TABLE OF CONTENTS I. INTRODUCTION II. BACKGROUND ON THE ISSUES A. Overview B. Search Engine Output Architecture and Labeling C. Keyword Auctions D. Search Engine Policies Regarding Trademark Usage 1. United States 2. European Union 3. Other Regions III. SEARCH CONTROVERSIES OVER TRADEMARKED KEYWORDS A. Case Law 1. Overview 2. Judicial Assumptions A. Consumer Goals and Expectations B. Intent C. Consumer Knowledge of and Attentiveness to Search Page Architecture and Labels D. Appearance of Plaintiff's Mark in Defendant's Ad Text E. Likelihood of Diversion F. Likelihood of Confusion B. Academic Scholarship C. Private Litigation--Consumer Surveys IV. EMPIRICAL FINDINGS A. Coding Study B. Surveys 1. Overview 2. Background on A. 1st Survey B. 2nd Survey C. 3rd Survey 3. Survey Findings A. Consumer Knowledge of Search Page Architecture B. Adequacy of Disclosure of Paid Links C. Consumer Preferences and Expectations D. Consumer Attentiveness to Search Page Architecture and Labels E. Paid Link Click-Through F. Diversion and Confusion G. Fairness Norms H. Regression Analysis V. DISCUSSION A. Limitations of Our Findings B. Framing of the Trademarks as Keywords Debate C. Search Page Architecture and Labels D. Trademark Inclusion in Ad Text E. Intent F. Diversion G. Likelihood of Confusion H. Survey Complexities I. Whither Trademark Law: Confusion, Free-Riding, or Both? J. The Perils of Casual Empiricism K. What is Really at Stake? VI. CONCLUSION APPENDIX ONE: TABLES
Google, Bing, and Yahoo are the primary gateways to the Internet for most people in the United States. (1) Google is worth more than $260 billion, and Yahoo is worth more than $25 billion. (2) These lofty market capitalizations are almost entirely attributable to the income generated by the advertising that accompanies search results. (3)
Most searches result in one or more paid ads appearing alongside the unpaid (organic or algorithmic) results. The specific ads that appear are selected because they relate to the search terms ("keywords") entered by the user. For example, a search for "bicycle" will return ads from stores and websites selling bikes as well as bike manufacturers. A search for "wedding" will return ads from stores and websites selling wedding supplies, wedding dresses, and wedding cakes. A search for "mesothelioma" will return ads from plaintiffs' attorneys. Each of these entities pays the search engine if its ad is clicked on, irrespective of whether a sale is ultimately made. (4)
When search engines began offering ads using trademarks as keywords, disgruntled trademark owners filed more than one hundred lawsuits in the United States and Europe. (5) Despite the volume of litigation, there has been little independent empirical work on consumer goals and expectations when using trademarks as search terms, on whether consumers are actually confused by search results, and on which entities are buying trademarks as keywords. Instead, judges have relied heavily on their own intuitions, based on little more than armchair empiricism, to resolve such matters.
We report on the results of a two-part study, including three online consumer surveys and a coding study of the results when 2500 trademarks were run through three search engines. Consumer goals and expectations turn out to be quite heterogeneous: a majority of consumers use brand names to search primarily for the branded goods, but most consumers are open to purchasing competing products. (6) We find little evidence of traditional actionable consumer confusion regarding the source of goods, but only a small minority of consumers correctly and consistently distinguished paid ads from unpaid search results, (7) or noticed the labels that search engines use to differentiate paid ads from unpaid search results. (8)
We also find that the aggregate risk of consumer confusion is low, because most of the ads triggered by the use of trademarks as keywords are for authorized sellers or the trademark owners themselves. (9) Perhaps our most intriguing finding is the sizeable mismatch between consumer sentiments and the protections provided by U.S. trademark law. After we excluded those who were unsure or had no opinion, survey respondents were evenly split on whether it was fair and appropriate for competitors to purchase one another's trademarks for use as keywords, even without any confusion as to source, sponsorship, or affiliation. (10) These findings may explain why European trademark law recognizes a cause of action for taking unfair advantage of a trademark. (11) Although the law need not precisely match common moral intuitions, our findings suggest that it may be desirable to create a similar cause of action under U.S. trademark law.
Although we do find some evidence of confusion, the types of confusion we document do not map neatly onto the categories recognized by U.S. trademark law. our findings suggest that the development of the doctrine in this area has not been well served by the reliance of judges on casual empiricism in resolving these disputes. Much remains to be done to ensure that trademark doctrine is empirically well-grounded, and "fits" the online context.
Part II provides some context for this dispute, including background on search engines and keyword searches. Part III outlines the extensive litigation, both foreign and domestic, over the use of trademarks as keywords, and identifies six assumptions that judges have made in resolving these cases. Part IV presents our empirical results. Part V discusses our findings, and Part VI concludes.
II. Background on the Issues
In 2004, Google started selling keywords that were also trademarks. (12) Conflict quickly arose. The problem is straightforward. If I run a search for American Airlines, and Delta Airlines appears in a paid ad on the search output page because Delta Airlines purchased "American Airlines" as a keyword, does American Airlines have any recourse? If so, against whom? Google? Delta Airlines? Both? Neither? What if Travelocity, which sells flights on both American Airlines and Delta Airlines, appears in a paid ad? Should the outcome turn on whether the paid ad uses the words "American Airlines" in the ad text?
Trademark law is primarily intended to prevent confusion about the origins of trademarked goods and services, but is it plausible that a consumer who searches for American Airlines and then buys a ticket on Delta Airlines was ever confused about which carrier he will be flying on? Even if the consumer is not confused about the airline he ultimately selects, should diversion of attention, however temporary, create a cause of action?
Should the mode or level of trade make any difference in the analysis? For example, searchers can obtain a reservation for a room at a Hyatt hotel either directly from Hyatt's website or from a travel website (e.g., Orbitz, Travelocity, getaroom.com, etc.). Hyatt makes more money if searchers deal directly with them, instead of going through a travel website. (13) Does Hyatt have a valid complaint if travel websites purchase the Hyatt trademark as a keyword? Should the outcome turn on where consumers ultimately make reservations after searching for "Hyatt"? Does the fact that Hyatt makes less money if reservations are made through travel websites have any legal significance?
Of course, Hyatt can capture some of these reservations if it bids on its own trademark, but should Hyatt have to pay for the use of a trademark it already owns? (14) Hyatt could prohibit travel websites with which it does business from purchasing its trademark as a keyword, but that would leave the field open to its competitors to buy higher placement for their ads. (15)
What if someone searches for Rolex, and three ads come up: one for Seiko, one for a seller of counterfeit Rolexes, and one for a store selling genuine Rolex and Seiko watches. Does Rolex have a case against any or all of them? If Rolex has a case, what needs to be established? Does the bare fact that a competitor or counterfeiter purchased another company's trademark as a keyword establish the necessary elements of trademark infringement? Are there any defenses available to those who purchased a trademark to use as a keyword, and to the search engine that sold it? Does it matter if the resulting ad is simply comparative? For example, "our watches have the same design features as a Rolex, but cost less." Does it matter that the store sells both Rolex and Seiko watches? Does it matter if the individual who conducted the search was using Rolex as a generic proxy term for "expensive watches"?
These questions are not law school hypotheticals. In well over one hundred cases in U.S. and foreign courts, disgruntled trademark owners have sued Google and other search engines, as well as the entities that have purchased trademarks as keywords. (16) Courts have varied in their approaches to these cases, in many instances showing considerable skepticism about the merits, (17) while in other instances expanding the law to include conduct outside the traditional ambit of trademark doctrine. (18) Judges in both camps have routinely engaged in armchair empiricism, making casual assumptions about such matters as why consumers use trademarks as search terms, consumer knowledge about the difference between paid and unpaid links, and the likelihood of confusion when competitors purchase one another's trademarks for use as search engine keywords. (19)
The sale of search engine keywords, whether trademarked or not, also raises interesting consumer protection issues. As noted previously, a search engine typically returns both paid and unpaid results. Consumer protection law requires a "clear and conspicuous" disclosure of paid content. (20) Are search engines complying with these requirements? How have search engines changed their descriptions of paid content over time? Do changes made by search engines result in greater awareness of the difference between paid and unpaid content? Do these changes affect click-through behavior? The Federal Trade Commission has expressed concern about the labeling and page architecture of search engine results, but has not brought any cases to date. (21)
We focus on Google in this Article because it is the dominant search engine. However, where appropriate, we describe differences in the way in which Google, Bing, and Yahoo present and label search results.
B. Search Engine Output Architecture and Labeling
We are confident that every reader of this Article has a general understanding of how Google organizes its search output. However, for those happy few who have no knowledge of Google's search page architecture, but for some reason have chosen to read this Article, Figure 1 is a screenshot of the results when "Mercedes" was used as a search term in 2011. (22)
[FIGURE 1 OMITTED]
As Figure 1 reflects, Google's search output from 2011 has several discrete sections. At the top of the page is a small search box, with links to various types of search output (e.g., web, images, videos, maps, news, shopping, mail, and "more"). To the left, a column largely replicates the links in the top zone, along with links to change the region that Google uses as the search location, and a tool with which to specify the time period searched. Below and to the right are three sections containing search information. The shaded top section, Section A, contains various search results and associated links, which are labeled "Ads." (23) The right-hand column, Section B, contains more search results and associated links, and is labeled "Ads" as well, though the background is not shaded. Finally, underneath Section A and to the left of Section B is Section C, with more search results and associated links.
The links in Section A and Section B are all paid ads. Each link that appears in these areas of the search results page is there because the relevant site won the right to appear by bidding in an auction of keywords run by Google. (24) Conversely, the links in Section C are unpaid "organic" or "algorithmic" content, which appear as a result of Google's search program. (25)
Bing labels its paid links using the same terminology ("Ads") and the same search page architecture. (26) Yahoo labels paid links with the term "Sponsored Results" but uses the same search page architecture as Bing. (27)
Prior to November 2010, Google labeled paid links as "Sponsored Links." (28) Prior to April 2011, Bing labeled paid links as "Sponsored Sites." (29) other search engines have used a wide array of labels to identify paid links at one time or another, including "Featured Listings," "Premier Listings," "Recommended Sites," "Search Partners," and "Spotlight." (30)
C. Keyword Auctions
Google began AdWords, a program of selling ads based on specific keywords, in 2000. (31) AdWords took its current payment-per-click form in 2002. (32) In 2004, Google eliminated many restrictions on the use of trademarks as keywords. (33) We discuss below the circumstances under which trademarks may be included in ad text.
Individuals and entities bid to have their ad appear when specified keywords are used as search terms. (34) Whether a particular ad appears depends on various factors, including the details of the search query, the amount that is bid, past performance of the ad in the context of such searches (i.e., click-through rates), and whether and how the bid is limited by the bidder. (35) For example, bidders can target their ads by location, time, search device employed, and language. (36) When consumers click on an ad, the entity that purchased the keyword in question pays Google the amount it bid, irrespective of whether any sale results. (37)
AdWords and the associated pay-per-click ("PPC") payment model are responsible for the overwhelming majority of Google's income and for Google's extraordinary market capitalization. For example, in 2010, revenue from advertising totaled $28.2 billion, 96% of Google's total revenues. (38) By 2011, revenue from advertising had climbed to more than $36 billion. (39) Bing and Yahoo use a similar PPC model. (40)
Trademarks account for a material share of this advertising revenue. According to an internal Google document, trademarked keywords accounted for 7% of Google's total keyword revenues in 2004, even though Google honored requests from trademark owners to disable the use of trademarks in keywords and ad text for part of that period. (41) In 2009, Google estimated that allowing the use of trademarks in ad text, which it had previously sharply limited, would result in at least $100 million in increased annual revenues. (42)
D. Search Engine Policies Regarding Trademark Usage
Google, Bing, and Yahoo have very detailed policies regarding trademark usage and infringement. Bing and Yahoo's policies are identical because of a search alliance agreement between Microsoft and Yahoo. (43) As of 2009, Google allowed trademarks to be purchased as keywords in more than 190 countries. (44) Because the policies vary by region, we have broken out our discussion accordingly.
To the extent these search engines police the use of trademarks--which varies based on the jurisdiction and on whether the trademark is being used as a keyword or in ad text--they do so using an approach analogous to the "notice and takedown" system through which online copyright infringement disputes are handled pursuant to the Digital Millennium Copyright Act. (45) Thus, search engines do not actively police the use of trademarks as keywords, but instead respond to complaints by trademark owners, and then only when those complaints fall within the boundaries that the search engines have set. (46)
1. United States
In the United States, all three major search engines allow trademarks to be purchased as keywords, and none of the three has a formal policy for investigating or disallowing future purchases in response to complaints by the trademark owner. (47) Google has had this policy since 2004, when it first allowed trademarks to be sold as key words, (48) while Bing and Yahoo did not formally announce this policy until 2011. (49)
Prior to 2009, Google did not allow trademarks to be used in ad text, and removed such ads in response to complaints by the trademark owner. (50) Since then, Google has allowed trademarks to appear in ad text so long as doing so constitutes "fair use" and the advertiser is an informational site, a reseller, or a seller of components, replacement parts, or compatible products. (51) If a trademark owner complains about the use of its trademark in ad text, Google will conduct an investigation to assess compliance with its policies. (52) Alternatively, a company may request that Google prohibit all use of its trademark in ad text by all advertisers. (53)
Like Google, Bing and Yahoo allow trademarks to appear in ad text so long as doing so constitutes fair use, although Bing and Yahoo's definition of fair use is more expansive than that of Google's. (54) Bing and Yahoo's enforcement policy is complaint-driven. (55)
2. European Union
Google began selling trademarks as keywords in the United Kingdom and Ireland in 2008. (56) It expanded this policy to other European countries in 2009, (57) and again in 2010. (58) If a trademark owner complains, Google will conduct a limited investigation to determine whether a specific ad in combination with a specific keyword creates confusion as to the origin of the advertised goods and services. (59) If Google concludes that the ad and keyword combination is confusing, it will remove the specific ad causing the confusion. (60)
In Europe, Bing and Yahoo have a policy on the use of trademarks as keywords only in the United Kingdom, France, Italy, and Ireland. (61) In these countries, Bing and Yahoo prohibit the use of a trademark as a keyword if the advertiser is a competitor of the trademark owner. (62) If the advertiser is a noncompeting third party, Bing and Yahoo explicitly permit the use of a keyword trademark so long as the advertiser's primary offering is not goods or services that compete with the trademark owner, and the advertiser is either an informational site or is using the term in a descriptive sense. (63) Bing and Yahoo also permit the use of a keyword that corresponds to a trade mark if the advertiser is selling authentic trademarked goods. (64)
For ad text, Google's policy in Canada, the United Kingdom, and Ireland has been identical to its policy in the United States since 2010. (65) In the rest of the European Union, Google restricts the use of trademarks in ad text. (66) If a trademark owner files a complaint, Google will investigate, and may disapprove future use of that ad. (67)
As with trademarks as keywords, Bing and Yahoo only have formal policies for ad text in the United Kingdom, France, Italy, and Ireland. (68) The policy is similar to that in the United States, but more restrictive, since the trademark may only be used in a descriptive sense or to advertise informational sites. (69) Thus, the website's principal offering must not be goods or services competitive with those of the trademark owner.
3. Other Regions
Google prohibits the use of trademarks as keywords in Australia, Brazil, China, Hong Kong, New Zealand, South Korea, Macau, and Taiwan. (70) However, a trademark owner must file a complaint before Google will disallow future use of that ad. (71) Google's review of the complaint is limited to determining whether the complainant's protected trademark has been purchased as a keyword. (72)
Google generally restricts the use of trademarks in ad text in all other regions. (73) Google does not proactively prevent the use of trademarks in ad text, but it will investigate once a trademark owner files a complaint with Google. If an advertiser is found to be using a trade mark in ad text, Google will disapprove future use of that ad. (74)
In Singapore, Bing and Yahoo's keyword policy and policy toward the use of trademarks in ad text mirrors its policies for the United Kingdom, France, Italy, and Ireland. (75) Bing and Yahoo have no other country-specific policies.
III. Search Controversies over Trademarked Keywords
As detailed below, the sale and purchase of trademarks as keywords has given rise to multiple lawsuits, both domestically and internationally. (76) In the United States, lawsuits have been framed around the issue of whether such transactions give rise to actionable confusion under trademark law. These disputes have also given rise to a massive outpouring of academic scholarship. Part III.A reviews the case law, and Part III.B reviews the academic scholarship. Part III.C analyzes the surveys of consumer confusion that private litigants have provided in connection with the litigation described in Part III.A.
A. Case Law
In the United States, no federal statute explicitly prohibits the use of trademarks as search engine keywords. (77) Trademark owners have accordingly turned to existing doctrines in trademark law for a potential remedy. There are two main doctrinal frameworks through which these cases may be viewed: (1) the likelihood of confusion (i.e., traditional trademark infringement analysis); or (2) the dilution cause of action. However, since at least 2006, dilution law in the United States has clearly been limited to cases of blurring or tarnishment, neither of which readily applies to most keyword cases. (78)
The paradigmatic example of blurring is when a company uses a close facsimile to a famous mark, like Google, for a wholly unrelated type of goods or services, like Googli Pancakes. In such a case, Google may be able to prove that consumers "think of Google when hearing or seeing the Googli Pancakes brand name. As a result, the Google mark is commercially diminished as a unique and strong brand identifier, because it is now associated with multiple sources in the minds of consumers. (79) If the pancakes taste bad, or are advertised with pornographic or violent images, Google may also have a tarnishment claim, again assuming consumers actually think of Google when they hear or see ads for Googli Pancakes. Dilution by blurring or tarnishment however, is specifically limited to nationally famous trademarks, (80) and it was designed for cases in which a third party uses a famous mark, or close facsimile thereof, on goods or services that are different than the goods or services upon which the famous mark owner is using them. (81) But, most of the litigation involving trademarks as search engine keywords features competitors who are selling similar goods to those bearing the trademark, which is quite different than the typical dilution by blurring or by tarnishment cause of action.
Not surprisingly, trademark owners have accordingly framed the dispute around traditional likelihood of consumer confusion trademark infringement, which turns on whether there is a likelihood of confusion as to "affiliation, sponsorship, or association." (82) To establish liability, the plaintiff must show that the defendant's use of a trademark leads to confusion as to source (i.e., some consumers think the defendant's goods actually are those of the plaintiff), sponsorship (i.e., that plaintiff has endorsed the defendant's goods), or affiliation (i.e., that the plaintiff and defendant are legally related entities). (83) Once likelihood of confusion is established, harm is usually presumed, and plaintiffs may obtain injunctive relief and provable damages, including the profits derived by the defendant from infringing sales. (84)
Courts have developed complex multifactor tests to assess whether there is a likelihood of confusion. The factors include: (1) the strength of the mark, (2) proximity of the goods, (3) similarity of the marks, (4) evidence of actual confusion, (5) marketing channels used, (6) type of goods and degree of purchaser care, (7) defendant's intent, and (8) likelihood of expansion of product lines. (85) To be sure, the circuit courts of appeals vary as to how many factors are included, and there is some evidence that district courts focus on only a few factors in deciding such cases. (86) Regardless, courts are more likely to find confusion if the marks are highly similar, the goods or services are alike, the defendant chose to use a similar mark in order to "free ride" on the goodwill of the plaintiff, the relevant consuming public is un sophisticated, and the parties sell to the same pool of customers. (87)
Courts have had difficulty applying this framework to keyword cases, and have responded by emphasizing the importance of some factors, and ignoring others. (88) Some courts have focused on "diversion," and imposed liability on that basis alone. (89) Other courts have said that mere diversion does not constitute trademark infringement; it must be shown that a searcher who was looking for X and was diverted to Y's website went there (at least initially) because she thought Y was affiliated with or sponsored by X. (90)
If mere diversion is insufficient, how much confusion must be shown to warrant relief? The plaintiff must show that an "appreciable" (i.e., more than trivial but less than substantial) number of relevant consumers are likely to be confused if defendant's activities are allowed to proceed. (91) Litigants may, but need not, offer survey evidence to prove or disprove the likelihood of confusion. (92) Courts have been unimpressed if the survey evidence shows that 10-15% of respondents are confused, but higher figures have been associated with greater success. (93)
2. Judicial Assumptions
In deciding keyword cases, judges have routinely made assumptions, based on casual or "armchair" empiricism, that have substantially affected the outcomes of the cases. We have identified six specific areas in which judges have made such assumptions: (1) consumer goals and expectations when trademarks are used as search terms, (2) advertiser or search engine intent when purchasing or selling a trademarked keyword, (3) consumer knowledge of and attentiveness to search page architecture and labeling of results, (4) the significance of the trademark appearing in the ad text, (5) the likelihood of diversion, and (6) the likelihood of confusion. We analyze each assumption in turn.
A. Consumer Goals and Expectations
Understanding consumer goals and expectations is critical in determining whether diversion is likely to occur, and is important but less critical in determining whether there is a likelihood of confusion. Consider two types of consumers: Consumer #1 has "narrow" preferences and Consumer #2 has "broad" preferences. When Consumer #1 types trademark X into a search engine, she is only looking for products bearing that trademark and expects to see such products and no others. Consumer #2, on the other hand, focuses on the product category rather than the specific brand. When Consumer #2 types trademark X into a search engine, she is using it as a generic proxy term to describe a category of goods. She could have equally well typed in trademarks Y and Z, which compete with products bearing trademark X. Consumer #2 welcomes information on products bearing trademarks Y and Z, and would be disappointed if her search engine provided only information on products bearing trademark X. (94)
It is certainly plausible, if not extremely likely, that Consumer #1 can be diverted by paid links for products bearing trademarks Y and Z. Conversely, it is implausible, if not impossible, for Consumer #2 to be diverted when presented with the same search output, since her original goals encompassed products bearing all three trademarks.
Matters get even more complex if consumers expect, based on past online and offline experiences, that a search for products bearing trademark X will also result in information for products bearing trademarks Y and Z. The case for diversion of Consumer #1 is weaker if she now expects, based on past experiences, to receive information on trademarks Y and Z, irrespective of the fact that she is only searching for products bearing trademark X. Stated more directly, for a meaningful claim of diversion to arise, the consumer must have both a fixed destination in mind, such as the specific branded product that she used as a search term, and must not have expected to encounter other branded products along the way. The relative proportion of consumers who have broad versus narrow preferences and/or expectations complicates matters further.
Consumer goals and expectations map less neatly onto the risk of confusion than they do onto the risk of diversion. However, it is plausible to assume that the likelihood of confusion is somewhat greater among those who are only interested in and expect to receive information about the branded good they searched for, while the likelihood of confusion is somewhat lessened among those who have more expansive goals or expectations. Thus, all else being equal, the distribution of goals and expectations within a population should have a material impact on the frequency with which courts find confusion.
Strikingly, rather than wrestle with these issues or demand that the parties provide direct evidence on these points, many judges have simply assumed that when a trademark is used as a search term, the consumer is interested only in goods bearing that trademark, or in the company that owns that trademark. For example, in Network Automation, Inc. v. Advanced Systems Concepts, Inc., the Ninth Circuit stated that consumers conducting a search with the trademark in question (Activebatch) "are presumably looking for [the] specific product, and not a category of goods." (95) The same assumption was made by the district courts in Rosetta Stone Ltd. v. Google, Inc., (96) Storus Corp. v. Aroa Marketing, (97) Harry J. Binder v. Disability Group, Inc., (98) and Australian Gold, Inc. v. Hatfield. (99) In Hearts on Fire Co. v. Blue Nile, Inc., the court did not go quite as far, but still noted that a factor to be considered in the confusion analysis was "the specific context of a consumer who has deliberately searched for trademarked diamonds only to find a sponsored link to a diamond retailer...." (100) This framing, and the underlying assumption on which it is based, makes it considerably easier for courts to conclude that actionable diversion has taken place, or that there is a likelihood of confusion. (101)
Ultimately, the distribution of goals and expectations among consumers is an empirical question. perhaps the stylized categories we have presented accurately capture what is going on. Alternatively, more categories may need to be employed, or it may be useful to think about consumer goals and expectations arrayed along a spectrum.
Consumers may also switch categories depending on what product or service they are searching for. For example, consumers may use trademarks as a generic proxy term for some categories of goods and services (e.g., Hertz for rental cars), but use other trademarks to search for specific products (e.g., Macbook for laptops made by Apple). Regardless, judges should not simply assume that consumers have homogenous goals and expectations, and are all equally susceptible to diversion or confusion.
In trademark infringement cases, courts have inferred "bad" intent when defendants knowingly use a mark that is similar or identical to that of a competitor. This approach has been transferred wholesale to the online world, with no consideration given to whether a different approach might be appropriate for keyword searches. For instance, in both Binder and Storus, the district courts treated the fact that another entity's trademark had been purchased as a keyword as dispositive on the issue of intent. (102) In Rosetta Stone, the Fourth Circuit concluded that a rational jury could infer bad intent from Google's financial incentive to sell trademarks as keywords, as well as Google's knowledge, based on its internal research, that doing so significantly heightened the risk of consumer confusion. (103) But if consumers routinely use trademarks as generic proxy terms, to refer to broad classes of goods, it does not make sense to automatically infer bad intent when advertisers use a trademark as a keyword.
C. Consumer Knowledge of and Attentiveness to Search Page Architecture and Labels
Courts routinely assume that consumers are knowledgeable about and rely upon search page architecture and the labels on paid links. Thus, in Playboy v. Netscape, the Ninth Circuit found a potential for initial interest confusion when consumers saw banner advertisements that were "confusingly labeled or not labeled at all." (104) The Ninth Circuit expressly observed that clear labeling "might eliminate the likelihood of initial interest confusion that exists in this case." (105) In Network Automation, the Ninth Circuit emphasized the importance of search page architecture and labeling:
[E]ven if [the Defendant] has not clearly identified itself in the text of its ads, Google and Bing have partitioned their search results pages so that the advertisements appear in separately labeled sections for "sponsored" links. The labeling and appearance of the advertisements as they appear on the results page ... must be considered as a whole. (106)
Indeed, the Ninth Circuit concluded that the "labeling and appearance of the advertisements and the surrounding context on the screen displaying the results page" was one of the most important factors in a trademark keyword case. (107) Similarly, in Rosetta Stone, the district court observed that confusion was unlikely because consumers "are able to distinguish between the Sponsored Links and organic results displayed on Google's search results page." (108)
These assumptions are material: consumers who understand search page architecture and can differentiate paid ads from unpaid search results are more likely to understand that ads may come from sources other than the trademark owner, even if they used the trademark as a search term in the first place. Conversely, if consumers do not understand search page architecture and the fact that some links are paid ads, they are arguably more prone to confusion. Because the degree of consumer knowledge about search page architecture and the fact that some links are paid ads is ultimately an empirical question, the assumption that consumers are knowledgeable about such matters weights the dice heavily against a finding of confusion.
D. Appearance of Plaintiff's Mark in Defendant's Ad Text
Courts have placed considerable weight on whether the ad heading or text includes the trademark in question, reasoning that the risk of confusion is far higher if the trademark actually appears in the ad itself. (109) Accordingly, courts have been more willing to infer confusion if the ad text includes the trademark in question. (110) Thus, in the GEICO trial, the district court dismissed the plaintiff's trademark infringement claims against Google for ad text that did not include the GEICO trademark, while finding that GEICO had demonstrated a likelihood of confusion, and therefore a violation of the Lanham Act, for sponsored links that use GEICO's trademarks in their headings or text. (111) As noted earlier, it is ultimately an empirical question whether the inclusion of the trademark in ad text increases the risk of confusion, but the judicial assumption that it does means that advertisers can dramatically lower the likelihood that courts will find confusion simply by omitting the trademark from ad text.
E. Likelihood of Diversion
Some courts have found in favor of the trademark owner when there was evidence of consumer diversion, but not much, if any, evidence of a likelihood of consumer confusion. (112) In these cases, courts have effectively assumed that diversion is a type of harm for which trademark law provides a remedy, virtually independent of evidence of consumer confusion. It is not entirely clear whether courts that take this step actually believe that diversion is legally equivalent to confusion, or are simply seeking to prohibit conduct they believe is normatively undesirable by stretching the existing doctrinal framework.
F. Likelihood of Confusion
The multifactor test described above in Part III.A.1 was developed to resolve disputes involving trademark infringement in the offline world. However, courts have used the same framework to resolve search engine keyword cases, discarding the factors that do not fit, and then applying the remaining factors. (113) Insufficient consideration has been given to whether the analysis should involve different criteria, tied to the realities of search behavior on the Internet. Courts have simply assumed that the same set of factors will work to cost-effectively identify confusion as to source, sponsorship, or affiliation on the Internet. But this approach means that factors that were developed to analyze likelihood of confusion in an offline world have effectively become dispositive endpoints in their own right, even when they are ill-suited to an online search environment.
B. Academic Scholarship
More than fifty law review articles and student notes have been written about trademark infringement in the context of keyword advertising. (114) We cannot begin to count the number of presentations made by practicing lawyers at CLE sessions on the subject.
Much of this work focuses on the "trademark use" controversy hotly debated at the outset of keyword litigation. (115) As that issue has waned in significance, articles and notes have increasingly focused on whether the initial interest confusion doctrine fits the online world. (116)
Strikingly, although there is a veritable mountain of materials on the legal issues raised by the use of trademarks as keywords, we have found very little empirical work on the subject, and none of it has been published in the law reviews. (117) O'Connor studied the use of trademarks as keywords for a sample of ninety hotels in Europe, Asia, and the United States, and found that "abuse is rampant," with ads for third-party websites appearing in a clear majority of searches. (118) Rosso and Jansen analyzed the frequency at which third parties' ads appeared in response to searches for 100 prominent trademarks and found that although such situations were common, occurring in 6493% of searches, competitors accounted for only 2.7-6.4% of "piggy backing" ads. (119) Further, very few of the piggybacking ads placed by competitors used the trademark in ad text. (120) Rosso and Jansen concluded that "competitive piggybacking does not appear to be a deceptive or widespread phenomenon." (121)
Several studies focused on other issues related to the use of trademarks as keywords. Chiou and Tucker studied the impact of including trademarks in ad text for hotel reservations and found that such advertising actually increased the demand for reservations from the hotel's own website--consumers clicked less often on paid ads, and more often on the organic link for the hotel itself. (122) They suggested that such "channel substitution" resulted from the fact that paid ads could no longer effectively differentiate themselves once all sites included the trademark. (123) In an unpublished doctoral dissertation, Shin developed a model for predicting when companies will and will not purchase their own trademarks as keywords. (124) Edelman and Gilchrist studied the impact of label text on click-throughs, and found that the use of "Paid Advertisements" resulted in a 25-27% lower click-through rate than "Sponsored links" or "Ads," respectively. (125)
Somewhat dated surveys also make it clear that consumers are not particularly familiar with the differences between paid and unpaid links, but distrust the former. In a 2004 survey, 62% of respondents were unaware that search engines provided both paid and unpaid search results. (126) Among those who were aware of the distinction, fewer than half (47%) said they could always tell which results were paid. (127) A 2003 survey found that respondents took little notice of labels and search page architecture, but they thought the term "sponsored" (128) was vague and confusing. Finally, a survey reported in a 2006 article found considerable suspicion about sponsored links, and "low expectation[s]" (129) about the value of such results. Not surprisingly, respondents reported that they preferred to click on unpaid links. (130)
C. Private Litigation--Consumer Surveys
Private litigation involving claims of trademark infringement has generated multiple surveys of consumer confusion. Table 1 specifies the rate of confusion quantified by thirteen expert reports offered in eleven different cases. For eight of these surveys, we had the full expert report, while for the other five surveys, we rely on the description of the survey in the court's opinion or in other materials.
Not surprisingly, defendants' experts invariably find low levels of confusion, while plaintiffs' experts invariably find higher levels of confusion. (144) In two cases, we have reports on the rate of confusion found by experts for each side. In American Airlines, the defense expert found a rate of confusion of 0-2%, while the plaintiff's expert found a rate of confusion of 20-32%. (145) In GEICO, the defense expert found a rate of confusion of < 10%, while the plaintiff's expert found a rate of confusion of 68-70%. (146)
Finally, the Fourth Circuit's opinion in Rosetta Stone references research conducted by Google in 2004. (147) Unfortunately, the actual study remains under seal, but the opinion states that Google found that the inclusion of a trademark in ad text, whether in the title or body, led to a very high degree of consumer confusion, noting that "94% of users were confused at least once" during the study. (148)
IV. Empirical Findings
A. Coding Study
As described previously, there have been numerous lawsuits arising out of the use of trademarks as keywords in Internet searches. But, little is known about the frequency with which such transactions occur, let alone who is doing the purchasing. We accordingly performed a study to determine who was purchasing trademarks as keywords. Because this study is the focus of another article, we provide only a brief description of this other study and one of its central findings here.
We obtained a list of approximately 2500 trademarks from the International Trademark Association, and developed a computer program to run an Internet search for each trademark through the three most prominent search engines (Bing, Google, and Yahoo). (149) For each trademark/search engine combination, the program captured a PDF file of the web page that would have been viewed had one clicked through each of the first ten paid links. We then developed a standardized coding protocol for classifying the search output, using eleven categories, including whether the paid link was for the trademark owner, an entity selling the trademarked goods as well as competing goods, or an entity selling competing goods exclusively. McCarthy Institute research fellows from the University of San Francisco coded the first five paid links for each trademark/search engine combination. (150) Table 2 contains details on the coding categories, and our results, sorted from most to least frequent.
As Table 2 indicates, vendors of the trademarked good and competing products account for 27% of paid links; collateral information/sales opportunity vendors (who provide a gateway through which to purchase the trademarked good) account for 24% of paid links; and the trademark owner accounts for 13% of paid links. Only 6% of paid links are purchased by entities selling exclusively competing goods. Thus, the overwhelming majority of paid links are unlikely to give rise to the types of consumer confusion at stake in the lawsuits that have been brought. Our findings are consistent with those of an earlier, smaller study of high-profile trademarks. (152)…