In a previous Memorandum Opinion addressing defendant’s Motion to Dismiss, the Court found that plaintiff had sufficiently alleged that defendant “uses” GEICO’s trademarks for purposes of the Lanham Act by allowing advertisers to pay to have their ads appear next to the organic listings that result when the marks are entered as search terms. GEICO, 330 F. Supp. 2d at 704. The Court also found that plaintiff had sufficiently alleged that such use was “in commerce” and “in connection with the sale, offering for sale, distribution, or advertising of goods and services,” rejecting defendant’s argument that the marks were used only in defendant’s computer coding and that such an invisible process could not constitute use under the Act. Id. at 704-05. As such, the Court denied defendant’s Motion To Dismiss because if the facts alleged in the Complaint were true, defendant could be liable, either directly or indirectly, for its use of plaintiff's trademarks to sell advertising and to place that advertising, labeled as Sponsored Links, alongside the organic listings resulting from searches on those marks. Id. at 704.
 

The Court, however, emphasized that its finding that Google uses the GEICO marks in commerce and “in connection with the sale, offering for sale, distribution, or advertising” of goods or services does not, without more, establish that defendant has violated or continues to violate the Lanham Act.⁷ Id. Rather, plaintiff was required to prove at trial that defendant’s use results, or at one time resulted, in a likelihood of confusion.
 

2. Likelihood of Confusion
a. The Applicable Standard

To establish a likelihood of confusion, a plaintiff must prove that the defendant’s use of the plaintiff’s trademark is “likely to confuse an ‘ordinary consumer’ as to the source or sponsorship of the goods.” PETA, 263 F.3d at 366. Of particular relevance to this action, in determining whether a likelihood of confusion exists, a court must consider the defendant’s use of the mark in its entirety, not in isolation. See id. The Fourth Circuit has adopted a list of seven factors to guide courts in this analysis: (1) the strength or distinctiveness of the plaintiff’s mark, (2) the similarity of the two marks, (3) the similarity of the goods and services that the two marks identify, (4) the similarity of the facilities that the two parties use in their businesses, (5) the similarity of the two parties’ advertising, (6) the defendant’s intent in adopting the same or similar mark and (7) the existence of actual confusion. Lone Star, 43 F.3d at 933. However, the Fourth Circuit has emphasized that likelihood of confusion is a highly factual issue, the assessment of which depends largely on the particular circumstances of each case, see Petro Stopping Centers, L.P. v. James River Petroleum, Inc., 130 F.3d 88, 92 (4th Cir. 1997), that not all of the factors are relevant in every case, and that the likelihood of confusion standard does not require that a plaintiff prove actual confusion. See Lone Star, 43 F.3d at 933. In addition, other factors, such as the sophistication of the consuming public or the quality of the defendant’s product, may be relevant to a given set of facts. Sara Lee Corp. v. Kayser-Roth Corp., 81 F.3d 455, 464-65 (4th Cir. 1996).

Recognizing that these traditional factors are not really applicable in this case, which does not involve confusion based on a defendant’s use of a mark identical or similar to a known mark to sell its own products, plaintiff argues that defendant’s use of its trademark causes “initial interest” confusion. In the Internet context, this term describes the distraction or diversion of a potential customer from the Web site he was initially seeking to another site, based on the user’s belief that the second site is associated with the one he originally sought. See Brookfield Communications, Inc. v. West Coast Entertainment Corp., 174 F.3d 1036, 1064 (9th Cir. 1999); Bihari v. Gross, 119 F. Supp. 2d 309, 320-21 (S.D.N.Y. 2000); see also PETA, 263 F.3d at 366 (discussing the appropriate boundaries of initial interest confusion with regard to parody). Inherent in this concept is the risk that the user will be satisfied with the second site or sufficiently distracted that he will not arrive at  or return to the site for which he was originally searching. Bihari, 119 F. Supp. 2d at 320-21. The risk of losing customers who are initially confused is lessened on the Internet as compared, for example, to when a billboard employs initial interest confusion to entice a customer down the wrong road because a customer can retrace his steps almost instantaneously online. See id. at 320 n.15. However, pointing to the unique nature of the car insurance business, GEICO contends that because customers seek an average of fewer than two quotes before purchasing car insurance, even in the Internet context the company loses significant business from the alleged initial confusion that misdirects potential customers who originally searched on “GEICO” to sites where they can obtain other companies’ quotes but not GEICO’s.
 

To prove likelihood or absence of confusion, initial or otherwise, parties commonly introduce the results of customer or potential customer surveys. See IDV North America, Inc. v. S&M Brands, Inc., 26 F. Supp. 2d 815, 829 (E.D. Va. 1998). Results indicating that a high percentage of customers are likely to be confused can establish this element of a Lanham Act violation, while results revealing a low potential for confusion can negate such allegations.⁸ Id. In assessing a survey’s probative value, a court must evaluate not only the percentages of respondents who
appeared likely to be confused but also the accuracy and reliability of the survey design and method. See id. at 829-32. Even if a survey is flawed in some respects, it can still support a finding of liability if the results reveal that an overwhelming percentage of respondents demonstrated confusion. See Sara Lee, 81 F.3d 455, 466-67.

b. Evidence of Confusion
(1) GEICO’s Survey Results
GEICO introduced an expert survey that it claimed revealed a strong likelihood of confusion caused by Google’s Adwords program and the Sponsored Links that the program places alongside the organic results of searches on the GEICO trademark.⁹ The survey was designed and evaluated by a professor of marketing at American University’s Kogod School of Business, who attempted to measure initial interest confusion experienced by potential customers¹⁰ by asking them to enter “GEICO” into the Google  search engine and then view a results page that showed five Sponsored Links alongside the organic listings.¹¹ The survey also tested a “control group,” the members of which also searched on “GEICO” but then saw a results page on which the Sponsored Links had been changed from advertisements for car insurance quotes, most of which mentioned GEICO, to advertisements regarding NIKE athletic apparel. After asking both groups a series of questions designed to measure the likelihood of confusion, the professor compared the degree of confusion demonstrated by those who viewed the insurance-related Sponsored Links to that demonstrated by the control group.¹²

According to the survey results, 67.6% of test group respondents expected that they would reach GEICO’s Web site if they clicked on the Sponsored Links, and 69.5.% thought that the Sponsored Links were either links to GEICO’s site or affiliated with GEICO in some way.¹³ In addition, 20.1% of test group respondents said that to purchase GEICO insurance they would click first on one of the Sponsored Links.¹⁴ From these results, the professor concluded that a substantial percentage of potential GEICO customers perceived the Sponsored Links as being
associated with GEICO and therefore were confused about whether they could get GEICO information or rate quotes from those links.
 

(2) Survey Weaknesses
The Court found that defendant’s cross examination of the professor revealed a number of weaknesses in GEICO’s survey evidence. First, the control did not successfully demonstrate the source of the test group’s confusion. As the survey expert admitted, an effective control should have removed from the page viewed by the test group the allegedly infringing elements for which GEICO wanted to measure confusion, such as the Sponsored
Links mentioning GEICO, while keeping the other elements as constant as possible. This would have allowed the evaluator to subtract any degree of confusion expressed by the control group from that expressed by the test group, with the resulting difference representing the confusion attributable to the eliminated elements–-similar to the manner in which medical researchers subtract out the “placebo effect” of a drug or procedure under examination. However, the survey’s control, which replaced the insurance-related Sponsored Links that appeared when respondents searched on “GEICO” with Sponsored links related to NIKE athletic apparel, did not function as an accurate measure of the confusion caused by non-infringing elements of the screen shot. As a threshold matter, the control retained the use of “GEICO” as a keyword, which itself was alleged to be a source of confusion. Further, instead of removing only the references to GEICO in the Sponsored Links, which would have measured whether the use of the trademarked keyword to place relevant Sponsored Links or the appearance of the mark in the ads was responsible for respondents’ confusion, the survey removed all references to car insurance and replaced them with clearly unrelated NIKE ads. Thus, the control did not reveal which aspects of the insurance-related Sponsored Links caused respondents’ confusion–-the use of GEICO’s mark in the ads or the ads’ mere reference to insurance.¹⁵ By not examining this  more subtle distinction, the survey did not produce evidence that the use of “GEICO” as a keyword, without more, causes respondents to be confused by the appearance of the Sponsored Links.
Second, the survey design introduced “demand effects” and “order effects” that could have tainted respondents’ answers that  appeared to indicate confusion. A demand effect results when th interviewer’s questions or other elements of the survey design influence participants’ responses by suggesting what the “correct” answers might be or by implying associations that might not otherwise occur to participants. An order effect results when a participant’s answer to one question affects his answers to subsequent questions. For obvious reasons, both effects can significantly bias the survey results. In this instance, as noted above, the interviewers repeatedly questioned respondents about their behavior and assumptions with regard to GEICO and obtaining GEICO quotes. Responses to subsequent questions, such as whether the participant thought the Sponsored Links were affiliated with any company, easily could have been influenced by the earlier questions about GEICO, making the participant more likely to assume that “GEICO” was the right answer or what the interviewer wanted to hear. This demand effect, along with the order effect, also undercut the assessment of confusion regarding the one insurance-related Sponsored Link shown to the test group that did not mention GEICO’s mark in its heading or text. Of the five Sponsored Links listed, only the last one did not refer to GEICO in its heading or text. By the time respondents considered this ad, they had already seen the four others that did contain the GEICO mark, calling into question whether they would have expressed similar confusion if faced only with insurance-related ads that appeared next to the organic results for “GEICO” but did not mention the mark itself.
Third, discrepancies between the Web page selected to be shown to the survey particpants and the actual pages users are likely to see when searching on “GEICO” further weakened the reliability of the results. As discussed above, the page chosen included more Sponsored Links than the average Google results page. In addition, and possibly more important to an accurate assessment of users’ confusion, the overall appearance of the outdated screen shot shown to survey participants differed markedly from the Google results page that a user would encounter in running a real search. Most notable, in the survey shot, the Sponsored Links were closer on the page to the organic listings
than on Google’s actual search results pages.¹⁶ In the Court’s view, the proximity of the Sponsored Links in the survey page increased confusion regarding the difference between the two lists of links and was suggestive of affiliations between the organic listings and the Sponsored Links. Especially when considered along with the other weaknesses in the survey, these differences give the Court serious doubts about the accuracy of the survey results’ reflection of actual users’ experiences with and reactions to the Sponsored Links.

c. Conclusions from Evidence of Confusion
Based on the above analysis, the Court finds that plaintiff has failed to establish a likelihood of confusion stemming from Google’s use of GEICO’s trademark as a keyword and has not produced sufficient evidence to proceed on the question of whether the Sponsored Links that do not reference GEICO’s marks in their headings or text create a sufficient likelihood of confusion to violate either the Lanham Act or Virginia common law. Despite the many flaws in its design, the survey’s results were sufficient to establish a likelihood of confusion regarding those Sponsored Links in which the trademark GEICO appears either in the heading or text of the ad. Based on this finding, Google may be liable for trademark infringement for the time period before it began blocking such usage or for such ads that have slipped or continue to slip through Google’s system for blocking the appearance of GEICO’s mark in Sponsored Links. For, despite the flaws in the survey, the extremely high percentages of respondents who experienced some degree of confusion when viewing such ads provides sufficient evidence to survive defendant’s Motion for Judgment. Further, having been advised by defendant that it has no evidence to introduce on this last issue, the Court finds that plaintiff has established a likelihood of confusion, and therefore a violation of the Lanham Act, solely with regard to those Sponsored Links that use GEICO’s trademarks in their headings or text.
 

Aware of the importance of these issues to the ongoing evolution of Internet business practices and to the application of traditional trademark principles to this new medium, the Court emphasizes that its ruling applies only to the specific facts of this case, which include the unique business model employed by plaintiff and the specific design of defendant’s advertising program and search results pages. In addition, the Court has not addressed several remaining legal issues, including whether Google itself is liable for the Lanham Act violations resulting from advertisers’ use of GEICO’s trademarks in the headings and text of their Sponsored Links, as accomplished through Google’s Adwords program. That significant issue remains to be resolved, either through an agreement by the parties or in a continuation of the trial. Also unresolved is the timeframe during which violations occurred and the measure of damages or other relief to which plaintiff may be entitled if Google were to be found liable.
 

III. Conclusion

Accordingly, for the reasons stated in open court, as amplified by this Memorandum Opinion, defendant’s Motion for Judgment pursuant to Fed. R. Civ. P. 52(c) has been granted in part and denied in part. The Court has temporarily stayed this civil action to give the parties time to consider these rulings and determine whether they can resolve the remaining issues of liability and damages. An appropriate Order will issue. The Clerk is directed to forward copies of this Memorandum Opinion to counsel of record.

Entered this 8th day of August, 2005.

Leonie M. Brinkema
United States District Judge
 

Alexandria, Virginia

1 The original Complaint was filed on May 4, 2004, against Google and Overture Services, Inc. (“Overture”). A First Amended Complaint was filed on May 14, 2004, and on November 29, 2004, Overture was dismissed as a defendant. Therefore, all references to the Complaint in this Opinion are to the First Amended Complaint, and only Google remains as a defendant.
In a previous Memorandum Opinion discussed more fully below, the Court dismissed Google’s Motion to Dismiss in part, finding that Google did “use” plaintiff’s trademark for purposes of the Lanham Act, although not necessarily improperly. The Court granted defendant’s Motion To Dismiss state law counts of tortious interference with prospective economic advantage and statutory civil business conspiracy and denied the Motion to the extent that it sought a ruling that Google did not “use” plaintiff’s trademark. See Gov’t Employees Ins. Co. v. Google, Inc., 330 F. Supp. 2d 700 (E.D. Va. 2004).

Later, the Court denied defendant’s Motion for Summary Judgment, which argued that Google’s use of GEICO’s marks does not create a likelihood of confusion, because such a highly factual question is inappropriate for resolution on summary judgment. See, e.g., National Fed’n of the Blind, Inc. v. Loompanics Enters., Inc., 936 F. Supp. 1232, 1241 (D. Md. 1996)(citing Country Floors, Inc. v. P’ship of Gepner & Ford, 930 F.2d 1056, 1062-63 (3d Cir. 1991)).

2 The Complaint alleges, and defendant does not dispute, that plaintiff has obtained federal trademark registration for “GEICO” and “GEICO DIRECT.”

3 Those advertisements appeared before Google began barring such use of GEICO’s trademark and may have continued to slip through Google’s screening system since the ban took effect.

4 Whether the advertisers themselves violated the Lanham Act is not before the Court because the advertisers are not parties to this action.

5 Conversely, Google has always allowed trademark holders to  request that their trademarks not appear in Sponsored Links’ headings or text. Thus, one point of dispute regarding Google’s potential damages exposure, should GEICO prevail on the remaining liability issue, concerns when GEICO first effectively requested that its marks be blocked in this fashion and whether Google responded appropriately.
That dispute involves the effect of a number of communications exchanged between the parties in 2002 and 2003 and the sufficiency of Google’s past and current efforts at policing violations of the ban on advertisers using GEICO’s marks in the headings and text of their ads. Because these issues are relevant to Google’s potential liability for damages, not to whether the use of GEICO’s marks violated the Lanham Act, they will not be addressed further in this Opinion.

6 The Complaint also alleges a common law claim of unfair competition. The elements of trademark infringement and unfair competition under the Lanham Act are identical to the elements of unfair competition under Virginia state law. See Lone Star Steakhouse & Saloon v. Alpha of Va., 43 F.3d 922, 930 n.10 (4th Cir. 1995). Accordingly, this Opinion’s discussion of the Lanham Act claims applies equally to plaintiff's unfair competition claim.

7 The Court found support for this distinction in McCarthy on  Trademarks and Unfair Competition: “[W]here keyword placement of . . . advertising is being sold, the portals and search engines are taking advantage of the drawing power and goodwill of these famous marks. The question is whether this activity is fair competition or whether it is a form of unfair free riding on the fame of well-known marks.” GEICO, 330 F. Supp. 2d at 704 (quoting J. Thomas McCarthy, McCarthy on Trademarks & Unfair Competition § 25:70.1 (2004)).

8 The Fourth Circuit has noted that “survey evidence clearly favors the defendant when it demonstrates a level of confusion much below ten percent” but that a number of courts have found actual confusion based on survey evidence indicating a ten to twelve percent rate of confusion. See Sara Lee, 81 F.3d at 467 n.15.

9 Plaintiff tested confusion with regard to the “GEICO” mark but advances the same arguments with regard to “GEICO DIRECT.” This survey represents GEICO’s only evidence of likelihood of confusion. Therefore, GEICO’s ability to survive defendant’s Motion depends on the strength and reliability of the survey results.

10 The survey was limited to drivers who said that they were likely to consider purchasing or renewing their auto insurance policy within the next six months and that they would use the Internet to search for providers.

11 The professor selected this screen shot from among a number of choices provided to him by plaintiff. The shot captured the results of a search run on “GEICO” in April 2004, which was before Google began barring advertisers from using GEICO’s marks in their ad headings and text. Plaintiff contended at trial that the use of an earlier shot was necessary to avoid having references to this action show up in the organic listings, a result that could have prejudiced participants’ responses.
Defendant objected to the choice of the particular screen, arguing that it skewed the survey results by showing five Sponsored Links, when the average results page includes fewer advertisements.

12 The questions asked of the participants were as follows:
1(a) If you wanted to purchase automobile insurance from GEICO, where on this page would you “click” first?
1(b) Why do you say that?
2(a) Now if you clicked on GEICO (interviewer pointed to the first of the Sponsored Links, which mentioned GEICO), what company’s or companies’ Web site would you expect to go to?
2(b) Why do you say that?
3(a) Do you think the company that sponsors this listing is associated or connected with any other company or companies? (If participants responded affirmatively, they were asked the following two questions.)
3(b) What company or companies are they associated or connected with?
3(c) Why do you say that? (After answering this question, participants were informed that the survey was complete.

13 These figures are based on an average of the confusion registered with regard to each of the five Sponsored Links that appeared on the Web page.
14 Defendant and the Court noted that many of those respondents explained that they would click first on one of the Sponsored Links to purchase insurance to save money or to get price quotes. These explanations belie the conclusion that they were confused about GEICO.

15 No respondent in the control group expressed confusion  when asked whether clicking on one of the NIKE Sponsored Links would direct him to GEICO’s Web site. The professor thus concluded that respondents’ confusion in the test group resulted from “the positioning, heading and/or content” in the insurancerelated ads. (Expert Report at 8). This absence of confusion when faced with NIKE ads alongside organic listings for “GEICO” also refutes the allegation that the use of the trademark as a keyword, without more, causes a likelihood of confusion.

16 Further complicating the analysis, certain Web browsers may significantly change the look of the pages seen by real users. For example, Internet Explorer removes shading and dividers.