original review content to develop a strong competitor in shopping. 222 Shortly thereafter, claiming that Amazon's data feed to Google Product hoppi ng was too limited, Google decided not to rely on the feed, but instead, crawled Amazo 's website to scrape the much more detailed product information - including star ratings and reviews. 223 Googlc also relied on Amazon's web pages that indicate the ranking of products within Amazon. Google used- and continues to usc - this information to determine the order in which to rank products within Google Product Search. In August2010, around the same time that Yelp requested that Google r ove Yelp's content from Google Local, Amazon requested that Googlc stop using Amazon' crawled 38 SJ W a i v s nt me u c o D comprchensive? 31 Google also attempted to pay a company to generate a new hierarchy, but this was taking a long time, and also was not sufficiently comprehensive.232 Ultimately, Google decided to cra\vl Amazon's product web pages, read mbedded information on Amazon's pages indicating Amazon's classification system, and to use that infom1ation to create Google's classification.233 This was critical to Googlc bec~use Froogle had failed partly due to Google 's inability to accurately classify the millions of roducts from feeds and crawls, and to return correct search results. 234 Amazon considers its c assification system an important competitive advantage that it spends tremendous resources o develop, and docs not approve ofGoogle's use of Amazon 's system to develop its O\V'll .rs .>r With Google's migration to a paid shopping modcl,236 Google has stated that it will only use reviews from companies that provide licensed feeds of their content. lt appears, however, that Google may continue to crawl, and rely upon, rivals ' product classifications to generate its own, and on rivals' ran kings to determine raukings of products with'n Google Shopping. c. Effects of Google's "Scraping" on Vertical Riv~ls Because Google scraped content from these vertical websites over an ex~ended period of time, it is difficult to point to declines in traffic that arc specifically attributable to Google's conduct. However, the natural and probable effect of Googlc 's condur is to diminish the incentives of companies like Yelp, TripAdvisor, CitySearch, and A~azon to invest in, and to develop, new and innovative content, as the companies cannot fully capture the benefits of their innovations. 237 40 SJ W a i v s nt e m Docu Initial ly, Google offered advertisers two ways to access the AdWords system and manage their campaigns: the AdWords Front End and the AdWords Editor. The Front End is a web page that advertisers could log into and manage their campaigns. The Editor is a program advertisers can download. It allows advertisers to download campaign inforn1ation from Googlc, make bulk changes offline, and then upload the changes back into AdWords. Thesetwo access points eventuall y proved to be insufficient because large adve~isers and agenc1es were taxmg the ex1stmg system. They would access the system and mlke so many changes to their campaigns that the system's capacity would be exceeded, causi~g it to be unavailable temporarily or even to crash. 240 In response, in 2004, Google introduced a third method for accessing the AdWords system: the AdWords APJ. The A PI (application programm ing interface) allows advertisers and agencies direct programmatic access to the AdWords platform. The API contains a set of specifications that allows advertisers and agencies to develop their own software programs to interact with the API and allow them to set up and optimize their ad campaigns. APls arc now an essential feature of campaign management for advertisers and agencies ran aging multiple accounts." ' All three major search advertising platforms (Google, Mi~osofi, and 242 Yahoo!) have APis that allow th1s dnect, automated mteract1on w1th ad platform featw-es. Google anticipated that the API would have several benefits, inctudjng: fl) reduced Google operating expenses (Google personnel having to provide manual proces~ing and troubleshooting for large bulk sheets); (2) increased advertiser spend due to reduced advertiser operating costs; and (3) rapid development of advertiser and third party tools supporting AdWords campaigns 243 42 SJ W a i v s nt me u c o D Amazon and eBay, can develop- and have developed - their own simultaneously manage campaigns across platforms? 48 The advertisers affected are those whose campaign volumes are large enough to benefit from using th e AdWords API, but too small to justify devoting the necessary resources to develop in-house the software and expertise to manage multiple search network ad campaigns. c. Effects of the Restrictive C onditions i. Effects on Advertisers and Search Engi e Marketers ("SEMs")249 As noted above, the immediate effect of the restrictive conditions has been to prevent the development and marketing of tools that would allow ad vertisers to manage ad campaigns on multiple search advertising networks simultan eously. Google routinely audits its API clients to determine compliance with the restrictive cond itions. On several occasions, Google bas required SEMs to remove functionality t hat would facilitate simultaneous management of search advertising campaigns.250 Other SEMs have stated that, but for the restrictive conditions, they too would develop and offer such functionality. 251 ey would also be freer to innovate the tools they offer based on their clients' demands. 252 Google anticipated that the restrictive conditions wou ld e liminate SEM incentives to iru ovate.253 Many advertisers have said they would be interested in buying a tool th homing functionality. 254 bad multi- Such functionality would be attractive to advertisers because it would reduce the costs of managing multiple ad campaigns, giving advertisers access to add itional advertising opportunities on multiple search advertising networks wi~h minimal additional investment of time. The advertisers who would bene fit from such a tool appear to be the medium-sized advertisers, whose advertising budgets are too small to jus ity hiring a 44 SJ W a i v s nt e m Docu ii. Effects on Competitors lt seems likely that the removal ofGoogle's API restrictions would increase the amount of advertising spend directed towards search networks that compete with Google. The rationale is that many advertisers would be w illing to advertise on Bing or Yahoo! if they could do so w ithout incurring significant transaction costs. As noted above, optimizing a search advertising campaign is time-intensive. lt may not be worthwhile investjng such efforts for additional, smaller search networks. Microsoft contends that if management tools that allowed advertisers to optimize their campaigns on multiple search networks simultaneously were available, many more advertisers would choose to advertise on the networks that comp ete with Google. Data on advertiser " multi-homing" may show some of the effects of the restrictive conditions. ''Multi~homing" refers to advertisers that advertise on multiple search networks. The data indicate that nearly all of the largest advertisers multi- home, but the percentage of multi-homing declines as the advertisers ' spend decreases. According to a 201 1 study by I Microsoft, which divided the advertiser base into deciles based on total number of clicks (such that the largest nine advertisers comprise a decile - or 10 pen;ent of total cl cks - unto themselves), the distribution of multi-homing was as follows: 262 Deci le Advertisers (from smallest %multi-homing'<> ! to largest) I 208980 2 18346 3 4876 83.0 4 1736 90.8 ( ---~ 31~ 8 ) '~ 46 SJ W a i v s nt e m Docu basis, the same advertisers optimize their Microsoft campaigns far less frequently, on a weekly or bi-weekly basis. 266 Staff conducted a series of interviews o f randomly selected small adverti l ers to gather their anecdotal perspective on these issues. These interviews strongly tend to su port the thesis that many small advertisers would extend their advertising to other search networks if they had access to a cross-platfom1 optimization too l. Nearly all small advertise s interviewed showed interest in such a too1. 267 They believed such a cross-platfo~m optimization tool would be central to addressing their core constraints: time, sophistication, and money. 268 When these transaction costs are coupled with Bing's limited volume, some small advertisers refrain from using Bing altogether? 69 Furthermore, even those that do use Bing may not be fully optimizing their Bing campaigns because the benefits of Bing's limited user volume may not out\veigh the transaction costs associated with full optimization. 270 d. Internal Coogle Discussions Regarding tbe Restrictions I ntemal Google documents support the notion that the removal of the res rictions would increase advertiser spend on competing networks. In 2007, when eonside ing whether to offer a cross-network management tool, an API product manager wrote (and director of product management Richard Holden endorsed): If we otTer cross-network SEM in [Europe), we will give a significant boost to our competitors. Most advertisers that I have talked to in [Europe] don't bother running campaigns on [Microsoft] or Yahoo because the additional overhead needed to manage these other networks outweighs the small amount of additional traffic. For this reason, [Microsoft} and Yahoo still have a.fi·action of the advertisers that we have in [Europe}, and they still have lower average CPAs [cost per acquisition}.271 This last point is significant. The success of Googlc's Ad Words auctions has ""rv•-·" to raise the costs of advertising on Google. With more advertisers entering the AdW 48 SJ W a i v s nt me u c o D and the plans to improve DART Seareh. 274 However, a series of documents- documents authored by Holden -explicitly link the two ideas. In December 2008, Holden, senior vice-president of ad products Susan W ~ cicki, and others met to discuss the issue. Of the meeting, Holden wrote: ij [O]ne debate we arc having is whether we should eliminate our A requirement that AW [AdWords] featu res not be co-mingled with ompetitor network featu res in SEM cross-network tools like DART Search. We are advocating that we eliminate this requirement and that we build a uch more :;treamlined and efficient DART Search otTcring and let SEM tool rovidcr competitors do the same. There was some debate about this, but 1 e concluded that it is better.for customers and the indust1y as a who e to make things more efficient and we will maximize our opportunity by mo ing quickly and providing the most robust offering. 27 In February 2009, Holden wrote the executive summary for a DART Search product review, in which he advocated that Googlc ·'alter the AdWords Ts&Cs to be less restrictive and produce the leading cross-network toolset that increases advertiser/agency efficiency." Such a move, he wrote, would "[r]educe friction in the search ads sales and management process and grow the industry fastcr." 276 In April 2009, in light of evident disapproval from Larry Page about the idea of removing the co-mingling restriction, Holden wrote: "We've heard that and we will focus on building the product to be industry-leading and wm evaluate il with him when it is done and then discuss co-mingling and enabling all to do it. '277 In September 2009, the API product manager again raised the possibility f eliminating the restrictive conditions as a way to help DART Search, this time wi h the added argument that DART Search was not able to compete effect ively against other SEM crossnetwork tools that might be violating those restrictive conditions.278 Before the i . ue was rnised up the Ladder to Susan Wojcicki, the API product manager asked Richard 1 oldcn's advice : 50 SJ W a i v s nt me u c o D advertising syndication (or "search intermediation"). We recommend that the Commission issue a complaint against Google for this conduct. a. Publishers and Market Structure The buyers of search and search advertising syndication ser vices are web ite publishers. In effect, any website that has content that it would like to monetize ia ads is a potential buyer of syndication services. Whi lc there arc thousands of these webs· es, a handful of the largest websites on the Internet account for the vast majority of sy dicated search traffic and revenue. 285 Google served approximately 118 billion AdSense (search syndication) queries in 201 1, but just 10 websites generated almost 80 percent of that tra flic. 286 The biggest customers for search and search advertising syndication services are ecommercc retailers (e.g., Amazon and eBay), traditional retailers with large associated websites (Wal-Mart, Target, Best Buy), and Internet Service Providers ("TSPs"),2 7 which operate their own web portals.288 Below this small group of very large publishers, there arc another roughly 25 companies with significant query volume. These mid-tier companies include ver ical ecommerce sites such as Kayak (travel), along with smaller retailers and smaller I Ps such as EarthLink. None of these mid-tier companies generate even one percent ofGoogle's total AdScnse query volume. Below these companies, publisher size drops off rapidly to well under 0.1 percent of Google's query volume. 289 The search provider pays the publisher (website) a percentage of the revenue generated from user ad clicks on the publisher's website. ln the industry, these agreements arc known as " revenue sharing" arrangements. The higher this percentage, the m re the 52 SJ W a i v s nt me u c o D domain related) advenising services. There are two main categories of AdSense agreements: AFS (search), which provides search advertising to publishers, and AFC (content), which provides contextual advertising to publishers. Staffs investigation has focused on Google's AFS agreements. Within the AFS category, there are two types of agreements: (i) Google erv1ce Agreements ("GSAs"), which are individually negotiated agreements with large partners; and (ii) standard online contracts, which are non-negotiable and non-exclusive agreements that any publisher can sign?96 Standard online agreements make up the bulk of Google's AFS partners, but only a small portion ofAFS rcvcnucs. 297 The bulk of the revenues lc ome from the GSAs with Google's 10 largest partners, which collectively comprise almost 80 percent ofGoogle's overall AFS query volume in 201 1.198 All ofGoogle's GSAs contain some form of exclusivity or "preferred placement" for Google, and the GSAs typically last from one to three years. 299 Google's exclusive AFS agreements effectively prohibit the use of non-CSoogle search and search advertising within the sites and pages designated in the agreement.300 Some exclusive agreements cover all properties held by a publisher globally; other agr ements provide for a property-by-property (or market-by-market) assignment. 30 1 By 2008, with its market presence clearly established, Googlc began to t11ligrate away from outright exclusivity in all of its agreements toward what Googlc terms "preferred placement" in many of its agreements.302 In essence, the "preferred placement" provision requires the publisher to display three Google ads or the same number of ads the publisher acquires from any competitor (whichever is greater)~ that Google's ads be displayed in an 54 SJ W a i v s nt e m Docu The customers generally confirmed Microsoft's claim that Bing's search syndication o~fer ing is inferior, at least in part, because Microsoft's network of advertisers is smaller than Google's. With a significantly larger advertiser base, Googlc is more likely to have a relevant, high-quality advertisement for any given query, which greatly improves its monetization rate relative to Microsoft. 307 A smaller publisher reported that, essentially, the only websites exclusiv ly using Bing's search syndication service today aTe those that have been kicked out ofGoogle's syndication network for violating its tenns of service. 308 While we know from other interviews that this comment is an exaggeration, it does capture the general tcno Iof the comments we received about the relative quality of Microsoft's search and search advertising syndication product. / Many publishers reported that M icrosoft was not aggressively trying to win their syndication business. One mid-tier publ isher stated that Microsoft did not even return its inquiry calls during the publisher's last contract renewa l discussions with Google.309 A Microsoft executive acknowledged that Bing needs a larger portfolio of advertisers in order to present a competitive offering to publishers, and so the company has not been ocused on wi nn ing new search syndication business .310 Another common theme we heard from many (but not all) of the publishJs is that serving advertising is a relatively minor part of their business and not a significant strategic focus for them. For example, Wal-Mart operates its website principally as an extension to its retail operations (letting Wal-Mart customers buy products either in-store or from the website at their preference). 311 Best Buy's principal goal for its website is to be the provider of presale information, as 60 percent of its customers do online research before coming to the 56 SJ W a i v s nt me u c o D reductions in their AdSense revenue share percentage as large enough to justify shifting their business to Bing or to begin serving more display advertisements instead of search ads. 321 Publishers ' Views of Exclusivity Provisions ii. When asked whether their AdSense contract with Google was exclusive, the publishers gave widely varied answers. A number of the large publishers reporter that their AdSensc contract with Google was exclusive,322 but some reported that their Adrnse contracts were not exclusive. 323 Most of the publ ishcrs that reported exclusivity ~rovisions did not complain to us about them. Staff s interviews did identify a fairly small, but significant, group of publishers that were deeply concerned by the exclusivity provisions in their Google AdSense agreements. All of these customers view search and search advertis ing syndication income as a substantial part of their business, and all have the technical sophistication to integrate multiple suppliers into their on-line properties. W e summarize these concerns below. eBay. eBay is Google's largest search and search advertising syndication partner, accounting for just over 27 percent of the syndicated U.S. queries answered by G ogle in 20 I l. 324 Section 14 of eBay' s AdSense agreement states that the agreement is n t exclusive. 325 However, the contract requires preferential treatment for Googlc A •Sense ads, which eBay has characterized as equivalent to exclusivity. 326 The preferential treatment terms include requirements that eBay show as many Googlc AdSense ads on each page as third-party advertisements, that no third party advcttiscmcnts appear above the Google AdSense advertisements, that Google AdSense advertisements cannot be interspersed with third party advertisements, and that Googlc AdScnsc advertisements cannot be less prominently displayed than third parry advertisements.327 58 SJ W a i v s nt me u c o D because it would have such poor placement on the NexTag site due to the Google contract restrictions.336 Business.com. Business.com is a " B2B" lead generation/vertical site. In effect, the site marries commercial customers looking for products (such as business phone systems) with providers ofthose products.337 Busincss.com is several orders ofmagnitud smaller than the other complainants, barely making it onto a list of the lop 60 providers or AdSense query volume. Googlc. 338 Busines~.com reports that it has an exclusive AdSense agreement with This agreement materially limits how Business.com can design its web pages. If Business.com were relieved from its exclusive arrangement, it would test Bi11g and Yahoo! by product category, and place their advertisements in a more prominent position in those categories where their perfonnancc warrantcd. 339 The company would also likely take advertisements from both Google and BingNahoo! , and show them on the same page, with placement dictated by relative performance in each category ? 40 Loosening up Googlc's exclusivity restrictions would allow Business.com to improve its revenue, and also allow it to introduce some new features that would make the site more accessible and user-Tendly. 341 Amazon. Amazon is the world's largest c-commcrcc sitc342 and the seco1d largest AFS customer after eBay. On a worldwide basis, Amazon earns roughly $ 175 m Ilion from search syndication services, with $169 million of that total coming from Google's AdSense search product. 343 Amazon does not have an exclusive agreement with Googlc, and actually splits its inventory among Google, Bing, and Yahoo! .3-14 However, Amazon finds that the Bing and Yahoo! 's advertisements monetize at about 46 percent the rate of Googie's advcrtiscments.345 Because of the very large monetization gap, Amazon can only afford to 3 46 use B ing and Yahoo! for a very small percentage of its total search syndication n~eds. 60 SJ W a i v s nt me u c o D during the negotiation period wanted an exclusive arrangement. 358 Ultimately, Google's offeri ng was the most lucrative, and lAC re-signed with Googlc. 359 However, lAC expressed concern about Google's requirement of exclusivity for subsidiary properties, such as local website CityGrid, that wanted to explore "mi -andmatch" options with other search advertising providers. Indeed, in 2008, lAC de lined to opt CityGrid into its larger exclusive agreement, attempting to forge an alternative ro 1te with other search advertising providers (including CityGrid's own ad network). Ultir ately, however, CityGrid determined that it could not completely replace Google's syn~ication network, even with a patchwork of other providers. Since then, CityGrid has been forced to "opt in" to lAC's larger exclusive agreement. Although CityGrid wants the option of using other networks (including its own), and supplementing those ads with Googlc ads, it cannot do so under lAC's existing agreement with Google. More generally, lAC expressed concern about the lack of competition in search and search advertising syndication because there are no good substitutes for search advertising. 360 While lAC initially seemed supportive of the story we heard from the oth r concerned publishers, during a recent follow-up call, lAC's tone changed substa tially. One of the key complainants on the initial call was the president of lAC subsidiary Ci Grid. That executive has since left lAC, and our more recent call was with another exequtive, who was in charge of business development for lAC. This executive was far less sanguine as to lAC 's likelihood of splitting their business in the absence of exclusivity. He noted that, while he was also concerned about the lack of competition in the market, he could not see moving incremental traffic to Bing or other search advertising providers unless the monetization gap narrowed significantly. The departure of the key executive with the closest 62 SJ W a i v s nt e m Docu An attempted monopolization claim requires a showing that (i) "the defendant has engaged in predatory or anticomperitive cooduct" with (ii) "a speci fic intent to t onopolize" and (iii) a dangerous probability of achieving or maintaining monopoly power. 3 A. 5 GOOGLE HAS MONOPOLY POWER IN RELEVANT M1 RKETS "A fi rm is a monopolist if it can profitably raise prices substantially abo e the competitive level. .. . [M]onopoly power may be inferred from a firm's posses ion of a dominant share of a relevant market that is protected by entry barriers." 366 Goog le has monopoly power in one or more properly defined markets. I. Relevant Markets and Market Shares A properly defined antitrust market consists of ''any grouping of sales whose sellers, if unified by a hypothetical cartel or merger, could profitably raise prices significantly above the competitive level." 367 Typically, a court examines "such practical indicia as industry or public recognition o f the submarket as a separate economic entity, the product 's peculiar characteristics and uses, un ique production fa cilities, d istinct customers, distinct prices, sensitivity to price changes, and specialized vendors."368 Staff has identitied three relevant antitmst markets. a. Horizontal Search Horizontal, algori thmic web search (hereafter "horizontal search") likely! constitutes a properly defined relevant market. As discussed earl ier, horizontal search engines, such as Google, attempt to cover the content of the Internet as widely as possible, and are specifically designed to return a comprehensive list of search results on any topic. By contrast, "vertical'' search engines focus on more narrowly-defined categories of content, such as product 64 SJ W a i v s nt me u c o D words, to the extent vertical wcbsites compete with horizontal search providers 'vithin their limited areas of competence, nothing prohibits price discrimination between thdse narrow areas and the broader web. Even in the narrow areas where vertical websites have subject matter co petence, they face challenges in competing effectively with horizontal search providers. This is because comprehensive coverage of all topic areas appears to be a very importa t driver of demand, even to websites focusing on specific topic areas. The abi lity to offer comprehensive search results was characterized as "fundamental" b_y Google's former CEO, Eric Schmidt. 372 Schmidt explained that the company needs to build brand equity with its customers by providing consistently good results regardless of the content of the query, and that strong results across-the-board lead to specifi c queries in commercial search: So if you, for example, are an academic researcher and you use Google 30 times for your academics, then perhaps you ' ll want to buy a camera ... So long as the product is very, very, very, very good, people will keep coming back ... The general product then creates the brand, creates demand and so forth. Then occasionally, these ads get clicked on.313 In effect, users are habituated into using Google for all their queries because of its comprehensive scope, and so they may be more likely to tum to Google when they have commercial queries, instead of starting at a vertical website. Schmidt's testimo?y is corroborated by the representations of several of the vertical search firms, who note tJ1at they arc dependent on horizontal search providers for significant amounts of their traffic, because even many vetiical search users tend to begin their search with a query on Google, Bing or Yahoo!. 314 When asked to identify his competitors in web search, Schmidt did not mention any vertical property: "[A]s far as I can tell, the industry has two main horizontal 66 you SJ W a i v s nt me u c o D properly define the scope of the geographic market for web search. Our investigation has uncovered no basis on which to deviate from this conclusion. 319 Google is clearly the dominant provider of "general search" services in tt c United States. Google's own sites have a 66.7 percent share of the market as of May 20 2, according to ComScore, a leading industry measurement finn. 31l0 Google also povides search services to two small, formerly independent web search operators (Ask.com 381 and AOL382), which collectively account for another 4.6 percent of the relevant marbt according to ComScore.m ln sum, the total Google-powered query share in the United Sta es is 71.3 percent, according to ComScore. 384 The balance of this market is controlled by the Microsoft/Yahoo! search alliance. Yahoo! holds approximately 15 percent of the market, and Bing (owned by M icrosoft), holds approximately 14 percent.385 As noted earlier, since 2009. Microsoft and Yahoo! have been partners in what essentially amounts to a long-tem1 joint venture for search, where Microsoft powers the algorithmic search results for both Yahoo! and Bing, while Yahoo! handles the direct relationships with large advertisers for the combined scrvicc. 386 Advertisers that want to purchase search advertising on Yahoo! or Bing cannot buy access to these pro erties separately, but rather must purchase advertisements that run on both sites simultreously. 387 So, in effect, there are just two providers of horizontal search: Google and the BTg!Yahoo! search alliance. Firm ComScore Market Share, May 2012 Google 71% Bing/Yahoo 29% 68 SJ W a i v s nt me u c o D are typically more interested in developing user interest- or "branding"- than in eliciting a direct response from the consumer, whereas the primary attraction of search advertising is its propensity to generate direct responses. 394 As Hal Varian, Google's chief ccono rust put it, "[o]nc way to think about the difference between search and display/brand adve ising is to say that 'search ads help satisfy demand' while 'brand advertising helps to crcat demand." 395 The different manners in which disp lay and search advertising arc priced ·s consistent with their disti nct overarching goals. Search advertisements - whose main goal is to directly drive user purchases- are priced on a "cost-per-cl ick" basis (i.e., an advertiser ooly pays if a user clicks on the ad). Conversely, display advertisements - whose main goal is to spark interest and drive awareness- are priced based on the number of times the ad is displayed. Display and search advertising arc also separately managed, measured, and tracked internally at Google. 396 Similarly, for advertisers and agencies, display and search are different eategories.397 The ad types require different creative, targets, budgets, ~d tracking. 398 Most advertisers spend in both categories, as they consider display a d search advcttising to be complements.399 Evidence suggests tbat search and display arc indeed complements rather than substitutes. Google has observed steep click declines when advertisers have atte~pted to shift budget to display advertising. For example, when automobile manufacturer Chevrolet decided to suspend its search advertising campaign for two weeks, and rely on display advertising alone, it lost 30 percent of total clicks on irs wcbsitc. 400 In recent years there has been some perceived convergence between the display and search advertising. 401 With varying degrees of success, both display nd search 70 SJ W a i v s nt me u c o D ofadvertising. 414 However, a minority of advertising agencies and advertisers sa·d they would move advertising dollars away from search advertising in response to a SSNIP. Google's internal documents and testimony confirm that there is 415 currcntl~ no viable substitute for search advertising. Both Ad Words vice-president of product manakement Nick Fox and chief economist Hal Varian have previously stated that search advertis+ spend docs not come at the expense of other advertising dollars.'" And former GoogleiCEO Eric Schmidt has twice testified unequivocally- in both this investigation and in a prior Depa1tment of Justice investigation- that search advertising is ·'the most effective tool for reaching the customers that arc actually prepared to buy,"41 7 and "has the best ROI of any advertising as best we can determine.',... 18 Both the Commission and the Department of Justice have previously found online " search advertising" to be a distinct product market. Specifically, in 2007, the Commission noted that ''advertisers purchase different types of ad inventory for different purposes," and concluded that "the sale of search advertising docs not operate as a significant constraint on prices or quality of other online advertising."419 The Department o f Justice found that search advertising was a relevant antitrust market in 2008, and again endorsed search a~vertising as a relevant market in 2010. 420 While no cowt has yet determined that search advertising constitutes a relevant market, courts have repeatedly recognized naJTOW advertising markets. For example, in Times-Picayune Publishing Co. v. United States,421 the Supreme Court identified newspaper advertising as a unique antitrust market. There, the Court held that there were two "separate though interdependent markets"- one market for selling news and ads to readers and a 72 e m Docu SJ W a i v s nt The consumers in this market arc the publisher wcbsitcs that wish to provide search services and return search advertisements on their websitcs, while the sellers are the horizontal search providers, Google, Bing, and Yahoo!. 43 1 Staff has interviewed a number of publishers of various sizes, and they rovide very consistent responses on the issue of cross-elasticity of demand. Publishers report that search and search advertising syndication monetizes better than display advertising or other content that they might place on their websites. 432 The publishers do not view other fo I s of advertising as viable substitutes for search and search advertising syndication.43 None of the publishers told us that a modest (5 to 10 percent increase) in the price for searc and search advcttising syndication would cause them to shift away from search and search advertising syndication in favor of other forms of advertising or web content. 434 Further support for this relevant market comes from Google's efforts to systematically reduce TAC, or the amount of money Googlc shares with the pu lisber from syndicated searches. A decline in revenue share is effectively a price increase t the publishers. A number of the publishers have seen their revenue share from Goo le decline significantly in recent years as a result ofGooglc's efforts.m Of the publishers ptaffhas interviewed, none have reduced or eliminated their usc of search and search adv~rtising syndication in response to these price increases. 436 In effect, Google's successful efforts to syslcmatically to reduce revenue share constitutes a natural experiment to deterline the I likely response to a SSNTP. The publishers' response to Google's price increases has been universally consistent with the proposition that search and search advertising syndication (search intermediation) is a relevant market. 74 SJ W a i v s nt e m Docu b. Substantial Upfrout Investment Along with specialized algorithms, search and search advertising platfonus require enormous investments in the technology and infrastructure required to crawl and categorize the entire lnternet. 443 For instance, in 20 11, Google spent more than $5 billion or research and development, although this figure is inclusive of all ofGoogle's divisions. 444 And, in 2010, Microsoft invested more than $4.5 billion into developing its algorithms and building t . I capacity . necessary to operate B.mg.445 1t1e phys1ca c. Scale Effects As discussed at length earlier, Internet search, search advertising, and search syndication arc markets that are characterized by substantial scale effects. As more consumers use a general search engine, its search algorithms are honed to improve its accuracy in retrieving the information that consumers want. More users also leads to an increased number of advertisers. And, as the number of advertisers that place ads- and the number of consumers who click on those ads - increases, the ad-serving algorithrs improve their ability to predict what advertisements stimulate consumer "clicks." This, i turn, inc~eases monetization .for the. search engine, its a~ve~tisers, and its syn~ication rrtn~rs, WhiCh leads to the cychcal cftect of greater partiCipatiOn by both advertiSers and rubhshers. This effect, which bas been termed the "virtuous cycle," represents a significant ~arricr for any potential entrant.446 Indeed, according to Microsoft, its greatest barrier is obtaining sufficient scale through its collection of search and advertising data, and it faces an enormous task in trying to catch up with Google. Despite substantial investments in technology and infrastructure, Microsoft has yet to make a significant dent in Google's market share, and has b Ien losing 76 SJ W a i v s nt me u c o D trafficked websitcs, which, in tum, magnifies the problems of scale effects. In addition, the exclusive agreements act as barriers to smaller, more specialized search advertising platforms (e.g., a network specializ ing in local Washington, D.C.-based advertising, or in specific categories, such as travel). B. COOGLE HAS ENGAGED IN EXCLUSIONARY CONDUOT Conduct may be judged exclusionary when it tends to exclude competitor "on some basis other than efficiency," i.e., when it "tends to impair the opportunities ofrii ls" but "either docs not further competition on the merits or does so in an wmecessarily restrictive way.'"' 52 In order for conduct to be condemned as "exclusionary," Staff must sh9w that Google's conduct likely impairs the ability of its rivals to compete effectively, and thus to constrain Google 's exercise of monopoly power. 453 1. Coogle's Preferencing of Google Vertical Properties Within Its SERP As described earlier, Staff has investigated whether Googlc is unlawfully prcfcrcncing its own vertical content over that of rivals, while simultaneously deP,oting rival vertical websitcs, in order to maintain, preserve, or enhance its monopoly power in the markets for search and search advertising. Althollgh we believe that this is a clol·e question, we conclude that Google's prc.fcrencing conduct docs not violate Section 2. a. Coogle's Product Design Impedes Vertical Competitors "As a general rule, courts are properly very skeptical about claims that competition has been harmed by a dominant fum's product design changes . . . . Judicial deference to product innovation, however, does not mean that a monopolist's product design decisions are per se Jawful.'"' 54 In United States v. Microsoft, the D.C. Circ uit conc luded that several 78 e m Docu SJ W a i v s nt The theory of harm to competition is mainly one of reduced innovation: that, when faced with Google 's seamless ability to enter into highly monetizable categories of commerce and simultaneously to disadvantage its competitors, existing competitors cannot innovate at the same pace; new or i1movative vertical websites will cease to entc the market; and consumers will be faced with a corresponding reduction in innovation and choice.459 b. Google's SERP Changes Have Resulted In Anti ompetitive Effects Google's conduct has resulted in significant harm to 1iva l vertical websit~s in a number of different categories. As described earlier, in the comparison shoppinJ category one of the first areas in which Google vigorously expanded its own offering, while simultaneously demoting rival offerings- many rival websitcs have experienced significant declines in traffic. Data obtained from NexTag and Shopping.com, among others, suggests that, as a result of Googlc's conduct, these web sites have experienced significant drops in traffic. Google's internal data confirms this impact.460 Simultaneously, Google's prominent placement and display of its Univer al Search properties led to gains in user share for its own properties. For example, Google s inclusion of Googlc Product Search as a Universal Search result turned a property that the Google product team could not even get indexed by Googlc's web search results into the number one viewed comparison shopping website on Googlc. 46 1 c. Google's Justifications for the Conduct Google claims that the conduct under review improves its product and benefits users. "[A] design change that improves a product by providing a new benefit to consumers does not violate Section 2 absent some associated anticompetitive conduct." Allied Orthopedic Appliances, Inc. v. Tyco Health Care Group LP, 592 F.3d 991, 998-99 (9th Cir. 2 10). 80 SJ W a i v s nt me u c o D results. Google measures the quality of its verticals by assigning relevance valu ( to each individual vertical result, i.e., to each merchant in a specific product search, or t each location in a local search (which may be ranked by popularity, rating, number ofGooglc reviews, distance, and other factors). Google's web search results, on the other hand, receive a score based on he text read from crawling the contents of the page. Based on the crawled text, the pages are rated using factors such as click-through rates (i.e., how often previous users clicked on the r ge), commerciality (i.e., whether the page has too many ads), and the page's PageRank.468 With Googlc's current algorithms, Google cannot directly compare, say, the ranking for a specific restaurant (in its own local results) to the ranking for an entire web page (in someone else's local results). 469 On the other hand, Microsoft has told us that Bing uses a single signal - click-through rate- to determine where to place the Universal Search content within the organic search results. 470 Google's justification for promoting its own properties above that of competing properties automatically when those properties appear (reca ll the algorithms that oosted Googlc Product Search to the top of the SERP whenever another comparison shrping website was deemed relevant) is not as strong, but still has some force. Google' justification for this conduct is that, if another vertical property is deemed rcleva1,1t by Google's algorithms, Google's vertical property must also have high quality results - and Google's rich Universal Search results are more helpful to the user than "blue links" to other comparison shopping websites. Google 's justification for surfacing only (or mainly) Google-sourced con ent- rather than third-party vertical content - w ithin its Universal Search results is less conv ncing. 82 SJ W a i v s nt me u c o D Google's defense for this conduct essentially boils down to "user expectations." Sergey Brio testified that Google's showcasing of its Universal Search results is not inconsistent with the demotion of other similar vertical content because Universal Search represents a "mode change" for users. 476 According to Brin: So when you search for products rather than searching for web p ges, I feel like that's more of a mode change. You know, you're switching in fact, you can switch .... You can switch to product mode. And I think th t would be confusing in the user interface if you were to just get a web link, ou know, that looked li ke a normal Googlc result and yet it takes you to an ther Google search. l think people understand mode changes. They might unacrstand resorting something in a different way. But I think ultimately whfn you click on an individual [web] link, you want to get an answer. You don t want to get another set of search results. 477 In other words, Google's position is that, if a user conducts a search on Google for a product, that user is looking for Google 's search results, not another list of search results from another search provider. However, Googlc has presented no evidence ofu~er expectations in this area. 4 78 Indeed, Google's vertical properties are typically not labeled as "Google" results, and thus, outwardly at least, provide no cue to a user that he or she is "switching" to a different mode of Google search.479 Nevertheless, Brin testifie~ that "the user interface is pretty clear'.480 - "the link that says 'shopping results for' is cle!rly a specialized part of the interface. It doesn't appea r to be just like another web pa e.''481 d. Google's Additional Legal Defenses Setting aside efficiency justifications, Googlc has argued - successfully b1 several litigations- that it owes no duty to assist in the promotion of a rival's website or search platform, and that it owes no duty to promote a rival's product offering over its own product offcrings. 4 82 1ndeed, one reading of Trinko and subsequent cases is that Google ·s privileged in blocking rivals from its search platform unless its conduct falls into in one of ·everal 84 SJ W a i v s nt e m Docu ln sum, Staff acknowledges the difficulties inherent in this area of the investigation, not only because of the legal hurdles we would face, but because ofthe strong procompetitive justifications Google has set forth. We arc faced with a set of facts that can most plausibly be accounted for by a nanative of mixed motives: one in which G ogle's course of conduct was premised on its desire to innovate and to produce a high q ality search product in the face of competition, blended with the desire to direct users to its o n vertical offeri ngs (instead of those of rivals) so as to increase its own revenues. Indeed, t e evidence paints a complex portrait of a company working toward an overall goal of maint3!ining its market share by providing the best user experience, while simultaneously engaging in tactics that resulted in harm to many vertical competitors, and likely helped to entrench poogle's monopoly power over search and search advertising. The determination that Google's conduct is anticompetitivc, and deserving of condemnation, would require an extensive balancing of these factors, a task that courts have been unwilling- in similar circumstances to perform under Section 2. Thus, although it is a close question, Staff does not recommend that the Commission move forward on this cause of action. 2. Google's "Scraping" of Rivals' Vertical Content As described earlier) Staff has investigated whether Google bas w1lawfb ll "scraped" or appropriated - the content of rival vertical wcbsitcs in order to improve its own vertical products, so as to maintain, preserve, or enhance its monopoly power in the markets for search and search advertising. We conclude that this conduct violates Section 2 and Section 5. 86 e m Docu SJ W a i v s nt ticket even ~f compensated at retail price revealed a distinctly anti competitive bent..t97 Appellate courts have focused upon Trinko 's reference to the "unilateral termination of a voluntary course of dealing," as a critical limitation upon a monopolist's discretion in determining whether to deal with a rival. For example, in American Central Eastern Texas Gas Co. v. Duke Energy Fuels LLC,498 the Fifth Circuit upheld an arbitrator's det rmination that the defendant natural gas processor's refusal to contract with a competitor fo additional processing capacity was unlawful. Plaintiff was both a "gatherer" and "processo " of natural gas. The plaintiff alleged that, because it was not economically feasible to open lrs own processing plant, it contracted with the defendant for processing capacity. 499 After two years of using the defendant's processing plant, when the plaintiff entered into renegotiations for additional capacity, the defendant proposed terms that it " knew were unrealistic or completely unviable" to the plaintiff, including a very high price, "in order to exclude [the plaintiff] from competition with [the defendant] in the ... gas processing market."500 The Fifth Circuit upheld the arbitrator's conclusion that the defendant unlawfully refused to deal with the plaintiff, acknowledging that, while courts "must be cautious in finding xception to the right to refuse to deal," here, the defendant's refusal, in the context of a "prio course of dealing" with plaintiff, supported a finding of liability. 50 1 Here, much like in Aspen Skiing and Duke Energy, there is a compelling narrative ---- regarding a prior voluntary course of dealing. Specifically, Google had long-established, - voluntary, and mutually beneficial licensing agreements with both Yelp and Tri_pAdvisor. Through its agreements with these (and other) third parties, Google secured relevant and high-quality content for its web search product. In exchange, through their prcse ce in 88 SJ W a i v s nt me u c o D collaborative networks if we permit one dominant firm to run away with all of the private gains once it is in a position to do so:.s07 There are some distinctions between the conduct in Aspen and Google's conduct here that bear mention. First, the exchange of value here is non-linancial. The bencfi to Google accrues fro m securing high-quality content, while Googlc's partners secure traffit However, this distinction appears to be insignificant. Whether the payment is in the form l dollars or other benefit is of little consequence to the purpose or effect of Googlc's threate ed refusal to deal. sos More importantly, Google ultimately did not "refuse" to deal with Yelp, and their relationship continues to this day. While Google never followed through on its tyreat to remove these websitcs entirely from its web search results, it is clear that Googles threat was intended to produce, and did produce, the desired effect (for a significant period of time), which was to coerce Yelp and TripAdvisor into backing down on their efforts to rave their valuable content removed from the Google Local product. Google's threat also sent a message to the broader marketplace that Google could, and would, use its mono oly power over search to extract the fi11its of its rivals' innovations. Consequently, Google' threat itself - although not a consummated refusal to deal - may be challenged as excl ionary conduct. 509 This theory of exclusion does not reach the search prefcrencing conduct we assessed supra at pp. 78-86. Here, we view the evidence of benefi t to Google stemming from its licensing agreements with third-party content providers as offering the critical distinction. Google's long-standing licensing agreements with parties such as Yelp and TripAd visor offer clear and convincing proof not just of an affirmative relationship between Googl and these I 90 SJ W a i v s nt me u c o D Amazon's product ranking information, which was never part of any licensing agreement between the parties). Under this approach, Googlc's conduct can be analogized to the imposition of higher costs, through onerous terms of dealing, on wcbsitcs wbos content Google deems the most valuable to its own web search product. 510 Viewed in is way, condemnation ofGooglc's conduct depertds not on any prior established relatio ship with the affected vertical websitcs, but rather, on Google's motivation in scraping conte t from highquality verti<.:al competitors- the motivation to keep vertical websites from sip oning users from Google's web search property (and thus, maintaining, preserving, or enha monopoly position in Lhe market for search). While a traditional Section 2 analysis relics on a prior course of dealing as a gatekeeper, or a bright line proxy, for showing that the defendant's purpose and effect was anticompetitive, Section 5 empowers the Commission to demonstrate harm to the competitive process in other ways. 5 11 For example, Googlc's threat (and willingness) to degrade its own web search product- by banishing high-quality vertical wcbsites from its web search results altogether- suggests that Google's motive in scraping high-quality con tent from its vettical competitors was not procompetitive. b. Google's "Scraping" Has Resulted In Anticomt etltive I. Effects As described earlier, Go ogle's "scrapi ng" of the content of rival vertical websites has resulted in harm to these vertical wcbsites and, more broadly, to the competitive process. Because Googlc scraped information over an extended period of time, it is diffi ult to point to declines in traffic that are specifically attributable to Google's conduct. How ver, Google's conduct has arguably lessened the incentives of vertical wcbsitcs like TripAdvisor, CitySearch, and Amazon to innovate. 92 elp, SJ W a i v s nt e m Docu In sum, the evidence shows that Google used its monopoly position in search to scrape content from rivals and to improve its own complementary vertical offerings, to the detriment of those rivals, and without a countervailing efficiency justification. ~oogle's scraping conduct has helped it to maintain, preserve, and enhance Google's mo opoly position in the markets for search and search advertising. Accordingly, we beli ve that this conduct should be condemned by the Commission. 3. Google's API Restrictions Staff has investigated whether Google has employed anticompctitivc contractual restrictions to prevent advertisers from using third-party tools to simultaneously manage campaigns on Google's search advertising platform (AdWords) and rival advertising platforms (e.g., Microsoft's AdCenter). As described earlier, Microsoft has alleged that Googlc is denying Microsoft critical scale by employing these restrictions, and thus impairing Microsoft's ability to compete effectively in the markets for general search and search advertising. We conclude that Google's API restrictions violate Section r- Google's introduction of the AdWords API was a clearly procompetitivj development that benefitted advertisers, SEMs, and Google alike. However, th restrictive conditions in the API usage agreement have anticompctitive effects without off: etting precompetitive benefits. They impede the efficient usc of advertisers' own camhaign data, creating additional, unnecessary transaction costs for advertisers that might wisli to use that data to run advertising campaigns on other search networks. The restrictive conr itions are not inherently tied to the product. Accordingly, we may evaluate Google's inclusion of the restrictive conditions as a stand-alone act and weigh their anticompetitive effects against any potential procompetitive bcnefits.517 94 e m Docu SJ W a i v s nt The restrictive conditions are unreasonable if their anticompetitive effecj outweigh their procompetitive virtues. Ou r investigation has shown that the restrictive cor ditions do not have any procompetitive virtues, whereas their anticompctitivc cf.Tccts, whil difficult to measure, are substantiaL b. The Restrictive Conditions Have Resulted In Anticompetitive Effects The restrictive conditions hann competition in three broad ways. They rl duce innovation, increase transaction costs, and degrade the quality ofGoogle's rivals in search and search advertising. As noted above, severa l SEMs have been forced to remove campaign cloning functionality by Google. Beyond removing these products from the marketplace, Googl e 's restrictive cond itior1s have created a profound d isincentive for tool developers to innovate in this area. A h igh performance cross-network campaign management tool would need to be a soph isticated product, able to allocate and adjust bids on keyword s in different auctions with different and rapidly shifting competitive environments. However well the first-generation tools performed, it seems obvious that their performance would only have imprTed as SEMs and their c lients tested these tools in the fie ld. Google's restrictive conditions stopped this market segment in its infancy. There would be little to no demand for a cross-network management tool without the prospect of accessing the dominant search network, AdWords. Google's imposition of the restrictive conditions has increased the transaction costs for all advertisers other than those large enough to make the internal investment~ to develop their own campaign management tools. 521 For the rest, they must devote additional staff time to manage multiple parallel campaigns. Some may choose to usc work-arounds, by which they download their AdWords campaigns into CSV (or plain-text) files, make th ~requisite 96 SJ W a i v s nt e m Docu Moreover, Google ignores the possibility that even larger advertisers that multi-borne would do so more without the restrictions. As described above, Microsoft's internal studies suggest that advertisers who advertise on both platforms do so unevenly and unequally, thus leading to better, more targeted, and more relevant ads on AdWords than on AdCenter. As described earlier, having the "right" ad for the " right" user at the " right" time is ' ritical to a search engine's abi lity to improve its ad-serving algorithms and its revenue-per-search (or RPS). 529 The lack of smaller advert ism, combined with the lack of regular optr ization by even the larger advertisers who advertise on both platforms, places Microsoft in a significantly inferior position to Google in tem1s of being able to provide that "right" ad for the "right" user at the "right" time. While the magnitude of these effects arc unclear, their direction is clear: advertisers arc spending less on the non-dominant search networks. For advertisers, this means forgone advertising opportunities that presumably would have been profitable, but for the restrictive conditions. For Google's rivals, the diminished spend resulting from the restrict ve conditions means lost revenue, which diminishes their ability to invest in quality improvements in search. The reduction in ads placed also reduces the overall qu lity of the ads served on the rival search networks, which reduces the usefulness of the ads erved to users, reducing, in turn, users' propensity to click on ads, an effect that broadly egrades the quality of the rival search network It is also possible, though more speculative, that reduced ad quality may modestly reduce the usefulness of the rival search engines, particularly on very commercial queries, which in tum may suppress the number of searches performed on the rival networks. The degradation of Google's rivals both as advertising platfQmls and as 98 SJ W a i v s nt me u c o D tools that perform well will lose clients. In fact, even if SEMs and agencies were in no danger oflosing their clients' business, they would still have a strong incentive to improve their clients' returns as a way to encourage their clients to spend more on searc advertising, increasing the third parties' commissions in the process. Jn a round-table discu sion hosted by Google, SEMs and agencies made this exact point to Google. 538 ln brief, the e third parties incentives are highly aligned with Googlc's interests, precisely the oppo ite of what Google contends.539 Google, meanwhile, is unable to identify any concrete examples of any i I effects from the purportedly misaligned incentives of SEMs and agencies. Google has represented to advertisers and agencies that " we have found that advertisers experience higher returns coherent manner." 540 owcvcr, Google has no such cvidence. 541 Go gle did investigate the potential innuence SEMs would have on the rate of spendi ng by their clients, and determined that the spend for advertisers represented by SEMs increased at a higher rate than did spend for other advertisers.542 Google has not engaged in any experiments to determine what effect relaxing the restrictive terms and conditions might havc. 543 Moreover, there is already a different provision in the APT AdWords Te Conditions that adequately addresses any concern about misaligned incentives. sa condition of using the APT, SEMs and other tool developers are required to expose a Googledefined set of minimum func~onality. 544 The required minimum functionality provision directly addresses any legitimate concerns that Googlc might have about SEMs1ailing to expose important features of AdWords to their advertiser clients. Google has n t explained how the required minimum functionality requirement is inadequate in this regar . 100 SJ W a i v s nt me u c o D Google deems important to the performance of AdWords. Although Google now claims that the required minimum functionality condition and the restrictive conditions are both aimed at the lowest common denominator concem,550 it cannot explain why the required minimum functiona lity requirement alone would not suffice to alleviate the lowest comm n denominator concern_ Indeed, this document suggests that the restrictive condi ·ons were actually designed specifically to reduce the likelihood that advertisers would ex end their campaigns to rival search networks. 55 ' In sum, the effects of these restrictive conditions, combined, have the te dency to preserve and enhance Google's dominant position in the search advertising market. Unjustified by any procompetitive benefits, we believe that Googlc's restrictive conditions should be condemned by the Commission.552 4. Coogle's Exclusive and Restrictive Syndication Agreements Staff has investigated whether Googlc has entered into anticompetiti ve, exclusionary agreements with websitcs for syndicated search and search advertising services (AdSense agreements) that serve to maintain, preserve, or enhance Googlc's monopoly power in the markets for search, search advertising, or search and search advertising syndica ion (search intermediation). We conclude that these agreements violate Section 2. a. ~ Google's Agreements Foreclose a Substantial ortion of the Relevant Market Exclusive deals by a monopolist harm competition by foreclosing rivals !from needed ~ relationships with distributors, suppliers, or end users. For example, in Microso t, then.J defendant Microsoft 's exclusive agreements with original equipment manufactu ers and software vendors were deemed anticompetitive where they were found to preve t third parties from installing rival browser Netscape, thus forec losing Nctscape from t e most 102 SJ W a i v s nt me u c o D party dataset routinely used in the industry to analyze query volumes and market shares. 561 As noted earlier, however, in a company data set provided by Microsoft, Yaboo! 's syndicated query volume is significantly higher than that reflected in ComScore.562 Relianc on the larger figure would clearly result in a dramatically lower foreclosure number for oogle's agreements. We are trying to get to the bottom of this discrepancy now. Howev r, based on our broader understanding of the market, we believe that the ComScorc set more accurately reflects the relative query shares of each party. 563 Below, Staff lays out three scenarios: the most conservative foreclosure scenario; the most aggressive foreclosure scenario; and the "intermediate"- or most likely defensibleforeclosure scenario. ln our most conservative estimate, the foreclosure rate is approximately 20 pcrcent.564 In our most aggressive estimate, the foreclosure rate is approximately 66 perccnt. 565 In the "intennediatc" scenario, the foreclosure rate is approximately 52 percent.566 Obviously, given the limitations of the various datasets, the calculated foreclosure rates arc of limited value. Nevertheless, it is clear that Google has tied up a substantial portion of this distribution channel with exclusivc and restrictive agreements. ln he market for search syndication, Google has exclusive or restrictive agreements with 12 of the top 20 companies (60 percent) and 4 of the top 5 (80 percent). The 20 largest companies account for 94 percent of total query volume.567 Courts have found that foreclosing rivals from the most effi cient means of distribution can be especially problematic.568 Access to these largest players is by far the most efficient method for Bing to gain query volume in the syndication channel. 569 104 SJ W a i v s nt e m Docu the publishers we interviewed did not object to exclusivity because they wanted to use Google for all their search syndication needs anyway. Our investigation indicates that this objection rests on a fallac ious assu ption: namely, that Bing's average monetization gap is derived from its consistent fai ings acrossthe-board. If, instead, that overall average is derived from sources of differing uality, that means Bing actually does have opportunities to pick off incremental business fr m Google in those areas where the monetization gap is lower, particularly where it can make up for some of its monetization deficiencies by otlcring higher revenue shares. Evidence fr m Microsoft indicates that there is indeed heterogeneity in the quality of its search advertising product, with comparative strength in certain c~mmercial categories, such as travel and people (social) scarch.571 Given this state of affairs, one likely path for Bing to win new syndication business is precisely the one blocked by the exclusivity provisions in Google's syndication agreements. All the publishers that expressed interest in using Bing told us that they want to split up their business, giving Bing opportunities where it can compete, and relying on Googlc for the balance of their needs. In addition to the immediate impact on Bing, our investigation suggests hat specialty search advertising platforms may emerge in the absence of Google's exclusivit provisions. For example, lAC's CityGrid property sought to build its own advertising platfQrm to serve advertising targeted to local markets. 572 CityGrid monetizes its websites from small''mom and pop'' stores, medium-sized businesses, and large chains to gain local eustomers. 573 CityGrid decided that it wanted to build its own network rather than "put all [its] eggs in one basket" by going with Google 106 are trying sing .IVC Iy. 574 SJ W a i v s nt me u c o D 5 to I 0 percent increase in its overall query traffic would be "very meaningful" because Bing is at the lower part of the scale curve where "each percentage point is critical."579 While there is not enough evidence on this point to reach definitive cone usions, internal Google documents suggest that Microsoft's view of things may be close to the truth. Googlc's interest in renewing deals with some of its largest syndication customers may have been, in part, to keep Microsoft from gaining scale. For example, an internal Google analysis of the 2010 AOL renewal explains: s marginal search share but represents sea e gains for a Microsoft + Yahoo! partnership.... AOL/Microsoft combinatiof has modest impact on market dynamics, but material increase in scale of Microsoft's search & ads platform. 580 When a senior Googlc executive was infonned that " Microsoft [is] aggressively wooing AOL with large guarantees," he responded that: I think the worse case scenario here is that AOL users get sent to Bing, so even if we make AOL a bit more c~mpeti tive relative to Google, fhat seems preferable to growing Bing.,') 81 Ac ord ing to Googlc documents, the company sought to pursue the AOL deal algressively ,582 While the evidence summarized above is consistent with the theory that esc exclusive dealing arrangements are creating anticompctitivc effects, there are nevertheless some significant limitations in this evidence. Perhaps our biggest concern is that today, so few publishers arc actively interested in using multiple suppliers. As noted earlier, we have identified only three companies that are subject to the exclusivity or "preferred placement" provisions today and clearly voicing unambiguous concerns: cBay, NexTag, and Business.com. In addition to these three companies, AmaLon is not foreclosed , but voiced very similar concerns and is very worried that it may be subject to cxclus vity in the 108 SJ W a i v s nt me u c o D dynamics as publishers have the opportunity to consider - and test - alternatives to Google's AdScnsc program. While the speed and strength of these long-term improvements cannot be accurately forecast today, this is a situation where the ncar-term competitive imp cts may be overshadowed by the long-term improvements, as competitive forces arc unleas led and additional dynamism emerges. c. Google's Agreements Are Not Justified By Effic'encies Google has offered three business justi.ftcations for its exclusive and restr clive syndication agreements with publishers. First, Google notes that there is a long- tanding industry practice in favor of exclusivity dating from the time when the publishers demanded large, guaranteed revenue share payments regardless of actual performance. However, guaranteed revenue shares are now virtually non-existent. A second, and related. justification is that Google is simply engaging in a vigorous competition with Microsoft for exclusive agreements. Although Microsoft asserts that it would like the opportunity to compete on a non-exclusive basis (and will happily serve even a sma ll portion of a website publ isher's queries), some publishers report that Microsoft itself sought various forms of exclusivity in contract negotiations.m Moreover, while r icrosoft has aggressively pursued some very large website publishers, it appears that Mic osoft is not generally pursuing the broader syndication busi11ess Loday. 586 Google may argue that the fac t that Microsoft is losing in a competitive bidding process (and indeed, nol compe ·ng as ' vigorously as it might otherwise) is not a basis on which to condemn Googlc. However, I Google has effectively created the rules oftoday's game, and Microsoft's subs tial mone tization disadvantage puts it in a poor competition position to compete on a aU-ornothing basis. 110 SJ W a i v s nt e m Docu While we acknowledge the limited effects here, it is worth noting that the market for search and search advertising syndication is, inarguably, not robustly competitive today. Googlc has been unilaterally reducing revenue share percentages to many of its syndication customers (in effect raising prices) with apparent impunity.589 One of the largest customers, Amazon, decided that it is in its long-term, strategic interest to fu nnel some que1 volume to Bing, even if it is losing money on each query. 590 Ama7.on is using multiple sup~liers just to try to foster a more competitive marketplace.59 1 Where markets are functioning Jo poorly, the rationale for government intervention is stronger, even in situations where th near-term competitive harm directly attributable to the challenged conduct may be small. Although this conduct presents a closer question, we believe that Google's exclusive and restrictive agreements have not only helped to maintain, preserve, and enhance Google's monopoly power in the market for search and search advertising syndication (search intermediation), but also in the underlying markets for search and search advertising. Therefore, we believe Lhat the Commission shou ld condemn Google's exclusive and restrictive syndica ion agreements. IV. POTENTIAL REMEDIES Staff has identified several possible remedies to Google 's conduct. These remedies arc described below. A. Scraping There are at least two possible remedies for Google's scraping conduct. First, Google could be required to provide an "opt-out" feature to remove "snippets" of website content (e.g., user reviews, ratings) from Google's vertical properties, but retain 112 ose SJ W a i v s nt me u c o D partners considering alternatives to AdSense may grow in the event that these agreements are enjoined.595 V. LITIGATION RISKS We have identified throughout this memorandum the many substantial ri ks associated with bringing a case against Googlc. On a global level, the record wit penni! Googlc to show substantial innovation, intense competition fi·om Microsoft and ~thers, and speculative long-run harm. Here, we highlight some specific facts that present t e greatest litigation risk: I. "Competition is just one click away. "596 Googlc does not charge consumers, and they are not locked into Google. The durability ofGoogle's monopoly power is questionable with an increasing number of wcbsites (e.g., Faeebook, Twitter) competing for user time and advertiser dollars. 2. Universal Search is a "product improvement" that has resulted in substantial benefit to its users. 3. Google's organization and aggregation o[ content from other wcbsites adds value to the product for consumers. 4. The largest advertisers (that produce the most revenue on Google's AdWords platform and Microsoft's AdCenter platform) already advertise on both AdWords and AdCenter. 5. The most efficient channel through which Bing can gain scale is Bing.com, not syndication or other distribution channels. 114 SJ W a i v s nt me u c o D VI. CONCLUSION Staff concludes that Google's conduct has resulted - and will result - in real harm to consumers and to innovation in the online search and advertising markets. Google has strengthened its monopolies over search and search advertising through anticompetitive means, and has forestalled competitors' and would-be competitors' ability to cha lenge those monopolies, and this will have lasting negative effects on consumer welfare. Sp~cifically, StalT believes that: 1. Google has unlawfully maintained its monopoly over general searcJ1 and search advertising, in violation of Section 2, or otherwise engaged in w1fair methods of competition, in violation of Section 5, by scraping content from rival vertical websites in order to improve its own product offerings. 2. Google has unlawfully maintained its monopoly over general search, search advertising, and search syndication, in violation of Section 2, or otherwise engaged in unfair methods of competition, in violation of Section 6, by entering into exclusive and highly restrictive agreements with we publishers that prevent publishers from displaying competing search results r search advertisements. 3. Google has unlawfully maintained its monopoly over general search and search advertising, in violation of Section 2, or otherwise engaged in unfair methods of competition, in violation of Section 5, by maintaining contractual restrictions that inhibit the cross-platform management of advcrtis~ng campaigns. 116 SJ W a i v s nt e m Docu 1 See Google Inc., File o. 111-0163, Resolution Authorizing Usc of Compulsory Process in Nonpublic Invest igation (Jun. 13, 20 II). 2 In total. the Commission has issued 20 subpoenas (to Googlc, Microsoft, Yahoo!, Amazon, eBay, NexTag, Thefind, Living Social, Yelp, Apple, Motorola Mobi lity, Samsung, Sony, Toshiba, LG Display, RIM, AT&T, Sprint Nextel, T-Mobile, and Verizon) and two voluntary access letters (to Expedia and Trip AdJisor). 3 TI1e investigational hearing of CEO and co-founder Larry Page, originally scheduled for Jun. 2~, has been delayed indefinitely due to the illness of Mr. Page. StafPs last scheduled investigational hearing a Google executive, Andy Rubin (Android founder and head ofGoogle's Android division), is slated for Aug. 23. 4 Letter from Joaquin Almunia, Vice-President o f the European Commission, to Eric Schmidt, G6ogle, dated May 2 1, 20 12 (copy of the letter is on file with Staff). I rr s !d. See Discussion Paper Submitted by Google on The Preliminary Concerns Identified by the European 6 Commission and Google's Proposed Solution, attachment to Letter from Maurits Dolmans eta!., t:leary Gottlieb Steen & Hamilton LLP, to Cecilio Madero Villarcjo, Deputy Director General, Europeal Commission (Jun. 30, 201 2) ("Google-EC Settlement Proposal"). 'The State of Mississippi is also conducting a separate investigation into Google, but is not wor~ng with the multi-state b'roup or w ith the Commission. The Commission declined to grant access to M ississi~pi due to the state's retainer of an outside law firm to conduct the investigation and the multi-state group's denial of access (on the same basis). 6 The states have jointly retained economist Rick Flyer as a consulting expert and, potentially, as a testifYing expert. 9 Kindemart.com LLC v. Google, Inc., 2006 U.S. Dist. LEX IS 8248 1, (N.D. Cal. Jul. 13, 2006); Search King, inc:. v. Google Tech., inc., 2003 U.S. Dist. LEXIS 27193, (W.D. Okla. May 27. 2003). 10 Person v. Coogle, Inc., 2007 U.S. Dist. LEXIS 47920. ( .D. Cal. Jun. 25, 2007); Google, inc. v. myTriggers.c:um, inc., Franklin County Ohio C ivil Division Case No. 09cvhl 0- 14836 (Aug. 31, 2011); TradeComer.com, U.C v. Google, inc., 693 F. Supp. 2d 370, 37R (S.D.N.Y. 2010) (a.ff'd, TradeC:omei.com LLC ''· Coogle, Inc., 647 f.3d 472,478 (2 nd Cir. 2011). 11 2006 U.S. Dist. LEXIS 82481, (N.D. Cal.Jul. 13, 2006). 1 ~ 2003 U.S. Dist LEXIS 27193, *2 (W.O. Okla. May 27, 2003 ). u !d. 14 Kinderstart.com, 2006 U.S. Dist. LEXIS at *28. 15 693 F. Supp. 2d 370,378 (S.D.N. Y. 2010). 16 Case No. 09cvhl 0- 14836 (Frankli n County Ohio C ivi I Division. Aug. 31, 20 t I). 17 2006 U.S. Dist. Ct. Pleadings 7297, *3 (N.D. Cal. Apr. 16. 2007). 1 ~ ld. at* 10. 19 Person v. Coogle. Inc., 346 Fed. Appx. 230, 23 1 (9th C ir. 2009). 20 Google owns and operates numerous websites, including: Google Alerts; Books: Finance; Gmail; Images; Maps; News; Google Plus; Product Search; and YouT ubc. 21 In a separate investigation, opened in Apr. 2012, FTC Staff is investigating whether Google viqlated commitments to various standard-setting organizations to license standard essential patents used \n the mobile industry on fair, reasonable, and non-discriminatory tcm1s. See Google-Motorola, File No. 121-0120, Resolutio n Autho rizing Use o f Compulsory Process inl\'onpublic Investigation (May 25, 2012). 22 Google FY 20 12 Form 10-K (Jan. 26, 2012), at 29, available at http://scc.gov/Archives/edgar/dma/1288776/ 000 119312512025336/d260 164d I Ok.htm#toc260164 8. ("G oog1e 2012 IOK"). 23 /d. at 25. 24 Press Release, Microsoft Corp .. MSN Significantly Upgrades MSN Search for Consumers with Major Performance and Relevancy Improvements (Jun. 30, 2004) http:!lwww.microsofl.com/ prcsspass/press/20041jun04106-30ImprovedSearcb2004PR.mspx. 25 Press Rele~tse, Microsoft Corp., Microsoft's New Search at Bing.com Helps People Make Bett r Decisions (May 28. 2009) http://www.microso rt.com/presspass/ press/2009/ may09/05-28NewSearchPR.msqx. 26 See Microsoft Corp., Complaint to the European Commission (Mar. 31. 2011) ('"Microsoft EC ubmission"). 118 e m Docu SJ W a i v s nt H Varian Tr. 88:16-89:7 ("Brand perception is driven to some extent by other [non-search] forms of advertising ... We know. for example, that display ads drive brand; brand drives clicks"); Schmidt Tr. 130:18-21 (a brand advertisement would be a Coca-Cola advertisement that is not trying to get you to buy a Coke, but trying to get you to think about Coke); Wal-Mart IR (Jan. 23, 2012): Fox Studios 1R (Jan. 20, 2012); Verizon IR (Nov. 1, 2011); EAS 1R(Feb. 23, 2012). 44 See Statement of Federal Trade Comm'n Concerning GooglefDoubleClick, FTC File No. 071- 170 (2007). at 5. See also lnterpublic lR (Oct. 20, 20 I I ); Didit.com !R (Dec. 27, 20 II ). 1 45 ContextuaJ ads are somewhat more successful at creating conversions than direct display ads, jut less successful than search. Contextual advertising is considered a c loser substitute for display adver ising than for search advertising in terms of function and performance. See, e.g, Group M TR (Oct. I I, 20 II); Expedia IR (Jan. 23, 20 12). See also, e.g., GOOG-ITA-03-0045511-17 (2009), a t 16 ("content conversions do not lead to sales Iike search conversions"); Brin Tr. 181:1-8 ("the conversions arc different. The click-t.hrot gh is also different .... So between the two of those, your average content page view is worth signiiicantf less than yo ur averagt: s~.:arch page, no question about it"). 46 Sec Didit.com IR (Dec. 27, 20 II); GOOGFOX-000073028 (2008), at 14. "Re-targeting" me ns serving ads to users that have abandoned purchases before completed, or who have visited ce1tain websites in the past. Like search, this type of ad is meant to elicit a direct response, but - unlike search ads- re-targeted ads are not shown in response to a user's declared intent Clickable IR (Oct 24, 2011) (re-targeted disp lay advertising requires advertisers to act on behavioral calculations and inferences from large troves of data, and does not generate leads or sales as well as search advertising). H See Didit.com IR (Dec. 27, 2011); Wai-Mart IR (Jan. 23, 20 12). Social media advertising appears to be more like display advertising in that it offers a large volume of impressions, but relatively few conversjons. See Facebook 1R (Jul. l, 20 I I); Facebook IR (Jan. 24, 20 12). See also Matt Lawson, How ro fmegrate Search and Social Media for Beller Resulls, Yiashable, Apr. 1, 2010, /um:l/mashable.com/2010/04/0llpaid-search-socialmcdia/ (Director of marketing for Marin Software discussing how to develop and integrate paid search and social media advertising strategies; social and search advertising are '"two distinctly different tactics- the bidbased, conversion-obsessed. ROI-driven world of paid search and the experimental, brand-building, bard-tomeasure world of social .. . each provide different benefits to your business, so you should leverage their strengths instead of trying to get them to deliver results that aren' t suited to the medium. Market,ers usually participate in social media to create an active dialogue with consumers around their products and services, with the main goal ofbuilding brand value, and a secondary goal of driving sales. On the other hand, marketers use paid search primarily to drive sales, leads, and conversions, and don't expect the s hort text ofth,r paid search ads w do much for brand ing''). 48 Contextual advertis ing is limited by the amount of advertising space ava ilable on web pages addressing any g iven topic, in which relevant ads can then be served. Re-targeted (or behavioral) adve rtising is } mited by the number of"cookies" users allow to be placed on their computers (and on how o ften those cooki~s are erased), and also requires guesswork and heavy analysis on the part of the advc11 iscr. See GOOG FOX-Q00073028 (2008), at 13; lnterpublic IR (Oct. 20, 201 1); FTC-EBA V-00000002 (20 12), at 31 (eBay and Shopping.com spent an ''insigni ticant" amount on contextual advertising). FTCNcxt-00000002 (2012), at 36 (nbn-search advertising cannot rep lace search advertising). Social media is sti ll a maturing market, which remains quite small. Moreover, neither Facebook nor Twitter has been very successful in genera ting conversions, despite the information they have available on the interests of their users (see Faccbook IR (Jul. l , 2011; Ja~. 24, 2012); Twitter 1R (Dec. 13, 201 1)), and both Living Social and General Motors have pulled the majority of their social media budgets based on u failure to achieve acceptable conversion rates. See Living Social CID Response (20 12), at 17; Joan Muller, GM Says Facebook Ad~ Don't Work. Pulls $10 Million Account, Forbes, May 15, 20 12, available at http://www.forbes.com.isites/ joannmuller/20 12105/15/gm-says-facebook-ads-dont-workpu lls-1 0-million-accountl. 49 Braddi Tr. 11:22-12:2. 30 !d. at 26:8-27:8. 1 ~ Staff continues to investigate Google's conduct in the mobile arena. and will address these issues in a supplemental memorandum. 32 Google purchased the Android business in 2005 . .lJ Since Google's release of the first commercially ava ilable mobile device running Android OS n October 2008, Android's market share has grown exponentially. ln Sep. 2009, Apple garnered 24. 1 perc nt share of 1 120 SJ W a i v s nt e m Docu sx See Brio Tr. 142:3-144:9, 169:1-19 (Google tracks user clicks to improve quality, citing early NavBoost algorithm as example of signal that relied heavily on user clicks); Schmidt Tr. 6 1:17-24 ("So clicks matter in tenns offeedback to the people who monitor these things. They say our algorithm needs to be improved"). See also e.g .. GOOGPAGE-000004652 (2008) ("[Click-tracking is) used to track which search resuhs a user selects. That information then feeds back into our search ranking"); GOOGBRJN-000005558 (2002), at 9 ("'Traffic/Quality Effect. The more traffic we generate and usage data we collect, the better ourtverall [ad) quality."); GOOCMAYE-000044916-21 (2004), at 18 (Brio notes that "(w]e could take advant ge of our scale more. [H) ave I 000 or 10000 people feeding information into our algorithms"). s9 Manbcr Tr. 54 :5-56:15 (describing various uses for experiments); Declaration of Satya Nadell , Senior Vice President, Online Services Division, Research and Development, In Rc Google/ITA (Dcpartmen~ of Justice) (2011 ), at4 ,,, O(d) ("Almost all innovations on the SERP ... go through a formal experimentati?n process before they are released, and often there are several rounds of experimentation") ("Nadella Decl.f ). 6 FTC-000023(,..44 (2005), at 38 (2005 Founders' Letter). 61 Manber Tr. 57:15-23 (when Manber mn the search quality team, Googlc was running approxi1 ately 5,000 ex perimcnts a year, or about 15 experiments per day, simultaneously); Brin Tr. 160:2-9 (multipl experiments are run simultaneously, with each typical experiment using approximately one to two percent of total user volume). Microsoft runs approximately ten experiments simultaneously. Microsoft Corp., "Mi osoft Response to DG Comp RFI" (Nov. 21, 20 11), at 78. 62 Susan Athey, "Scale in Online Search" (Mar. I 0, 2012), at I 0-1 1; Nadella Dec!. at 4 ~IO(d). 63 Susan Athey, "Scale in Online Search" (Mar. 10, 2012), at I 0-1 1 (''Today, Microsoft has relatively few users it can use for experiments and there is a limit to the number of parallel experiments that a single query can be ~an of without compromising the robustness of the results"). Microsoft asserts that additional query volume will also help its algorithms to detem1ine what web pages to crawl and index. based on observed user interest of similar web pages. Susan Athey, "Scale in Online Search" (Mar. 10, 2012), at 6-7; Microsoft Corp., "Microsoft Response to DG Comp RFI" (Nov. 21, 201 1), at 63 ("Queries are a critical component of the user data necessary to identify and rank URLs and documents for inclusion in a search index"). Moreover, while Bing maintains an index of approximately 43 billion documents (as of l\ovember 20 II), it "serves'' only 16 billion of those documents. The remaining 27 billion web pages have not been clicked on recently enough (if ever) to give Bing's algorithms a sense as to "whe'1er they are suitable" or relevant to user queries. !d. at 63. Google served more than 200 billion doeuments~ lat last estimate, accordi ng to Sergey Brin, who testified that Googlc reached this point several years ago. Brin T . 339:14-23. lt does not appear that Google relies on query volwne in order to determine what to index. Udi M nber testified that Google indexes everything it can. Manber Tr. 34:24-25. 6 s CX-1 29 (GOOGMANB-000029871-75) (2009), at 73. 6Q See Schmidt Tr. 119:24-120:8 (" .... Think of it this way, advertisers don't put in one ad. Th y put in a thousand ads against different keywords and different combinations. So if you have a thousand dvertisers and a thousand such combinations, you have a million ads that you can choose from. So that's clearey better than having a hundred ads - right- because you can [pick) the one which is you know, the person .Who wanl<; campi~g equipment that's blue i~1 New Hamps~ire"); Bri~ Tr. 192:10-1.4 ("IIaving a good sele~t\on ~f advertisers to choose from defimtely helps havmg the option of producmg a good ad, no questiOn"); td. at 193:20-24 (agreeing that having more ads means that Google is more likely to have the right ad for the right user at the right time). See also, e.g, CX-81 (COOGROSE-000013304-12) (2004), at 6 ("More advertisers (and the ads they bring with them) increase overall ads quality by increasing the number of total choices.' This is yet another exmnple of a positive feedback and/or scale effect"). 67 See Schmidt Tr. 73 :2-23 ("Having more advertisers fills out your otTering.... [l]fyou have one advertiser, only one, and then the ad is - is - the ~'rong ad - obviously, more advertisers up to some point of diminishing returns does acrually kind of fill out your ponfolio"); Wojcicki Tr. 110:16-22 ("Well, I think whTn we have more advertisers we're able to cover more topics''). See also e.g., GOOCBRJN-00001977 1 (tmdated, c. 2004), at 51 ("More advenisers improves partner monetization: more ads on more queries (coverage, CTR). More competitive auction (CPC). Overall, higher moneti2ation (RP\4)"). 6 x See Urin Tr. 171:24-173:6 (Google relies on what ads a user clicks on and how the user engag s with the ad to determine whether to show an ad, how to rank the ad, and how to price the ad); Schmidt Tr. 7 · : 13-22 (more ads gives a search engine "more at-bats," or "more opportunities to show that ad"); Wojcicki Tr. 104:17-19, I05:20-106:9 (testifying that "we determine relevance mostly by do we see the users have clicke on these ° 122 e m Docu SJ W a i v s nt publishers and advertisers"). C.f GOOGFOX-000025982-83 (20 I 0), at 82 (noting that recent p ess article is "premised o n the notion that MSFT and Yahoo are not able to take full ad revenue advantage ofjeir search query share, which may be true"). 711 See Schmidt Tr. 74:3-8 (agreeing generally with the concept of the "virtuous cycle," and testifying that "[t) hese are scale business(es). You want to get to scale. . .. Larger indices; more advertisers; o~viously, more revenue; more reach .. . those sorts of things."); id. at 85:8-87:20; Brin Tr. 225: 17-227:4 (agreeif"g generally with the concept of the "virtuous cycle"). See also Preston McAfee, Presentation, "Scale, Data, pnd Machine Learning: Solving the Search Problem" (20 I I), at 6 ("scale, liquidity, and access to data results~ a virtuous cycle"); Microsoft EC Submission at 17 ("for smaller search engines, scale generates a 'virtuous cycle' t.h.at mpidly improves quality"). 77 Sea Schmidt Tr. 178:17- 179:5 (" .. . . There's some evidence ... that we're past the point wh re there's any particular benefit of using the user . .. inforn1ation to improve [search quality on tail queries). In other words, we have enough users already that more users don't make it much better."); id. at 284:3-286:1 8 Qsame); Brin Tr. 145:7-153:6 (discussing scale curve and diminishing returns; testicying that, while data sources i e "still valuable, but you know, ... you'd have to like double or ten times them to get you know, materi lly better''; agreeing that Google' s search quality will no t improve significantly based on additional queries oday; and testifying that ifGoogle had 10-20 percent fewer queries today, this would create a "pretty margfnal difference" in search quality); Manber Tr. 150:14-23 ("Well, obviously, after a while, there's a diminishing re turn for data."). See also e.g., CX-129 (GOOGMANB-000029871-75) (2009), at 73 (Google chief economist Hal Varian argues that increases in data are subj ect to "diminishing returns''): Michael L. Katz et al., "An Economic Analysis of Microsoft 's Allegations that Google's Conduct Hanns Competition by Reducing Birtg's Scale" (May 14, 20 12), at 46 ("Benefits of scale in search are subject to diminishing ren1rns. Click-and-query data are an important input to Google·s search algorithms, but the value of incremental data in providing relevant search results decreases as the amount of data available to those algorithms increases"); id. at I 04 ("the effect of incremental advertisers on search monetization are subj ect 10 di minishing returns"). 18 Brin Tr. 154:5-14. Brin did not state this premise as a mathematical certainty, only as an illustration of the "diminishing returns" curve. Preston McAfee, Yahoo''s former chief economist, suggested that "having 2-3 times as many user observations," particularly for ''tail" queries, would result in substantially more than a one percent increase in quality- indeed, doubling a search engine's queries WOuld be "an enorn10US rdvantage." McAfee suggested that a 3-to- 1 advantage in query volume could result in a 70 percent increase ~n "precision" for thlll search engine's ability to answer unique queries. Preston McAfee, Yahoo!, Presentatio1 "Scale, Data, and Machine Learning: Solving the Search Problem" (20 II ), at 8. 79 See Brin Tr. 154: 15-158: 18 (testif)'ing that, based on publicly available infonnation of Microsoft's query volume, he doesn't believe that additional query volume would significantly improve Microsoft'~ search quality). See also Michael L. Katz et al., "An Economic Analysis of Microsoft 's A llegations th~t Google's Condtrcl Tlarms Competition by Reducing Bing 's Scale" (May 14, 20 12), at 47 (arguing that, "because of the diminishing value of additional click-and-q uery data and Bing's substantial and growing query \jolume, it is unlikely that query data from Google's exclusive syndication and distribution arrangements wou~d provide any considerable value to Bing"); id. at 104-105 (Mierosofi already has a significant number of advertisers; any increase in ads volume or clicks would result in insignificant additio nal yield). 80 See Microsoft EC Submission, at 26 ("The marginal relums for additional scale decrease once a platfonn reaches a certain scale") (Mar. 31, 20 11); Susan Athey, '·Scale in Online Search" (Mar. 10, 2012), at 9 ("as query volume grows, RPS grows quickly at first and then becomes flatter, as more and more of the most importnnt advcrti ~crs have already been attracted to the platfom1"). Rr Mic rosoft estimates that, in 1997, the size of the world wide web was approximately 200 million web pages; by 2008, the figure wa~ approximately I trillion web pages; and today, there are anywhere between 5 and 2.0 trillion web pages. Susan Athey, ·'Scale in Online Search" (Mar. I 0, 20 12), at 11 . See also e.g., Schmidt Tr. 33: 15-25 (''the mte of growth of the Internet appears to be accelerating, so it's getting- it's getting worse faster, if you will, prinlarily because of generation of ... user content"). 8 ~ Susan Athey, "Scale in Online Search" (Mar. 10, 20 12), at I I. MJ Jd. To this end, Microsoft conducted an experiment in 2008 that tested the effect on user enga~ement of reversing algorithmic improvements. Microsoft found that, when it moved back to two-year-o1d algorithms (essentially eliminating two years' worth of user data), the search engine "s ignificantly reduced f1ser engagement" with Microsoft 's search engine. Jd. at 13. Google came to the same conclusion w en it removed 124 e m Docu SJ W a i v s nt Sherman, Coogle Introduces Book Searches, Search Engine Watch. Dec. 16, 2003, http://searchcngincwatch.com/article/2065619/Google-lntroduccs-Book-Scarches (Google launched Google Print on Dec. 16, 2003):Tbe abo'e citations are linked-to from, Google's Official Webpage. Our History In Depth, http://ww\\ .google.<.-om/about/company/history/#>20 12 (last visited Jul. 3 1, 20 12). 100 Mayer Tr. 67:3-18. 101 GOOG-Texas-1325832-33 (2010), at 33. 102 COOG-Tcxas-1486915-70, at 28-29. llmerous documents demonstrate Google's recogniti n of thjs vertical threat. See, e.g., GOOG-ITA-05-0012603-16 (2009), at 4-5 ("Some vertical aggregator are building brands and garnering an increasing% of traffic directly (vs. through Google): ... Strong content!is improving aggregator organk rankings and generating higher quality scores, giving them more free and/or \ow-CPC tratlic; . .. A growing% of finance/travel category queries are navigational vs. generic (e.g., so'1thwest.com vs. cheap airfare). This demonstrates the power of these brands and risk to our monetizable traffic"); COOG-ITA04-0004120-46 (undated, c. Feb. 2009), at II (presentation discussing the "vertical specialist ch!lenge," and noti~g that the "potent~a l thr~a ts to Google" incl ~ded ~·generic prod~ct sea~che.~· moving fro~11 G ogle ... to Vert1cal aggregators," 'Vert1cal Aggregators takmg lugher share of last cl1cks before sale,' an "merchants increasing% ofspend on aggregulors . . . vs. Google") (emphasis in original). CX-158 (GOOG ITA-060021809-13) (2005), at 10 (email from Bill Brougher, a Google product manager, stating, "what ·s the real threat if we don't execute on verticals? (a) loss of traffic from google.com because folks search elsewhere for some queries; (b) related revenue loss for high spend verticals like travel; (c) missing Opt)' if someone else creates the platform to build verticals; (d) if one of our big competitors builds a constellation of high quality verticals, we are hurt badly."); GOOGWRIG-000069488-524 (2008), at 489 ("Google's core business is moneti..:ing commercial queries. If users go to competitors such as Amazon to do product queries, long-term revenue will suffer."); GOOG-Tcxas-0274944-47 (2009), at 44 (discussing creation of a slide for the Google Board of Directors about verticals from a search perspective, i.e., "users going to aggregators rather than [G]ooglc for specific queries'' and an ads perspective). Ol See, e.g.• GOOG-JTA-04-0063246-55 (2009), at 47 (presentation laying out "four key vertical growth opportllnities;• including finance (EU), travel, local, and retail). Most recently, Google bas introduced its own social product, Google Plus, which competes with Faccbook, Twitter. and other social networking sites. See Twitter IR (Dec. 13, 2011); Facebook IR (Jan. 24, 2012). 1 ~ By V1arch of2004, Googlc had launched the Beta form of a local vertical property to handle tbcal queries. Juab Carlos Perez, Google Offers New Local Search Service , Info World, Mar.l7, 2004, http://www. in foworld.comit/ap pl icat ions/ QOO glc-o ffcn;-ncw-1oca1-scarch-scrvicl:-56 I . ' 65 See, e.g, COOG-Tcxas-0197410-14 (2008). at I 0 (preparing presentation for executives showing Amazon queries increasing and Google's fl at or declining, as "a strategic justification for the Product Search Movement"); GOOG-Tcxas-0971713-27 (2008), at 13 (presentation discussing investing in content acquisition to w_in mn~s/loca l).. For example, in shopping, Go ogle comm i~ted to. ma~si_ve inv.estment, introd~ced a new vers ron of 11s shoppmg vert1cal, and mtroduced new ways of d1splaymg mformatton from the v_e~1cal. In 2006, Cioogle decided that its comparison shopping site, Froogle, hnd failed, and decided to stop wor~:g on or promoting Froogle, while it worked on its new shopping product, Google Product Search. GOOf EC0076341-42 (2006), at 41; GOOC-Tcxas-0213227 (2006). Accordingly, the traflie to Froogle fi II dramatically. Google launched Google Product Searc h to re place Froogle in Apr. 2007. Danny ullivan, Goodbye Frougle, Hello Google Pmduct Search, Search Engine Land Apr. 18, 2007, http://searchemrineland.com/goodbye-frooele-hello-googlc-product-scarch- 1 I00 I; GOOG-Tex~s-0216363 (2007) (discussing launch of Coogle Product Search the following week). Shortly afterward, in fyfay 2007, Google launched the product llniversal. See Press Release, Google Inc. Google Begins Move to Universal Search, Release (May 16, 2007), previously available at http://www.google.com/intllen/press/pressrel!universalsearch 200705 16 (accessed Feb. 1, 2012, ince removed from this URL. copy saved by Staff); David Bailey, Anln:.ider "s View o.fGooxle ·s Universal Se rch, previously available at http://searchengineland.com/an-insiders-view-of-google-universal-search (accessed February I, 20 12, since removed from thjs URL, copy of article saved by Staft). 06 See CX-157 (COOG-Texas-0213579-580) (Oct. 5, 2009), at 580 (including slide for Google oard presentation, "Vertjcal Search: Makillg progress on multi ple fronts." identifying Google venicals in images, books, products. news, travel, health, real estate, finance, and mortgages). Google launched G<>ed in many vertical areas. Regarding Google Product Search, see, e.g.. Mayer Tr. 272:7- 277:8 (explaining that Google used the occurrence of comparisons opp engines at positions 1-3 in the web ranking to boost Google's product universal to position one, because a C . E would appear if it has a highly relevant product to the q1.1ery, and, thus, Google Product Search must also ave a higll.ly relevant prod\1ct) (citing GOOG-Texas-0214363 (2009)). See alw GOOGLR-00161 8-8Q. (2009),.at (launch report entitled "Product universal top promotion based on shopping comparis n [site]prescnce"~hat relies on a 162103 (20~9) (listing list of"blessed sites" to trigger top promotion of product universals); GOOG sites). Regarding Google Local, a local sites trigger - using, for example, Cit Search and Yelp appears to have been introduced in 2007, see GOOGLR-00297666-69 (2007), at 06 ("adde a 'cooccurrin sites' signal to bias ourselves toward triggering when a local-oriented aggregator site (i.e. City, arch) shows u~ in the web results''); GOOG-Texas-1324737-39 (2009), at 38-39 ("final trigger . .. inclu es web-based sigtlals such as yelp et al"). Regarding Google Books, Google used Amazon as a trigger, see GOOG-Texas-01 Y,6298 (2009) (For books, we use Amazon as co-occurring site.''). Google appears to have considered a trigger for the finauc "OneBox" based on the presence of finance sites in organic results, but it is not clear it was launched. GOOGLR-00257663-75 (2008), at 68. u 7 See GOOGEC-0066150 (2009); GOOGLR-00162615-17 (2009), at 15. Googlc has provided some evidence that it has discontinued this practice with respect to Google Product Search in Dec. 20l0. 13 ~ Google did, at times, lower the position of certain Universal Search results. for example, in 2008, Google's search quality team recognized that Google Product Search results were often of poor quality. See CX-168 (GOOG-Texas-0214363) (2009); GOOGWRJG-000041022-23 (2009), at 22; GOOG-Texas-0197396 (2009); GOOG-Tcxas-0180522 (2008), at 22 ("With regard to middle/top threshold, raters say it goes at the top but clicks metrics suggest middle"). Around the same time. the Google advertising team expressed concern that the photo , pricing infom1ation, and other rich data provided by the Coogle Product Search diverted users' attention from ads, resulting in fewer clicks on ads. In the spring of2008, Google estjmated that the top placement of Google Product Search would lead to an "annualized loss of$ 154 million'' on product queries. GOOG-Texas0178597-607 (2008), at 598 ("Product Search Universal Holdback Experiment''). In response to both concerns, Google launched a series of"aggressive demotions" to move most Google Product Search result;s down a few positions on the SERP. See GOOG-Texas-0178597-607 (2008), at 59S ("Product Search Universal Holdback Experiment") ("We are executing an aggressive plan to further improve google.com user exper~·nee tor products that we estimate will reduce annualized loss from - $130mm to - $45MM within 4 wee · s"); GOOGTexas.-0214409-11 (2008), at 9 (Nick Fox writes that "the product search team said they were g ing to do a bunch of thi ngs to dramatically reduce the negative [revenue] impact of the product .... "); GO G-Texas017859?· 607 (~008), at 605 (estin~a. ting tha~ thes~ changes would rc~ult in the percentages of G~ogl~ Product Search m posmons 1, 4, and I 0 gomg from ·'85/0t 15" to ''40/35/25,' and a correspondmg reduc~on m loss of advertising revenue from $154 million to $70 million). Speci fically, in Jul. 2008, Google made ~hree algorithm ch~mges to "aggressively demote" more top OncBoxes to middle OneBoxes. GOOGMANB-00pos6049-S4 (20015), at 50. These were: (I) "Product Search Universal Triggering 2.0 [which] mainly moves them to a lower position", id., (2) "Using Exact Corpusboost to Trigger Product Onebox", which compares quer~es with que1ies on Google Shopping, triggers the Product OneBox if the same query is often searched in Google Shopping, and automatically places the universal in position 4, GOOGLR-00330279-80 (2008), at 79 (Launch Report for algorithm change); and (3) "Aggressive Demotion to Middle for Product Universal," which demotes from position one to position four if the product One Box does not meet a higher relevance threshold, the first web result is navigational with high probability, or rwo out of the top three resulis are for a manufacturer. Th.is change demoted about 51 percent of top product OneBoxes to the middle). GOOGMANB-000055473-76 (2008), at 73-74 (Launch Report for algorithm change). See CX-168 (GOOG-T cxas-0214363) (2009); GOOG-Texas-0197396 (2009). The "aggressiveness" of the demotion effort is debatable, as Google continued to display Google Product Search results in the fourth position. And even t11ese tni nor demotions were apparently quite controversial within Google. For example, Marissa Mayer "threatened to come to quality launch review to defend keeping product w1iversal at [position] 1." GOOGWRIG-000041022-~3 (2009), at 22. In any event, these demotion efforts were short-lived, as Google quickly moved Google Proruct Search 78 130 e m Docu SJ W a i v s nt Google does not allow comparison shopping sites to advertise in ads with graphics such as Produ,ct Listing Ads and Product Extension Ads which have higher clicks and conversions than text ads). 151 See Response to the Microsoft Economist Report on ''Anticompetitive Organic Search Manip\)lation" (Jul. 7, 20 I I) (stating the Panda update "was designed to e nsure a higher ranking for high-quality sites with original content and information and reduce the ranking of, inter alia, 'content fam1s,' i.e., low-quality adtoriented wcbsites. typically containing content copied from other websites.''); Economic Response to the omplaints by Found em and Ejustice.fr- RB B Economics (May 12, 20 I 0) (" ... Google applies a set of rules de igned to prevent sites that contai11 inappropriate content, mal ware or non-original content from showing u high in it'! search and ad results."). See also Google's Webmaster Guidelines, Little or No Original Conten http:l/support.google.com/webmasters/bin/answer.py?hl=cn&answcr-66361 (last visited Jul. 2, 0 12) 154 Although Google originally sought to demote all comparison shopping websites, after Google raters provided negative feedback to such a widespread demotion, Google implemented the current iter tion of its socalled "diversity" algorithm. See GOOG-Texas-0 179485-92 (2006), at 85 (identifYing shoppin comparison s ites for demotion); GOOCEC-0148152-56 (2007), at 53 (testing algorithm that would result in 'SERP declines between 8 and 20 percent" for shopping comparison sites); GOOGMANB-000007246-·p (2007), at 46 (launching the algorithm in Dec. 2007). Googlc claimed that the goal of this algorithm was to :increase the 1 diversity of Google ·s search results for product related queries." See Response of Google to DG 'Comp (Nov. 22, 2010), at§ 2.2, p.l. Initially, Google compiled a list of target comparison shopping s ites and demoted them from the top 10 web results, but users preferred comparison shopping sites to the merc hant sites that were often boosted by the demotion. GOOCSI~G-0000141 16-17 (2006), at 16- 17 (''We had moderate losses when we promoted an etailer page which listed a single product because the raters thought this was worse than a bizrate or nextag page which listed several similar products. Etailer pages which listed multiple products fared better but were still not considered better than the meta-shopping pages like bizrate or nextag..... "). Google then tried an algorithm that wo uld demote the CSEs, but not below sites of a certain relevance. GOOGEC-0168032-33 (2006), at 32. Again, the experiment £:1iled, because users liked the quality of the CSE sites. COOGSING-000014375-76 (2006). at 75 ("The bizratelnextaglepinions pages are decently good results. They are usually we~-forma[t)ted, rarely broken, load quickly ru1d usually on-topic. Raters tend to like them. I make this point becaase the replacement pages that we promote are occasionally off-topic or dead links. Another positive as~ct of the meta-shopping pages is that they usually give a variety of choices. ... The single etailer pages te d to be single product pages. For a more general query, raters like the variety of choices the meta-shopping site seems to give.") Google tried another experiment which kept a CSE within the top 5 results if it was alreatly there, but demoted others "aggressively." !d. at 76. This too resulted in slightly negative results. /d. Unable to get positive reviews from raters when Google demo ted comparison shopping sites, Google changed the raters' criteria to try to get positive results. Previously, raters judged new algori thm~ by looking at search results before and after the change ''side-by-side" (SxS), and rated which search result wa~ more relevant in ench position. COOGEC-0 168014-27 (2007), at 25. After the fi rst set of results, Google ask1d the users to · instead focus on the diversity and utility of the whole set of results, 111ther than resul t by •·esult, te ling users explicitly that ''if two results on the same side have very similar content then having those two re ults may not be mo re valuable than just having one." !d. at 23. When Googlc u·ied the new rating criteria with an algorithm which demoted CSEs such that sometimes no CSEs remained in the top I 0, the test again came back "solidly negative." ld. ut 19. Google again changed its algorithm to demote CSEs only if more than two appeared in the top 10 results, and then, only demoting those beyond the top two. With this change, Google finally got a slightly positive rating in its "diversity test" from its raters. /d. at 16; C OOGEC-0148 152-56 (2007), at 52 ("Launch Report: Shopping Comparison Demotion''). Google finally launched this algorithm change in Jul. 2007. GOOGEC-0014649 (2007) (launching at one Google data center); GOOCMAN B-<100007246-47 (2007), at 46 (launching to all remaining Google data center). 1 << Brent Rangen, Google Goes Boom on Low Quality Siles . .. So They Say, Search Engine Watch. Feb. 25, 20 II , http:l/searchcnginewatch.com/article/2049961/Googlc-Gocs-Boom-on-Low-Ouality-Sites.J.So-They-Sav; Am it Singhal & Matt Cutts, Finding More High Quality Sites in Search, Google Blogspot, Feb. , 20 II , h£tp://googleblog.blogspot.com/20 11/02/finding-more-hieh-gualitv-sites-in.html. 156 Google determined which websites would be demoted in two ways. First, Google had a grou of "spam mters" manually rate whether certain websites would be labeled as "content farms,'' and thus, su ·ect to demotion. GOOGHUFF-000089790-93 (2011). at 91. Google provided specific instructions fo its spam L32 e m Docu SJ W a i v s nt off by 31% and 25% respectively"); Shopping.com Data Submission (20 12) (showing drop in visits from 1.62 million to 1. 17 million for the weeks before and after the second Panda algorithm launched in the first week of Apr. 20 11 ); Dealtime Data Submission (2012) (showing drop in visits from 1.38 million to 0.508 million in the weeks before and after Panda initially launched in Feb. 20 I I). The drop in tr-affic to those websites also affects merchants, who prefer getting traffic from multiple sources. The monthly traffic from Pricegrabber and Shopping.com to Amazon dropped from the end of feb. 201 1 through the end of Oct. by, respectively, 35 percent and 30 percent. Amazon CID Response at 13. In addition, while traffic fi·om Feb 20 I 0 to 20 II incrcas(;d 99 percent, traffic from May 20 l 0 to 2011 decreased by 12 percent. Jd. at 14. Staff has collected evidence of several declines in traffic to other competing verticals due to changes to Google's SERP. See, e.g., FTC-NEXT-00000005 (20 12), at 70 (2007 search result page removal resulted in drop from ab9ut 900,000 to about 500,000 visits). 168 See, e.g., GOOGEC-1068069-72 (2009), at 70 (Comparison Shopping Dcmoti.on- "This pn?ject is likely to affect 1n1f~c flow to c?.mparison shopping.sites. ~he docum~nt lo~ated at [ci.ted document] giv9s a. detaile~ account of how tllJS affects the number of 101presswns of vanous s 1tcs. The Sites that lose the m~st 1mpress10ns are, as expected, comparison shopping sites. The sites gaining impressions arc retailers and eve1} some government and cdu sites."); GOOGEC-0148152-56 (2007), at 53 (Comparison Shopping De~otion - "The large comparison s hopping sites see SERP declines between 8 and 20%"); GOOGEC-0015560 66 (2007), at 60 (With respect to removing search result pages from the index, "In the end here the various G9ogle impressions the stores will be losing (not necessarily traffic to the stores, but correlated): ebay - 3.6M impressions, amazon - 2.3M, dealtime- 150K, epinions - 200K, ke lkoo - 620K, overstock- SOK, pricegrabber 70K, shopping.com - 500K"). 169 See, e.g., GOOG-T exas-1265906 (2010) (email noting that Google's local property now "dwarfs all other local sites in the world"); GOOGFOX-U00029790 (201 1) (discussing traffic increase since launch of Google Advisor vertical). 170 GOOG-Texas..()l99877-910 (2008), at 906. In its new iteration, Google Product Search took traffic from competing comparison shopping sites, despite some "pretty terribly embarr-assing failures" with regard to returning relevant product results. See GOOGWRIG-000041 022-23 (2009), at 22. See also GOOG-T exas0 192014-18 (201 0), at 16 (email noting that Google's product universal has increased shopping queries on Google} and, related, GOOG-Texas-0004101-04 (20 10) ("Product OncBox Traffic Impact Analysis"). 171 GOOG-Tcxas-U199877-910 (2008), at 907. 112 GOOG-Texas-0265014-16 (2010), at 14. 173 NexTag CID Response at 13. 114 /d. at 12. 175 Websites engaged in " scraping,'' according to Google's launch report fo r '"scraper demotion" are sites "that have authored less than 15% of their content .... " GOOGMANll-000037864-75 (2011), at 65. 1 176 Sec, e.g., GOOG-Tcxas-1 380771-73 (Jun. 2009), at 72 (email exchange discussing "scraping" review content from Yelp in lieu of reaching distribution agreement with Yelp); see also Yelp IR (Jul. ~2, 2011); Trip/\dvisor IR (.Jul. 6, 20 I I); Amazon IR (Nov. 18, 20 II ). 177 See, e.g, GOOG-Tcxas-1380771-73 (2009), at 71-72 (discussing imp01tance of Google Pla9es canying better review content from Yelp). Google has since ceased scraping content (as of Jul. 2011), illj a "voluntary" move a llegedly designed to transition its own local ve11ical property into focusing on "original content." See Google IR (Jul. 20, 20 I I). I 17 ~ Sc:e, e.g., FTC-YELPTX-00000163 and FTC-YELPTX-00000164 (2010) (email from Goolle to Yelp attaching standard Google license agreement). I?? See, e.g., Shopzilla lR (Feb. I, 2012) (stating that Shopzilla does not have the leverage to ne otiate the terms of the feed license; it is a take-it-<>r-leave-it agreement). 180 GOOG-Texas-0240698 (2009). 1 1 ~ GOOG-Texas-0182336-38 (2009), at 36-37 (discussing Googlc's usc of "scraping" Amazon's website to obtain Amazon Sales Rank of products, not available via Amazon's feed). 1112 See supra note 165: see also e.g., TripAdvisor JR (Mar 12, 2012) (web publishers "depend on search engines to gain vis ibility. Otherwise they just remain as tiny blips of information. Without the card catalogue, nothing is going to get found in the library. Because Google is dominanr in organic earch, the ecosystef depends on its services"). Websites believe that they need to make all o f their content available for Google o crawl because this will improve their traffic from Google. First. websites believe that the more origin· I content they I 134 SJ W a i v s nt e m Docu after we replicate their features)"); Hanke Tr. I 07:6-I 09:7 (citing CX-0055 and discussing risk tbat if Google launched its own site, pam1ers pulled their review content. and users didn't contribute reviews, then Google would risk having no review solution). 193 COOG-Texas-0996561-62 (2007), at 61; see also COOG-Texas-1074268-69 (2007), at 69 (email from Yelp CEO Stoppelman to Google's John Hanke upon learning about "the Google review feature in ~·1aps", "In the interest of giving us enough time to negotiate in good faith, I'd like to request that you remove our review and photo content from Google Maps before launching your feature next week. We're very uncomfortable with Coogle launching a directly competitive feature and we'd like to opt out while discussing what might be done to alleviate our concerns."). l'>4 COOGROSE-000082811-48 (2009), at 41 ("We have pa1t ially e nded up where we feared we would in 2007 1 ... 3~'~~ party content providers abandon Coogle .. . Limited success with our Reviews . .. Users b gin to start at review sites for key categories/regions ..."). 19 ' See Yelp IR (Mar. 5, 20 12). · 196 GOOG-Texas-0863053 (2009) (Eric Schmidt noting, when Yelp tumed clown Google's offep, "as you can see the deal is apparently off ... [instead we need to] continue to build a g reat reviews product h~re at Google." To this John Hanke responded "we'll come to the oc in jan w/ a plan. my sense is that we shou19 be prepared to invest some real money ($1OOM?) building this up. It will require us spending on things (comm~ nity managers as well as technologists, city-by-city community building, city-by-city marketing) that have beem hard for us to wrap our am1s around and commit to in the past. ... " Eric Schmidt responded, ''Thanks. I completely agree with your approach here and will definitely fund it !! thanks"). 197 John Hanke, introducing Google Places, Google Blogspot, Apr. 20, 20 I 0, htto:/ /googlcblog.blogswtcom/20 10/04/introduc ing-google-places.htm I# !/20 I 0/04/i ntroducing-googleRiaccs.html. 98 See Joh n Hanke, introducing Google Places, Google Blogspot, Apr. 20, 20 I 0, htt p://goog lcblog.blogsoot. com/20 I 0/04!introduc ing-google-places.htm l#!/20 I 0/04/introducing -googleRiaccs.html. i)i) GOOG-Tcxas-1363574 (Jul. 26, 20 I0) (" ... J noticed you're still using excerpts of our review content in locnl without license and counting them as Googte ' reviews' , yet you've demoted Yelp to the bottom regardless of freshness (happy to discuss, but we're not ok with this use of our content)"). 200 TripAdvisor IR (Mar. 12, 20 12). 201 ld. (explaining that a lthough TripAdvisor received some traffic from Google's Places property, once Google became competi tive with T ripAdvisor, T ripAdvisor had a reason to terminate the license, and the loss of traffic was very small). 202 203 ld. See CX.-67 (Google Blog, "Place Search: a taster, easier way to find loca l information") (2010) ("Today we' re i~~troducing Place Search, a new kind oflocal search result that organizes the world's. in_fofmation around plnces. ); GOOG-Texas-I 012889-92 (201 0), at 89 ("[Manssa Mayer's] current proposal dtstmgmshes between Search and 'Content' (Non-Search] pages, and accurately deems our 'current ' Place and Produc\ Pages to be 'Content' [Non-Search] pages, and concludes: partne rs should be allowed to choose whether the~ want to be included in such pages. I believe we all agree with Marissa on these (and all other) ideas .. ."). Websites pem1it or block web crawlers from crawling their sites by including a robots.txt fi le on their web site See, e.g., w\.vw.yelp.contlrobots.txt; www.amazon.contlrohots.txt; www.google.contlrobots.txt. These files provide very crude capabilities, telling crawlers whether they can crawl data or not, not how the sites may use that crawled data. Web:;ites that are not crawled are not included in Googk web index and do not show up in organic search results. Coogle's Webmaster Tools, Block or Remove Pages Using a Robots. txt File, http:l/support.googlt:.com/webma-;tersrbinlanswer.pv?hl =en&answer- 156449&topic= 1724262&ctx=topic (last visited Jul. 2, 20 12). 2<14 GOOG-Texas-1041511-1 2 (2010), at 12 ("remove blacklist of yelp [reviews] from Web-extracted Reviews once provider ba!.ed Ul live"); GOOG-Texas-1417391-403 (20 10), at 394 ("stating that Googi; should wait to gublish a blog post on the new U1 until the change to " unblacklist Yelp" is "live"). 05 GOOC-Texas-0222679 (2010) ("The competition in this space comes from two weaknesses: I. We do not have much user-user or user-business communication on the Google platform. This is both a culf'ral and technological issue. 2. We do not have a complete solution wrt local businesses. We nm the risk that competitors like facebook, twitter and yelp become the site where local businesses are discovere and interacted 136 e m Docu SJ W a i v s nt 216 GOOGMAYE-000062536-537 (20 11), at 36 (Marissa Mayer wrote to Jeremy Stoppelman, "we do not have the ability to immediately customize which search features a website is included in."). 217 See, e.g.. Hanke Tr. 143:20-144:8 (citing CX-61, GOOG-Texas-0864517-518 (2009), on providing perdomain blacklisting for Google local); Goodrow Tr. 116:12-119: II (discussing a few methods of preventi ng product content from appearing in Google Product ~earch). Moreover, Googlc has also proposed to adhere 10 commit to precisely such an "opt-out" feature in its proposal to the EC. See Googlc-EC Settlement Proposal at 15-16. 213 Mayer Tr. 223: 1·1-224:7. m Avni Shah, The Ongoing Evolution ofPlace Pages, Google Lat-Long Blogspot, Jul, 21, 20 I I, htttJ://googlelatlong.blogspot.com/2011/07/ongoine-evolution-of-place-pages.html. ("Based on careful thought about the future direction of Places pages, and feedback we've heard over the past few months, review sui pets from other web sources have now been removed from Place pages. Rnting and review counts reflect nly those that've been written by fellow Google users, and as a part of our continued commitment to help' g you find what y0\1want on the web, we're continuing to provide lin ks to other review sites so you can ge a comprehensive view of locations across the globe.") 220 See, e.g., Goodrow Tr. 35: 18-22; &0: 11-22; & l: 11-23; I 09:18-1 10:7. 221 See. e.g.. Goodrow Tr. 77:2-16; 114:2-12; 164:18- J 65:9; 185: 14- 186: I I. 222 Amazon CID Response at 28. 223 /d. m GOOG BRA0-000049034-35 (2010) (including email from Amazon executive Steven Shure regarding ··Google's use of Amazon 's customer product reviews and ratings"). 2 ~ GOOG-Texas-1039100-101 (2010), at 100 ("As I said on our call, we wo uld like Google to no longer display or incorporate the Amazon product reviews information, including text and stars/ratings, which it ingests [lhrough] its natural search crawl, within Google Product Search.... We ask thai you remove the review excerpts fro m the display and the star ratings from your overall product rating calculation. Their current use is without Amazon's pem1ission .. .. We would like you to get back to us in a week, by September 3ro, with a date by which Google will be able to remove Amazon review information from product search. . .. 0'1 the surface it would seem that we are simply asking you to make a change which directly parallels the recent ~hanges Google has made in displaying Yelp reviews in Coogle Places .... Amazon's product review content represents a similar proprietary asset and we do nm want it to appear in Coogle Product Search."). I 226 GOOGROSE-000078506-08 (20 10), at 6 (''We arc preparing to remove Amazon's product Jeviews since they gave us until Friday of next week.''); GOOG-Texas-1012889-92 (20 I 0), at 90 ("Amazon - let's tell Amazon lhat we were planning to change (the user interfacel anyway, but s ince we a re a few w~eks away from making revisions and because of (technical unccrtaintyj we will in the meantime take their cont~nt out of Product pages by [date] ... stress that we're doing this out of respect fo r the relationshi p, but thlour decision ~doesn't represent a change in policy]"). 27 Amazon CID Response at 15 (explaining the value of Amazon's "massive amounts of custOJ er ratings for the millions of products in its catalof is that "they accur~tely aggr.egate customers' ~·eviews a~ t any given product and enable consumers to qwckly assess the percetved quality of a producl w1thout havt g to read often lengthy text reviews''). 228 !d. at 29. 229 /d. at 28-29. 2'11/d. 231 Goodrow Tr. 4 7:2-49: 13. 232 /d. at 67:6-68: I. 233 /d. at 74:5-79:20. 2 \.1 See. e.g., Goodrow Tr. 44:5-46: ll (describing benefit of having a product catalog and that with Froogle, Google tried one method of developing a product catalog, but were not successful in the method that they chose, namely, clustering); GOOGEC-0 134533-631, at 617 ("We' vc demonstrated that unsupervised clustering doesn' t work."). m Amazon CID Response at 32-34 (detailing the considerable resources Amazon has expended in developing its comprehensive and user-friendly product catalog of over a billion unique items for sale: effOiitS include obtaining and developing content from merchants, vendors, and Amazon employees; entering i~ko various types of business relationships to obtain catalog information from mcrchanls and vendors; developinglthe appropriate 138 SJ W a i v s nt e m Docu 262 Keystone, Advertiser Multi-homing in Online Search Advertising in Europe (June 26, 2011), at 9 (data discussed in the text is U.S. data). w The e figure represent th e upper bound estimates of the percentage of multi-homing advertisers in each docile. The likely actual percentage will be lower. 2 1>4 This is important because the availability of advertisements from smaller advertisers fills out a search engine's coverage of queries, particularly for "tail" queries. See inji-a p. 98. 26 s According to Microsoft, approximately 49 percent of keywords with l 00 impressions or fewer per month are bid for only on AdWords; for "high scale" keywords, approximately 78 percent are bid for on both AdWords and AdCenter. Susan Athey, Presentation, ''The Role of Scale in Competing in Online Search" ! arch 26, 2012), at 9. 266 See Microsoft: IR (Jun. 11 , 2012). This claim may not stand up to scrutiny, however. Despit numerous requests, Microsoft has not produced data to suppott th is assertion. In ad.dition, it is unclear on hat basis M icrosoft is able to estimate the level of optimization advertisers perform on the ir Ad Words canlpaigns. 267 See Brew Gadgets IR (Jan. 30, 20 12); National Relief IR (Feb. 15, 20 12); Phoenix East Aviat110n IR (Feb. 29, 2012); Speedy Soft IR (Feb. 6, 2012); Top Ilat lmagewear IR (Feb. 22, 2012); Yarn Market IR (~an. 13, 2012). While it is true that some of the small advertisers interviewed were not interested in a cross-platform optimization tool, their limited interest can be explained by unverified assumptions about a eros -platform tool's u ltimate fi.tnctionality and varying opinions on cross-plattonn management's current trans~ction costs. Sec Ekinoks and Lab Test Florida IR (Feb. 10, 2012): Portadam IR (Feb. 13, 2012); Wyzant fR CJan. 20, 2012). lbS See Green Paper Products lR (Feb. 9 & 10, 20 12); Puppet U IR (Jan. 31 , 20 12); Top Jlat lrnagewear lR (Feb. 22, 20 12). 2 9 " See Brew Gadgets IR (Feb. 2, 2012); Top Hat lmagewear IR (Feb. 22, 2012). 210 See Phoenix Aviation IR (Feb. 25, 2012). 211 CX-36 (GOOGWOJC -000044501-05) (2007), at 3: see also GOOGAROR-000007146 {Sep. 25, 2007), at slide 13 (emphasis added). 212 CX-41 (GOOGFOX-00128077-81) (2009), at 79. m Holden Tr. 50:3-21. m Sec id. at II 0, 122-123, 185-186. m CX-40 (GOOG-JTA-25~0064254-55) (2008), at 54 (emphasis added). 276 CX-39 (COOCWOJC -000009350-53) (2009). at 51. 277 CX-47 (GOOGEC-0181955-59) (2009), at 56. Making explicit the connection between the iscussion of relnxing the restrictive conditions and contemplated new functionality for DART Search that would otherwise violate those conditions, the engineer responsible for DART Search rep!ied "[w)e aren't ready t9 build a comingli ng product now.'' CX-0046 (GOOGWOJC-000058344-47) (2009), at 44. 278 CX-42 (GOOGEC-0180380-85) (2009), a.t 84. 279 CX-43 (G OOGEC-0180407-11) (2009) at 7. 28 CX-45 (COOGEC-0180400-06) (undated), at 5. I lolden was not certa in of his response to,, e original question posed by the API product manager. Holden Tr. 166:12-13. But, he did believe th at C -0045 was the document presented to Larry Page. Holden Tr. 174:6-20. . 2 1 R CX-44 (GOOCWOJC-000059695-97) (201 0), at 95 ("As we expected, Larry was OK with tihe status quo ns outlined in the presentation"). m See supra note 3. m Holden Tr. 175:24-25. 2 4 s CX-182 (GOOG-JTA-09-0057720) (20 I 0) (I !olden writing to Wojcicki, "We didn't take notes for obvious reasons (hence why I' m not elaborating too much here in email) but happy to brief you more verbally"). This document is an unredacted version of CX-44 (GOOGWOJC-000059695-97) (Jan. 21, 20 I 0). During the hearing. counsel for Google indicated that the redaction was improper. Holden Tr. 197: I 2-24. 285 For a detailed overview of Google 's AdSense partners, see Appendi x I (Table listing exclusive agreements) and Appendix 2 (Table listing preferred placement agreements). 2 1 M' Google Data Submission (Jul. 31, 20 12). m Braddi Tr. 22:1 1-15. 2 ~R ln the early 2000s, Google identified these partners as important sources of user traffic becau~e the search bar on the lSP/ portal page was the first thing the user often saw when tuming on the computer. $ee COOGPAGE-000009322 (2004), at 3-24 (discussing Google's ISP access strategy in 2004). I I ° 140 SJ W a i v s nt e m Docu 310 Microsoft IR (Jun. I I , 20 12). Wai-Man IR ( May 30, 2012). 312 Best Buy IR (Jun. 14, 20 12). 'n See Kayak IR (Jun. 20, 2012) (characterizing the abilit y to serve some Bing or Yahoo advertisements alongside Google search ads as "wonbless" because B ing monetizes so poorly in relation to Google). 314 LAC IR (Dec. R, 20 12) (Microso ft sought an exclusive deal); Ama.~;on IR (Feb. 15, 20 12) (Microsoft and Yahoo! both require page-based exclusivity so their ads cannot be mixed and matched with the advertisements of their competitors.) 315 W1ll-Mart IR (May 30, 2012); Best Buy !R (Jun. l4, 20 12). 316 Amazon lR (feb. 15, 2012). Microsoft and Google apparen tly do have the ability to provide publishers with technical assistance to avoid duplication, but none of the publishers that identified this concern rppo1ted receivi ng such assistance. See CX-113 (FTC-0000093-228) (2008), at I J0 (Google/Yahoo! proposed agreement at §2.12, explaining that Google would use "commercially reasonable efforts" to exc~de AFS Ads that contain URLs from corresponding results provided by Yahoo!); Microsoft !R (Jul. 20, 2012). 317 See, e.g. COOCKAP0-000006280-95 (2010), at 83 (discussing revenue improvements frorh lowering revenue share and standard. iz. ing AdSense agreements with publishers.)~ CX-102 (COOGBRTNl 000025680-83) (2008), at 80 ("Our general philosophy with renewals has been to reduce TAC across the board"~; COOGBRA0-0000 12890-944 (2007), at 13 (AFS strategy discussed in the 2008 AdSense Business Rev iew, "we are instituting stricter AFS Direct revenue-share tiering guidelines by region . .. Our overall goal is to achieve better AFS economics for both new and renewing pan:ners."): CX- 106 (COOCKAP0-000006280-95) (20 I 0), at 83 ("2009 Traffic Acquisition Cost (TAC) was down 3 percentage points from 2008 attributable to the application of s tandardized revenue share guidelines for renewals and new partnerships ... "). ~ 18 See, e.g. Business.com !R (Jun. 15, 2012); Time Warner Cable IR (Sep. 8, 2011 ). 319 CX-104 (GOOCBRAD-000048209) (May 3, 2010), at4 (Ql 10 Googlc TAC Summary). J~ See, e.g., Business.com IR (Jun. 15, 2012); GOOG-A FS-000004666-68 (2007). at68; COOG-AFS000000316-27 (No\'. 4, 20 I 0) at 27 (2007 GSA had a 3-tiered revenue share of 80, 85, and 87.5 percent; the 201 0 re newal had corresponding tiers of73, 75, and 77 percent). 1 -'~ See, e.g. Time Warner Cable IR (Sep. 8, 201 1) (search adve rtising typically generates revenue well above display advertising). 322 See, e.g., Comcast IR (Nov. 15, 2011 ); AOL IR (Dec. I , 20 I I ); lAC IR (Dec. 8, 20 II). J2.l Best Buy lR (Jun. 14, 20 12) (contract is not exclusive); Kayak JR (Jun. 20, 20 12) (contract is not exclusive); Amazon IR (Feb. 15, 2012} (contract is not exclusive- Amazo n res isred Google's attempt to impose exclusivity); Wai-Mart lR (May 30, 2012) (describing the contract as not exclusive b11t noting t !at Google requires preferred placement ifWal-Mart uses Yahoo! or M icrosoti). ·12~ See Google Data Submission (Jul. 31 , 20 12). m cBay IR (Oct. 27,20 1 J). 111 J26 !d. m Jd. 328 /d . .I2Q !d. JJO Jd. ~31 NexTag I R {May I0, 20 II). m Td. 333 /d. JH /d. m Jd. 136 !d. m Busincss.com IR (Jun. I 5. 20 12). mtfd. Jd. 340 Id. '-~1 Td. 3 2 ~ Amazo n IR (Feb. 15,201 2). 141 /d. 339 142 SJ W a i v s nt e m Docu 37 g Brin Tr. 3 19:13-320:15. Brin testified that Google has done some analysis of Facebook and {\mazon at some point in the past (although not regularly), and does not recall getting regular reports on an~ other vertical competitor. According to Brin, "it's definitely [a] much harder comparison to make." Brin Tr. :. BI:IS-22. 379 See Department of Justice, Recommendation to Challenge Google/Yahoo Services Agreeme~t, 39 (Sep. 22, 2008) ('"[u]sers do not substitute foreign search engines tor U.S. engines, becau:;e foreign engines are not designed to deliver relevant information for a U.S. user''). See also Microsotl EC Submission 8~ (noting that the relevant markets at issue in this investigation should be defined by "national or linguistic bo~ndaries"). 3 soSee Press Release, comScore, comScore Releases May 2012 U.S. Search Engine Rankings (Jt~n. 22, 2012) ht ://www.comscorc.com!Press Events/Press Releases/2012/6/cornScore Releases Mav 201 U.S. Search Engine Ran kings.Google's market share has been measured as a share of the total voh1me ohm que searches in the United States conducted across traditional search engines, as well as other "leaui ng" sites su has Facebook and Wikipedia. Coogle's internal figures reflect slig htly higher market shares for Google, see, g., CX-183 (GOOGWRIG-000086779-81 ) (2011) (reporting monthly market shares in Coogle's internal n'etrics ranging between 69.4 and 83.5 percent, while the equivalent comScore nu mber for the :;arne period is 6$. 1 percent). According to Sergey Brin, Google relies on both internal and external data sources when ex.ami~ing its market shares, although all of the souJces "have their problems, their challenges." Brin Tr. 3 15:9-316:2. Brin testified that he is more concemed with whether all of the data sources are consistent in their "trends," 1., whether Google's share is going up or down, than the exact numbers. Brin Tr. 321: 11-23. 38 1 See Brad Stone & Brett Pulley, lAC's Bany Diller Surrenders to Google, Ends Ask. com 's Se rch Effort, Bloomberg, Nov. 9, 20 I 0, htlp://~·v.:w.bloomberg.com/news/20 l 0-11-09/iac-s-diller-surrenders-to-googlejuggernaut-ends-ask-com-search-effort.html. m Google has an intermediation agreement with AOL, whereby Google provides AOL with Google search and search advertising functionality. Microsoft EC Submission at 23. m See Press Release, cornScore, comScore Releases .May 2012 U.S. Search Engine Rankings (JUJ1. 22, 20 12) hllp://www.comscore.com/Press Events/Press Releases/20 12/6/comScore Releases Mav 2012 U.S. Search En~ine Rankings . .>s.a This number should be viewed with some caution, both because there can often be shifts of a percent or two in the monthly comScore data, and also because there is really no good WHY to measure search share with high precision. All oft he measures of search share have various methodological problems and limitations. See. e.g. Brin Tr. 315:9-3 16:2 (noting that all of the internal and external market share numbers havetissues); Schmidt Tr. 53: l 0-55:2 (noting that Google's view is that comScore numbers are always wro ng ; but see GOOGMANB-000095004-07 (2011), at4 (Hal Varian, Google's chief economist, writes: "Tho 1gh I would agree that ComScore is unreliable, it's not at all obvious to me that this matte rs much to us. Fro 1 an antitrust £crspcctive, I'm happy to see them underestimate our share.") liS See Press Release, comScore, comScore Releases May 2012 U.S. Search Engine Rankings (Jlm. 22, 2012) htt ://www.comscorc.com/Press Events/Press Releases/2 12/ I omSc re Releases Ma 20 1 U.S. Search Engine Rankings. JN(I Microsoft IR (Jun. II, 2012) . i.l I !d. See Bepco, Inc. v. Allied-Signal, Jm:., 106 F. Supp. 814, 830 (M.D.N.C. 2000) (70-75 percent~. See also, e.g.• Exxon C01p. v. Berwick Bay Real Esrare Partners, 748 F.2d 937,940 (5"' Cir. 1984) (per curiam) ("monopoliz..'ltion is rarely found when the defendant's share of the relevant market is below 70%"); Colo. interstate Ga.1· Co. v. Natural Gas Pipeline Co. ofAm., 885 F.2d 683, 694 n.l8 ( l 0'11 Cir. 1989) (in order to .\87 JHS cstnblish monopoly power, "lower courts generally req uire a minimum market share of between 70% and 80%") (internal citation omitted). 389 See, e.g., Oahu Gas Service, Inc. v. Pacific Resources. inc., 838 F.2d 360,366 (9th Cir. 1988) (affirmingjury finding that defendant had monopoly power despite steadily decl ining market share frum 100 percent to 68.2 percent at time oflaws uit). While Judge Learned Hand was '·doubtful whether sixty or sixty-foqr percent would be enough;· see United States v. Aluminum Co. ofAm., 148 F.2d 416.424 (2d Cir. 1945), and the Third Circui t has suggested that ''a share significantly larger than 55% has been required to establish prima facie market PQWer." United Stutes v. Dentsp~y lm'l, Inc., 399 F.3d 181 , 187 (3d Cir. 2005), no minimum threshold has ever been established. See Broadway Delivery C01p. v. United Parcel Serv. ofAm., 651 F.2d 122, 130 (2d Cir. 198 1). cert. denied, 454 U.S. 968 ( 1982) (holding that, while, "[s]ometimes, but not inevitably, it will be useful to suggest that a market share below 50% is rarely evidence of monopoly power, a share betwee 50% and 70% 144 e m Docu SJ W a i v s nt ~See, e.g., Amazon CID Response at 38; Clickablc IR (Oct. 24, 2011); Living Social IR (Mar) 3, 20 11). See also, e.g., Brin Tr. 178:5-21 (testifying that search ads convert much better than other types of advertising); Schmidt Tr. 125:21-126:9 (same); GOOG-ITA-03-004551 1-18 (2009), at 13 ("Content conve1ions do not lead to sales like search conversions," attributing the difference to where display reaches users in the buying cycle versus where search reaches users in the buying cycle); GOOG-ITA-13-00 00937-41 (2009), at 37 (Hal Varian stating, "don't lump search advertising in with everything else- treat it as a separate category," further noting that the recession has cut far more significantly into d isplay conversions than into search ad cot~ ersions). 40 ' Gian Fuglioni, Who Will Rid us of this Meddlesome Click?, comScore, Dec. 7, 2010, httr:/Jhlog.comscore.com/2010/ 12/ri d meddlesome click.html ("TI1c average click rate (define~ as the percent of paid ads that were clicked on) for paid search campaigns (3.5%) is massively (35x) higher than for display ad campaigns"). 4 0Q See wpra p. 9- J I & notes 43-48. 407 FTC-EBA Y -00000002 (20 12), at 3 I; A mazon CID Response at 38. 408 Booyah lR (Jan. 25, 20 12); Comcast IR (Nov. 15,201 2); iCrossing IR (Apr. 9, 2012) (seardj advertising is a lone at. the bottom of the marketing funnel, keyed to user intent); Core-Metrics (IHM) lR (Novl 4, 20 1 I); Comcast IR (Nov. 15, 201 1); Priceline TR (Oct. 18, 20 12) (search has surgical precision and is Jnlike other advertising); Amazon CTD Response at 38. 409 Brin Tr. 178: 16-21 (search ads convert much better than other types of ads); G O OG-TTA-0 -0045511- l 8 (2009), at 16 ("content conversions do not lead to sales like search conversions"); GOOG-ITA-01-0364176205 (20 I 0), at 95 (a picture depicting a hierarchy of conversion attribution placing paid search at the top followed by organic search, display, affiliates, social networks, email marketing, direct visitation, and offline); Group M IR (Oct. ll , 201 I) (contextual advettising is better than display, but not as effective as search at ~cncrating conversions). 10 Brin Tr. 181:2-8 ("[Y]our average content page view is worth significantly less than your average search page, no question about it."); Schmidt Tr. 129:6-130:5 (testifying that for advertisers that want to generdtc sales, their money should go to search advertising first and then other forms of online advertising and then offline advertising "[s]o the general feeling- and again this is con finned by experience - is that you would always put text ads lirst and lhen display second which is still online.''). 41 1 Amazon CJD Response at 38 and Table 9 .2. See also e.g., Living Social C ID Response at I~ (no substitute for search advert is ing); Group M IR (Oct. 11 2011 ); cBay IR (Nov. 4, 20 II); Didit.com fR (De~. 27, 20 12); lAC IR (Dec. 8, 2011); AOL IR (Dec. 9, 2011); Demand Media IR (Dec. 9, 2011); Kayak IR (Dec. 20, 2011). m See, e.g., Demand Media IR (Dec. 9, 2011) (price increase will not cause shift to other forms of advertising); EAS IR (Feb. 24, 20 12) (same); Kayak IR (Dec. 20, 20 11) (price increase would not cause Kay~k to spend less on search advertising); Boo yah IR (Jan. 25, 20 12) (if prices went up I 0 percent on Google paid search, the advertiser would not like it, but would pay it); Clickable lR (Oct. 24, 2011) (cannot divett adveiising dollars from Google to other p latfonns); Wyzant IR (Jan. 20, 20 12) (would pay an increase of 10 perce t rather than shift spend away from Googlc Adwords); Comcast IR (Nov. 15, 20 I I) (it would take a price in rease of more than 50 percent to move any money from search advertising). 4 1 t. Living Social CID Response at 16; Did it Draft DecI. (2008) (would aftim1 to the principle thf t there are "no good substitutes for paid search'' in 201 2); Amazon CID Response at 38-39; FTC -EXPE-0000 002 (2012), at 15-16. 41 ~ Livi ng Social CID Response at 16. 4 15 Apollo IR (Jan. 4, 2012); fox Studios IR (Jan. 20, 2012); Havas JR {Oct. 5, 2012); Sound World Instruments IR (Jun. 24, 20 12). Generally speaking, it was difficult for many advertisers to answer the hypothetical- "what would you do in the face of an across-the-board price increase?"- because of the unique manner in which search advertising is priced. Pricing of search advertising is based on what is known as a " Vickrey second auction" model. The idea behind this auction is !0 give advertisers the incentive to bid their maximum bid, rather than tly to game the auction to pay as little as possible. In this type of auction, an advertiser is only requi red to pay $.0 I more than the next lowest bidder. For example, three sports retailers are bidding on the keyword "sneakers." Reta iler A bids a maximum of$1.00; Retailer B bids $0.50; and Retailer C bids $0.25. All other th ings being equal (i.e., controlling for Google's quality score adjustments), Retailer A. will "win" the top spot in the auction, but will only pay $0.51 to Google if a user clicks on Retailer A's ad. ~~~his way, the auction itself drives up the prices, and Google's "control" of prices is more indirect (although lfogle sets minimum bids and establishes quality scores that sets each advertiser's baseline bid). Notably, each time a I 146 e m Docu SJ W a i v s nt 8.1% This Year. Emarketer.com, Mar. 23,201 1, httl://www.emarketer.com/blo • index. h I ~ick-stat­ vahoos-search-ad-revenue-share-fall-81-vear/ (Emarketer.com estimates Google's share to be 7 percent in 2010 and 80 percent in 20 11 ); See ADV Media Productions, Google Dominates Search Adverti~ing With 800/o Market Share Unaffected by The Rise of Bing, http://www.advmediaproductions.com/blog/google-dominatespaid-seMch-advertising-with-80-market-share-unaffected-hy-the-rise-of-bing/ (last visited Jul. 16, 2012). See also GOOGMA YE-000035824 (2009), at 8 (in 2009 Google estimated its market share 71.3 p11rcent). We understand that BE Sta!T may be measuring Google 's share of the search advertising market basJe d on ad clicks or impressions. We are unclear as to why BE would rely on this metric because a click on an ad does not actually tell you anything about how much an advertiser is spending on any g iven ad on any giv~n platform. The logical metric for estimating adve1tising share is advertiser spend (or advertising revenues),lwhich is the metric relied upon by all of the industry sources (see above) - and Google itself. See, e.g., CX- 16 (GOOGTexas-14R915-70) (2009), at 19-20 (evaluating "market share by size of ad revenue captured"). 42 R Google Data Submission (Jan. 10, 2012) ( listing I ,280,983,000 advertisers in 2011 ). 429 Michael Liedtke, Microsoft Takes $6.2 Billion Hit On aQuanrive Online Ad Woes, Huffington Post, Jul. 2, 20 12, http://www.huf fingtonpost.com/20 12/07/03/microsoft-aquantive-online-ads n l645696.tltml (attributing the growth to loss of share to Yahoo with Microsoft holding steady at 7 percent); Covario.com, Covario Finds High Tech Global Paid Search Spend Rose 22 Percent in Ql over the Same Period Last Year, I httn://www.covario.com/ news-and-views!newsroonv press-releClses!SIS-covario-finds-high-techcglobal-paidsearch-spend-rose-22-percent-in-q 1-over-the-same-period-la.<;t-ycar-) (last visited Jul. 16, 2012) (estimating Microsoft and Yahoo!'s share of search advertising market to be a combined 13 percent). The remaining 4-12 percent of the search advertising market appears to be controlled by AOL and Ask, both powere See, e.g., Cab levis ion IR (Jun. 20, 20 12); Business.com IR (Jun. 15, 20 12). 437 Department of Justice, Recommendation to Challenge Google/Yahoo Services Agreement, 5 -55 (Sep. 22, 2008). The Department of Justice defined "search syndication" to inc lude both syndicated sear!h and search advertising, wherein intermediaries such as Google stmck agreements with website publishers t provide both functionalities. Jd. 41 ' R See supra p. 67 (relevant geographic market for horizontal search is li mited to the United St!\tes) and p. 73 (sume fo r search advertising). See Department of Justice, Recornmcndat'ion to Challenge Googl'e!Yahoo Services Agreement, 39 (Sep. 22, 2008) ("[u]sers do not substitute foreign search engines tor U Is. engines, bcc $100 million by 20 12). See also e.g., Brin Tr. 58:7-19 " ... if we're serving our users poorly in whatever subsets of queries, we would definitely face significant re.Jenue erosion as we got less usage.''; Schmidt Tr. 160:25- 161: I 0, 226: I 0-228:25, 229:23-230:25, 234:13-234:22 235:2-235:8, 236:20-237:5, 294:1-295:1 8 (".. . it's opportunity lost... And in our industry, it's important to &o very well ... There was a concern that the aggregators were doing a good job in an area where we were not a - doing a good enough job ... We want to compete. So that drove a - a discussion."). 458 Se11 Microsofi Corp., Microsoft Complaint to the European Commission (Mar. 3 I, 20 II). 'f1'is theory di rectly tracks the Department of Justice's theory on the role of middleware in Mieroso.fi. Therf.' it was argued, middleware represented a threat to Microsoft's operating system dominance not because the mi?dleware would itselfreplace the underlying operating system, but because middleware provided an alternati vellatform onto which applications could be written, which could be run irrespective of the underlying operatin system. Lowering this so-called applications barrier to entry, in turn, lowered the costs for other firms t introduce rival operating systems that could directly challe11ge Microsoft's dom inance over lntel-compatible o~erating systems. Sim ilarly, here, Microsoft argues that a "key component" of its strategy in attracting users ha~en to partner with vertical websites so that Bing can offer a "differentiated generctl search experience to com ete with improving Google.'' !d. See also e.g., GOOG-Texas-1325832-33 (201 0), at 33 ("Oing has explicitly m verticals a key pa•t of their strategy to beat Coogle"); GOOG-JTA-0 1-033 1214 (2009) (email oting that Bing is focused on competing against Googlc in its "two top vert icals," shopping and travel). 459 Wh ile reduced innovation is at the heart of this theory, the role of pricing cannot be ignored, in that (as with other theories described later in this memorandum), the broader availabi lity of alternative search advertis ing r:oatforms would operate as a constraint on Google's ability to raise prices to its advertisers. See supra p. 30. . 461 See supra p. 30-3 1. 462 In Micro ·oft. the government's argument that product improvement could be outweighed by anticompetitive effects did not fare well. The en hanc court considered a claim that Microsoft had designed certain software in a way that made Java a pplications both faster on its operating system and incompatible with rival operating systems. Although the opinion stated that the appl icable test was th at "lhe incompatible produc~ must have an anticompetitive effect that outweighs any procompetitive justifi cation for the design," it held thft the fact that product ran taster on Microsoft machines sufficed to make it legal standing alone and did not a{/pear to try to balance that benefit aga inst anticompetiti vc effects. Microsoft, 253 P.3d at 74-75. Similarly, w ' le the D.C. Circuit up held the lower court's ruling against Microsoft on the company's e fforts to integrate t e In ternet browser with the operating system, it did so on particular integration aspects for which Microso t could provide no justification. Where M icrosoft did provide a justification (namely, in oveniding users' choi e of a default browser), the comt found no liability. 253 F.3d at 67-68. 463 See Response of Go ogle to DG Comp (Jul. I, 20 II ), at 2. 4 ~ See Shashi Seth, Beyond the Search Box, Yahoo Search B log, Jun. I 0, 20 I 0, http://wwv,.,ysearchblog.c.om/201 0/06/1 0/bevond-the-search-box/. ("People no longer search t find a list of blue links; they search to find answers in the shortest amount of time possible. We believe that surfacing the right information at the right time is more important than the number of total results delivered or number of traditional queries conducted"); Greg R. Notess, Microsoft's New Bing - The 'Decision Engine, Information Today lm:., Jun. 8, 2009, http://newsbreaks.infotodav.com/NcwsBrcaks/Microsof\s-Ncw-BingThc-DecisionEngine-54514.asp. (noting that :vticrosoft rebranded its MS search engine as Bing in 2009, du~bed it the ''decision engine," and began incorporating universal blends similar to those used by Google and Yahoo!). 465 Google Search Innovation White Paper at 56- 58. 466 /d. at 40. 461 The OneBox, predecessor to the Universal Search ''blend,'' showcased Google's vertical content in a box at the top of the Google search results page. See id. at 34-45. 4 6lC Page Rank "relies on the uniquely democrdtic nature of the web by using its vast link structure as an indicator of an individual page's value. In essence, Google interprets a link from page A to page B as a by page A, receives; for for page B. But, Google looks at considerably more than the sheer vo lume of votes, or links a example, it also analyzes the page that casts the vote. Votes cast by pages that are themselves rtant" 150 e m Docu SJ W a i v s nt design changes by monopolists that substantially disadvantage rivals or nascent threats, even w ere that conduct does not rise to a Section 2 violation. Professor Herbert Hovcnkamp believes that the area ofmpnopoly leveraging in industries characterized by network effects may be a type of exclusionary conducd uniquely suited to standalone Section 5 competition enforcement. Herbert J. IIovenkamp, The Federal Trade O{Jmmission Act and the Sherman Act, 62 Fla. L. Rev. 871, 885-87 (20 I 0). Hovckamp lauded the FTC's decisiob to challenge [ntcl's conduct with respect to graphic chips in the intel matter because he fe lt that Section 5 wjs uniquely suited to deal w ith thorny issues relating to design changes by monopolists that disadvantage ri als, id., and because liability under Section 5 does not lead to the imposition of treble damages, and is appli d by an agency that is able to develop expertise about partt' cularly complex issues such as design changes that nfbgatively impact rivals. See also Areeda & Hovenkamp, Antilntsf Law, ,] 772h ("Another possibility is use of§ of the Federal Trade Commission Act, whose prohibition of unfair methods competition can reach instances o leveraging act ivity relating monopolized and nonmonopolized markets in circumstances where § 2 of the S 1errnan Act cannot."). 45 ti See generally Eugene Volkh and Donald M . Falk, Mayer Brown LLP, "First Amendment Pr tection for Search Engine Search Results" (Apr. 20, 20 12). 48 ~ 2003 U.S. Dist. LEXlS 27193 (W.D. Okla. 2003). 487 !d. at *3. 4 ~R See Kinderstart LLC v. Goog/e, Inc., 2006 U.S. Oist. LEXlS 8248 1, at *30 n.6 (N.D. Cal. Ju . 13, 2006) (although not specifically reaching the issue, noting that Goog1c's manipulation of it.s search re Its might be distinguishable from other fotms of protected expression because Google is not a media defend~nt, and website ranking may be of little or no public concern, citing Jefferson County School Dist. No. R-1 v. Moody's Investor's Services, inc. , 175 F.3d 848, 852 (I 0'11 Cir. 1999)). 4 9 a See Cemral 1/udson Gas & Elec. Corp. v. Puhlic Se11>. Comm'n. 447 U.S. 557,667 (1980). 490 Edenfield v. Fane, 507 U.S. 761 , 767 (1993) (quoting Ohralik v. Ohio State Bar Assn., 436 U.S. 447,457 (1978)). 491 540 u.s. 398 (2004). 492 /d. at 408 (quoting United States v. Colgate & Co., 250 U.S. 300. 307 ( 1919)): :.ee Pac. Bell Tel. Co. v. Link/ine Communications. inc., 129 S.Ct. II 09, 1118 (2009) ("As a general rule, businesses are free to choose the parties with whom they will deal, as well as the prices, tem1s, and conditions of that dea ling!') . 493 !d. at 408; see Lin/dine. 129 S.Ct. at 1118 (acknowhedging "limited c ircumstances in which~ fim1's unilateral refusal to deal with its rivals can give rise to antitrust liability"). 4 14 472 U.S. 5~5 ( 19S5). ' 49 .1 Trinko, 540 U.S. at 409. 49 ~ Trinko, 540 U.S. at 408-09 (describing Aspen Skiing). 497 hi. at 409 (emphasis in original). 498 93 Fed. Appx. l (5'" Cir. 2004) (unpubl ished op.). 499 fd. at 3. 500 /d. at 4. 501 Jd. at 9-10. See also, e.g., Creative Copier Servs. v. Xerox Corp., 344 F.Supp.2d 858, 866 ( . Conn. 2004) ~a ll owing plaintifrs ref~tsa l t_o deal clain~s t? go fotwa~·d where plaintiff alleg~d tha_t defen?ant ¥erox eng~ged 111 a voluntary course of dealmg wtth plamttff, then unilaterally "stopped dealmg wtth [plamtt11l or made 11 difficu lt for [plaintifl] to deal with Xerox" without a legitimate business justification). Conversely, several courts have dismissed complaints that have failed to properly allege a " unilateral termination ofla vohmtary course of dealing." See, e.g., Covad Communications Co. v. Bell Atlantic Corp., 39R F.3d 666, 673 (D.C. Cir. 2005) (upholding dismissal of complaint where, among other things, plaintiff failed to allege that "the defendant had previously engaged in a course of dealing with it<> rivals, or that it would ever have done so absent statutory compulsion" (in ternal quotation omitted); LiveUniverse. Inc. v. MySpace. inc. , 304 Fed. Appx. 554, 556 (9'11 Cir. 1009) (unpublished op.) ('holding that a refusal to dea l claim requires a prior affim1ative decision or agreement to cooperate, and upholding dismissal of complaint where voluntary, aflirrnative prior course of dealing was not alleged); In re Elevator Amitmst Litig., 503 FJd 47, 52-53 (2d Cir. 2007) (interpreting the "sole exception'' to a defendant's right to refuse to deal as '\.vhen a monopolist seeks to terminate a prior (volu ntary) course of dealing with a competitor," and dismissing complaint where plaintiff failed to allege this). Sec also Areeda & Hovenkamp, 772h ("As a general marter, court-imposed sharing obligationlcreated under the very general provisions o f the antitrust laws must be restricted to circumstances where the d(endant !52 e m Docu SJ W a i v s nt offer to deal with a competitor on unreasonable terms and conditions can amount to a practical r.efusal tO dear'); Aspen Skiing, 472 U.S. at 592-93 (noting that defendant offered plaintiff joint ticket deal provitd that plaintiff agreed to accept a fixed percentage of profits considerably below plaintilis historical average, at a member of defendant's board of directors admitted that defendant made an offer it knew plaintiff would no accept, and that on those facts, plaintiff did reject defendant's offer); Duke Energy, 93 fed. Appx. at 4 (premisi g liability for refusal to deal on offer with terms that defendant "knew were unrealistic or completely unviable'' to plaintiff); Creative Copier Servs., 344 F.Supp.2d at 866 (allowing refusal to deal claim to proceed based defendant's delays in shipping, making certain parts unavailable, and raising prices on other parts). See als Areeda & l lovenkamp ~772cl (noting that, in Aspen Skiing, defendant did not actually refuse to deal with plaintiff, but kept trying to reduce plaintiff's share of the profits until it "finally made an offer that [plaintifl] vould and did find unacceptable"). ~ 1 ° C.f in the Matte1· of Intel Corp., 128 FTC Decisio ns 213 ( 1999) (challenging lntd 's threat to ut off customers from critical technical information unless those customers granted Imel Iicenses tote hnology developed and owned by the customers). Sl l See, e.g. , In the Mauer o,/Negotiated Data Solutions UC, FTC File No. 05 1-0094 (2008) (c9ndemning, as unfair method of competition under Section 5, N-Data's reneging on prior patent owner's pri cin~ commitments to stand.ard-setting organizatio~'. where (i) the conduct caused "subs t~~tial consumer injury" tha~ (ii) was "~-~t : . . outweighed by any countervallmg benefits to consumers or compenllon that the practice prodt~;ces," and (m) 1t was an injury that "consumers themselves could not reasonably have avoided") (quoting Orkin Exterminating Co. v. FTC, 849 F.2d 1354, 1364 (lith Cir. 1988). 512 See Analysis of Proposed Consent Order to Aid Public Conunent, In the Malter ofImel Cor ., 128 FTC Decisions 213 ( I 999), at *3 ("Unjustified conduct by a monopolist that removes the incentive t~ ... compet[e] by depriving innovators of their reward or otherwise tilting the playing field against new entrants or fringe competitors ... has a direct and substantial impact upon future consumers"). 513 See, e.g., .vticrosoft IR (J ul. 23. 20 12} (Qi Lu referencing well-known Silicon Valley investor who has allegedly pulled funding from a variety of vertical websites). 14 ~ Mayer Tr. 152: I 9-24 (".. . it's not possible to be dropped in one place and not the other"). SIS See supra p. 37. Similarly, Google's almost immediate removal of Amazon product reviews from Google Product Search indicates that technical barriers were quickly surmounted when Google desired o accommodate a partner. 516 Google allows newspapers to choose to be indexed for Google's web search results, but not~y Google News. Sec Jonathan Simon, New User Agent for News, Google WcbmasLer Central. Dec. 2, 20 9, http://googlewebmastercentral.blogspot.com/2009/ 12/new-uscr-agcnt-for-news.hunt; David Sm dra, Googh~ News Now Crawling with Googlehot, Google Webmaster Central , Aug. 25, 20 I I, htt :// • o •lewe mastercentral.blo s ot.comi2011108/ oo le-news-now-crawlin -with- oo lebot.html; Vanessa Fox, Google Retires the Googlebot-News Bot, Search Engine Land, http;//scarchcngi neland.com/google-retires-the- googlebot-news-bot-90607. The primary di fler nee between Google News and the affected verticals here is that Google makes little money from Google Ne~s as a standalone product. Presumably, this lower-value vertical is one in which Google was willing to make certain concessions that it was not willing to make in higher-value vertical areas. m As demonstrated in the Microsoft opinion, courts are deferential in their treatment ofproduc innovations with gen uine procompetitive qualities. See Microsoji, 253 F.3d at 75-76 (reversing finding of l i~bility with respect to Microsoft development of a java script that allowed improved pe1formance, but was ihcompatible with the j ava script pioneered by Sun .vticroSystem, Inc.); sea a/s(J Allied Orthopedic, 592 F.3d at 998-1002 (finding that the inlroduction of improved sensors that were incompatible with competitors' monitoring systems was no t antioompctitive). However, when evaluating comracrual restrictions attached to the product. the Microsoft court had no trouble evaluating those contractual restrictions separately from the products they were attached to. See Microsoft, 253 F.3d at 59-63 (condemning licensing restrictions for harming rivals, «not by improving its own product, but, rather, by preventing OE.vts from taking actions that could increa;;e rivals' share of usage"). This distinction demonstrates why the consumer choice model described in the BE Staff Memo of Jan. 3 I, 20 I 2, at 23-24, frames a false choice. With the model. BE Staff compared overall welfare of advertisers with the API plus restrictions versus their welfare if no API existed. There is no support in the case law for limiting the choice in such a way when there is a third choice: the API without the restrictive conditions. The analogous argument in the Microsoft case would have prevented the courts from considering the possibility 154 SJ W a i v s nt e m Docu correct ROI to their client and would thus be out-of-business.''); Clickable IR (Oct. 24, 20 11) ("Although advertising across multiple platforms requires Clickable to use additional resources, Clickable wants to encourage this behavior nonetheless ... [a<;] advertising across multiple platforms helps its clients achieve the highest return on investment (ROT).") Didir.com (Dec. 27, 2011) ("Didit manages client campaigns to maximize ROI."); lnterpublic IR (Oct. 20, 201 1) (noting that "t he search advertising market is 'effectiveness driven,' ... ."); Kenshoo lR (f"ov. 9, 2011) ("Kenshoo's software is structured to primarily emphasize returp on investment (ROI) and scale, and to secondarily address brand awareness and exposure."); Raven Tools lR (Feb. 28, 2012) ("Raven is limited on what they can do, so they focus their energy o n where lhcy see the most return.") Reach Local IR (Jan. 12, 20 12) ("[t]he value in Reach Local's advertising campaigns stems from the return on investment, time and opp01tunity cost savings, access to technology and software, and the kno\~fedge of its staff.''). m lloldt:n Tr. 64:20-65:9. A searc h for SEMs reveals hundreds of firms offering these services see also, Varian Tr. I 07:4-1 08:4 (explaining that ad agencies act in a non-zero sum gam~; and their role i a positive one for Google); Varian Tr. 149:22-150:11 (where there are numerous advc1tising agencies "they w · uld try to compete in providing functionality and, of course, costs of developing tools that are appropriate to the needs of their c lients ... like any competitive market, they would try to address the needs of their potent~~ I customers."). ~JR CX-37 (GOOGWOJC-000031755-64) (2008), at 58 ("Market forces are going to protect G0ogle. Their (3'd party, agencies) customers will drop that customer/agency. To the extent that someone is add in~ spammy stuff -they are going to worsen their own performance and this won't work out in the long run."), at 59 ("Won' t market forces drive developers to adopt [all AdWords functionality]? C ustomers will hound yoL or leave if you don't offer it"). 9 s> To the extent SEMs and agencies have misaligned incentives, it would be with non-dominant search networks, because the third parties' ft.rst priority would be to improve their clients' returns on Ad Words, the largest search network, before optimizing on others. See, e.g., Varian Tr. 135:11-17. 540 GOOGAROR-000018605-16 (2006), at II (emphasis added). ~~ Holden Tr. 31:19-32: 16 (Google does not have reliable infom1ation about the ROI of advertisers using agencies and SEMs); id. at 129: I 0- 130:14 (no record of any harm to Google from SEMs that were violating terms and conditions). ~ 2 Holden Tr. 31:22-32:7 ("typically our assessments come back that rate of spend increases on advertisers working through agencies."). See also CX-41 (GOOGFOX-000128077-80) (2009), at 77 (Goile study limling lhat advertisers who use SEM tools have about I 3% higher spend growth than advertise who only use the AdWords Front End). 543 Holden Tr. J 2S:7-IJO: 14; cf Google Submission to the EC, "Google's /\dWords API Terms and Conditions Do Not Have Anti-Competitive Foreclosure Effects - An Analytic Framework" (Sep. 23, 20 11 ), at 19 ("Google 's online Ad Words guide explains: "getting the most ow ofAdWorcf.y requires ongoing exferimentation. ") (emphasis in the original). 51 ' AdWords Terms and Conditions, 111.2.f. provides: "All AdWords API Clients must expose ~t least as much functionality as is set forth in the RMF List. If the RMF List includes a particular fimction, all a pects of that function a nd all API calls related to that function must be enabled and exposed. AdWords API :tients will need to expose any add itional functionality added to the RMF List within 4 months after those functi9nalities are added to the RMF List." The list of requirements is updated pe1iodically and posted by Google. See Google Developers, Required Minimum Functionality, ht s://develo ers. oo le.com/adwords!a i/docs 're uirements past visited Jul. 25, 2012). 4 ~ CX-192 (GOOGVARI-000006959R-61R) (2004), ar 6 1R. Later in that thread, J..Tal Varian ;s noted as saying, "We're the dominant incumbent in this industry; the folks pushing us to develop our APJ will be the underdogs trying to unseat us." !d. at 60R. <46 GOOGKAMA-000004812-13 (2004), at 12; see also GOOGKAMA-000015528 (2006), at~ (in response to concern about Google advertisers migTating to MSN AdCenter. Google's response is "fight commodi tization of search networks by enforcing Ad Words API T &Cs with SEMs"). · 547 GOOGKAMA-000004815 (2004), at I. ~4ll ld. 49 s Ad 'Words API Tenns and Conditions, section 111.2.f ("All Ad Words API Clients must expos at least as ll'lUCh functionality as is set forth in the Required Minimum Functionality List."). 156 e m Docu SJ W a i v s nt y.a The ''conservative" estimate includes in the '"foreclosed" group only the companies that have explicitly complained to the Commission that agreements foreclose them from using a rival syndication service, and that they would like to do so, but for their current agreement with Googlc. This group includes only cBay. It should also include ?\exTag and Business.com, however the comScore dataset does not provide numbe for these firms. The comScore dataset suggests that, under this scenario, 8,653.366,936 queries, or some 19.6 percent of the market, is foreclosed. (If lAC is included within this group, the foreclosed query volume in~reases to 16,447,977,342, or some 37.3 percent of the market.) This is an extremely conservative estimatf because, as noted above, courts routinely include all sales made pursuant to an exclusive agreement as bein foreclosed. sns The "aggressive" estimate includes in the ·'foreclosed" group every company that is party to n exclusive agreement with Googlc (sec Appendix I, table showing exclusive agreements), as well as every company that is party to an agreement with the challenged "preferred placement" provision (.1ee Appendix 2, tab e showing "preferred placement" agree ments), except for any party that has explicitly told us that they do ot view the "preferred placement" provision as a baiTier to the use of a rival's syndication service. The excl1ded group includes Amazon, Wai-Mart, and Google's online partners. Also excluded is Ear1hlink, althoug the comScore dataset does not provide numbers for this finn. The comScore dataset suggests that, under this cenario, 29, 133,927,g82 queries, or some 66. 1 percent of the market is foreclosed. 566 The "intermediate'' estimate includes in the "foreclosed" group every company that is party t an exclusive agreement with Google, as well as any company that is party to "preferred placement" terms an~ has explicitly complained to the Commission that these terms foreclose them from using a rival syndication service, and has stated that they would like to do so, but for their current agreement with Google. In addition to all partners with an exclusive agreement (see Appendix I, table showing exclusive agreements), this group includes: eBay (and s hould include NexTag and Business.com, but does not, given the limitations of the comScore dataset, described above). The comScore dataset suggests that, under this scenario. 22.804,213.204 queries, or some 51.5 percent of the market, is foreclosed. We believe that this is the most defensible position because it takes into account both the exclusive agreeme nts- those companies literally forcclo~ed to competitors on the face of their agreements - as well as any "preferred placement" agreements for companies that have explicitly complained about the de facto exclusive effect of s uch agreements. Staff believes that this approach is consistent with case law. See Omega Environmental. Inc. 127 F.3d at I I 62; Still Spark Plug Co., 840 F.2d at 1258. ~"1 20 I I comScore qSearcb20 Repon. ~~ Microsoft, 253 F.3d at 64; see also Robert II. Uork, The Antitrust Paradox 158 ( 1978) ("But here is no doubt that predation can succeed when the distribution pattern is so much more ellicient than the alternative that those forced out of the pattern cannot compete"). SM Ryko Mfg. Co. v. Eden Senw .. 823 F.2d I 215, 1233 (8th Cir. 1987) ("When the degree offoreclosu.re caused by the exclusivity provisions is so great that it invariably indicates that the supplier imposing the provisions has market power, we may rely on the foreclosu re rate alone to establish the violation."). 70 s See Tampa Elec·tric Co. v Nasf1vilfe Coaf Co., 365 U.S. 320, 329 ( 1961); In re Belfone Elect1onics Cmp., I 00 FTC 68, 204 { 1982). m See Micro oft IR (Jul. 20, 2012); Microsoft IR (Qi Liu, Jul. 23, 2012) (reporting that Microsoft's people search program is better than Google's because Uing has access to Faccbook data and that Bing built a better search system for travel queries than Google has.) ~72 lAC IR (Dec. 8, 20 11 ). m !d. indeed, as CityGrid explained, there are approximately 15- 17 million individual local businesses that hope to attract local customers throughout the United States. These local businesses are potential advertisers for which search advertising (particularly search advertising serving specialized or "tail'' queries) can deliver a very high return lor investment. As such, these markets are highly lucrative for Google. and competition for this advertising revenue from specialized web-sites, such as CityGrid and UrbanSpoon, aggregately poses a significant competitive threat to Google. For reference, competition in serving these local and specialized (vertical) markets is the same competitive threat Google contemplated it its 2007 EU planning document entitled, "Online Advertising Challenges: Rise of the Aggregators." wherein Google saw local advertising saw as a markets in Europe as having many companies experimenting to lure ad\'ertisers it what Google "winner take all" market. See CX- 116 (GO OG-Tcxas-14869 15-70) (2009), at 21. m lAC IR (Dec. 8, 2011). m id. 158 e m Docu SJ W a i v s nt ourside of major syndication platfonns (i.e., those with sii,'llificant query volume, such as AOL and lAC), the company has not been focused on winning new search syndication business. Microsoft IR (Jun. II, 2012). 581 See Yahoo! IR (Scp. 14, 2011). Google and Yahoo! abandoned their proposed arrangement in the face of a threatened challenge by the Department of J1.1stice. Ultimately, Microsoft and Yahoo! entered i1o a similar arrdngement in 20 I0. m See :.upra p. 55. 59 lowering H See, e.g. GOOGKAP0-000006280-95 (2010), at 83 (discussing revenue improvements fro revenue share and standardizing AdSense agreements with publishers); Business.com IR (Jun. I , 20 12); Time Warner Cable IR (Sep. 8, 20 I I). 590 Amazon IR (Feb. 15, 2012). 591 /d. Google has offered this remedy to the European Commission as part of its settlement propos I. See GoogleEC Seulcment Proposal at 15-16. ~ 91 Google has offered this remedy to the European Commission as part of its settlement propos I. See GoogleEC Settlement Proposal at 26-27. 594 Acquisio IR (Sep. 12, 201 1); Resolution Media 1R (Nov. 7, 2011); Microsoft IR (Sep. 23, 20~ 1). s9s Google has offered some version of a non-exclusivity remedy to the European Commission as part of its settlement proposal, but has excluded ce11ain classes of syndication partners from its proposal. $ee Google-EC Settlement Proposal at 21-22. As such, we do not believe that Google's offer is sufficient to rerbedy the conduct addressed in this memorandum. <% Adam Kovacevich, Coogle ·s Approach 10 Comperirion. Google Public Policy Blogspot, Ma)! 8, 2009, http://googlepublicpo1icv.blogspot.com12009/05/googles-approach-to-compctition.html. I Wl 160