Full opinion text
HARTZ, Circuit Judge. The Lanham Act, 15 U.S.C. §§ 1051-1127, prohibits the infringement of trademarks (used to identify products) and service marks (used to identify services). It was enacted in 1946, but because it speaks in general terms it can be applied to technologies unimagined at the time of enactment. One such technology, the Internet, has created a number of challenging issues. The case before us concerns Internet search engines, which present advertisers with new means of targeting prospective customers and therefore new possibilities for claims under the Lanham Act. The dispute arises out of advertising through AdWords, a program offered by the Internet search engine Google. An advertiser using AdWords pays Google to feature one of its ads onscreen whenever a designated term, known as a keyword, is used in a Google search. We must resolve whether the Lanham Act was violated by an advertiser’s use of keywords that resembled a competitor’s service mark. For the most part, we hold that there was no violation. Plaintiff 1-800 Contacts, Inc. (1-800) dominates the retail market for replacement contact lenses. It owns the federally registered service mark 1800CONTACTS. Defendant Lens.com, Inc. is one of 1-800’s competitors. To police the use of its mark, 1-800 enters different variations of the mark into Google searches and monitors what search results are displayed. When 1-800 found that several searches generated paid ads for Lens.com’s websites, it concluded that Lens.com had reserved the mark as a keyword. After attempting to resolve the situation informally, 1-800 sued Lens.com for service-mark infringement. Its primary claim was that Lens.com itself had infringed the 1800CONTACTS mark by purchasing keywords resembling the mark. According to 1-800, this conduct had directed potential customers for 1-800 to Lens.com by creating what is known as “initial-interest confusion,” which can be actionable under the Lanham Act. As the case progressed, 1-800 supplemented its claim of direct infringement by alleging that certain third-party marketers hired by Lens.com, known as affiliates, had also purchased keywords resembling the mark and that at least one affiliate was using the mark in the text of its online ads. 1-800 sought to hold Lens.com secondarily liable for its affiliates’ conduct. The theories of secondary liability, which will be discussed more fully below, were common-law agency and contributory infringement. The district court awarded summary judgment to Lens.com on all claims. On the direct-liability claim and most of the secondary-liability claims, the court ruled that 1-800 had raised no genuine issue of fact regarding the likelihood of initial-interest confusion. On the remaining secondary-liability claims — which concerned the use of 1-800’s mark in the content of ads displayed on Google’s site — the court ruled that 1-800’s evidence was insufficient to hold Lens.com liable for any misconduct of its affiliates. 1-800 appeals the summary judgment. To the extent that the court based summary judgment on the ground that no likelihood of confusion existed, we affirm. Traditional analysis and actual marketplace data reveal that the keyword use by Lens.com and its affiliates was highly unlikely to divert consumers. As for the remaining secondary-liability claims, we affirm the denial of liability under agency law because the affiliates, even if agents (or more precisely, subagents) of Lens, com, lacked authority to include 1-800’s mark in ads for Lens.com. But we reverse the denial of liability for contributory infringement because the evidence could support a reasonable finding that Lens, com did not take reasonable steps to halt the display of 1-800’s marks in affiliate ads once it learned of such display. Also, we affirm the discovery sanction challenged by Lens.com on cross-appeal (but decline to award 1-800 its attorney fees for defending the sanction in this court), and we affirm the denial of Lens, corn’s district-court motion for attorney fees. I. BACKGROUND A. The Dispute 1-800 is the world’s leading retailer of replacement contact lenses. It sells lenses via telephone,- by mail order, and over the Internet. In 2003 it registered with the federal trademark register the nonstylized word mark “1800CONTACTS” as one of its service marks. Aplt.App., Vol. 6 at 1001. The mark achieved incontestable status under 15 U.S.C. § 1065 in 2008. Lens.com is one of 1-800’s competitors in the replacement-lens retail market. Unlike 1-800, which advertises through several different media and which derived approximately 40% of its gross sales from sources other than Internet orders in 2007, Lens.com advertises and does business almost exclusively online. This dispute arose in the summer of 2005, when 1-800 discovered that paid advertisements for Lens.com appeared when one searched for the phrase “1800 CONTACTS” on Google. Id., Vol. 11 at 2654. 1-800 concluded that Lens.com was using the 1800CONTACTS mark in its online marketing. To explain this concern properly, we must first review some mechanics of Internet advertising through search engines. Because 1-800’s arguments on appeal focus solely on Lens.com’s use of Ad-Words, a program offered by Google, we describe only AdWords and no other search engines or advertising services. At the time of the proceedings below, a typical Google search simultaneously yielded two different kinds of results: organic results and sponsored links. Organic results were the links generated by Google’s search algorithms, which sorted web pages according to their relevance to the user’s search as well as their quality. An advertiser could not pay Google to have its web page displayed among the organic results. Through AdWords, however, an advertiser could pay to be displayed as a sponsored link. A sponsored link would include advertising copy and the advertiser’s website address. A user who clicked on the ad would be connected to the website. Sponsored links usually appeared either above or to the right of the organic results. The notice “Sponsored Links” was displayed next to each cluster of ads. Google placed background shading behind several of the sponsored links to set them apart visually from the organic results, which appeared on a plain white background. For its ad to appear as a sponsored link when a user initiated a Google search, an advertiser had to bid to reserve a particular word or phrase — known as a keyword — that would trigger the display of its ad. The advertiser specified whether its ad should appear as the result of (1) a broad match — that is, whenever a Google search contained a phrase that was either similar to or a relevant variation of the keyword; (2) a phrase match — whenever the search contained the exact keyword; or (3) an exact match — whenever the search contained the exact keyword and nothing more. The advertiser could also use negative matching, which instructed Google not to display the ad when a certain search term was used. Negative matching allowed the advertiser to filter out irrelevant searches. For example, if a seller of contact lenses had purchased the keyword contacts, it might have wanted to exclude searches for marketing contacts. The display of a sponsored link in response to a user’s search was known as an impression. An advertiser paid Google only if the user actually clicked on its impression; its bid for the keyword represented the amount per click that it was willing to pay. Advertisers who bid higher amounts generally received superior placement among the sponsored links. A click that led to a sale through the advertiser’s web page was called a conversion, which did not incur an additional charge to the advertiser from Google. 1-800 apparently reasoned that a Google search for “1800 CONTACTS” could generate an ad for Lens.com only if Lens, com — or someone working on its behalf— had bid on that exact term or on some phrase containing that exact term. In September 2005 it sent Lens.com two letters reporting that online searches for that term were resulting in ads for Lens.com. One of the letters was accompanied by screenshots that showed Google search results for the phrases “1-800 contacts,” “1-800-contacts,” and “1800contacts.” Id. at 2657-59. In each screenshot an ad for Lens.com appeared among the sponsored links, along with ads for 1-800 and other retailers. Lens.com responded that it had looked into the matter, had determined who appeared to be responsible, and would advise them not to bid on “1-800-CON-TACTS” as a keyword in the future. Id. at 2663. The parties who appeared to be responsible, Lens.com told 1-800, were affiliates. Advertisers like Lens.com might pay third-party affiliates to publish ads for them through AdWords and other search-engine programs. An Internet user who clicked on an ad published by a Lens.com affiliate would be routed directly to one of Lens, corn’s four websites — www.Lens.com, www.JustLenses.com, www.l-800GetLens. com, and www.ContactsAmerica.com — or instead would be taken to the affiliate’s own website, where links to Lens.com’s websites were displayed. When the user made a purchase at one of Lens.com’s websites as a result of clicking on the affiliate’s ad, the affiliate earned a commission. Lens.com did not recruit individual affiliates directly; rather, it worked with Commission Junction (CJ), which managed a network of affiliates. Under the arrangement in this case, CJ agreed to pay the commissions to the affiliates for their conversions, and Lens.com agreed to reimburse CJ. According to Lens.com’s chief executive officer, Lens.com had four different accounts with CJ in 2009, and through those accounts more than 10,000 affiliates were signed up to promote Lens.com and its brands. Whatever action Lens.com took in response to 1-800’s September 2005 notices, 1-800 continued to express concerns. In November and December 2005 it again contacted Lens.com and advised that Google searches for “1800contacts,” “1800 contacts,” “1-800-contacts,” and “1-800 contacts” were still generating Lens.com’s ads. Id. at 2695-98. Lens.com replied that it would try to determine who was publishing the ads in question. The next relevant communication did not occur until April 2007, when 1-800’s counsel emailed Lens.com’s counsel once more to complain that the problem was recurring. Attached to the email were screenshots of search results from Google and another search engine. Lens.com’s counsel replied that he would confer with his client to see whether the problem could be fixed. 1-800 filed a complaint against Lens.com in August 2007 in the United States District Court for the District of Utah. The complaint stated that 1-800 had “discovered that Lens.com had purchased sponsored advertisements from Google, and other search engines, for Plaintiffs Marks to trigger advertising and/or a link to the Lens.com Websites.” Id., Vol. 1 at 42. It further alleged that Lens.com had “use[d] the 1800 CONTACTS trademark as a triggering keyword to display and promote Lens.com’s directly competitive goods and services.” Id at 43. To support this allegation, the complaint included a screenshot of Google search results for the term “1800 CONTACTS” in which an ad for Lens.com was featured. Id. The complaint also alleged that Lens, com had used the 1800CONTACTS mark in its advertising copy, and it included a second screenshot that, unlike any of the screenshots that it had previously disclosed to Lens.com, showed a sponsored link featuring the term “1-800 Contacts” in the ad’s text. Id. at 44. The Internet address beneath this text was www.Just Lenses.com, one of Lens.com’s websites. 1-800’s chief legal claims were that Lens.com had infringed on its 1800CON-TACTS mark under § 32 of the Lanham Act, 15 U.S.C. § 1114(1), which provides a cause of action for the infringement of a federally registered mark, and § 43(a), 15 U.S.C. § 1125(a), which provides a cause of action for the infringement of unregistered as well as registered marks. As discovery proceeded, 1-800 learned that Lens.com itself had bid on the following nine terms (the Challenged Keywords) as AdWords keywords: “1-800 contact lenses”; “1800 contact lenses”; “800 contact lenses”; “800comtacts.com”; “800con-tacta.com”; “800contavts.com”; “800con-taxts.com”; “800contzcts.com”; and “800conyacts.com.” ApltApp., Vol. 9 at 1922-23. Lens.com does not dispute that it bid on the Challenged Keywords, nor does 1-800 contend on appeal that Lens, com ever bid on the 1800CONTACTS mark itself. Additionally, 1-800 does not claim that any impressions created by Lens.com featured the 1800CONTACTS mark in their text. Discovery revealed, however, that two Lens.com affiliates, Dusty Goggans and Ryan McCoy, had bid on the keyword “1800Contacts” and close variations of 1-800’s mark. Id, Vol. 5 at 507. And McCoy had published at least one ad for www.JustLenses.com (one of Lens.com’s websites) that featured the phrase “1800 Contacts” in its advertising copy. Id. at 508. In light of this discovery, 1-800 amended its complaint to convey two theories of how Lens.com had violated §§ 32 and 43(a) of the Lanham Act: (1) that Lens.com had directly infringed on the 1800CONTACTS mark by purchasing the Challenged Keywords; and (2) that Lens, corn’s affiliates had infringed on the mark and that Lens.com was secondarily liable for their infringement. It advanced two separate grounds for secondary liability. The first — vicarious infringement — imposes liability on a principal for the infringing acts of its agent. The second — contributory infringement — is analogous to aiding and abetting. Before discussing the district court’s rulings on 1-800’s claims of direct and secondary liability, we briefly review some fundamentals of service-mark infringement under federal law. B. Service-Mark Infringement Under the Lanham Act A service mark, similar to a trademark, is defined by the Lanham Act as “any word, name, symbol, or device, or any combination thereof’ that is used “to identify and distinguish the services of one person, including a unique service, from the services of others and to indicate the source of the services, even if that source is unknown.” 15 U.S.C. § 1127 (2006). 1800CONTACTS is such a mark. The Lanham Act’s private causes of action for trademark infringement are available to the owners of service marks. See Vail Assocs., Inc. v. Vend-Tel-Co., Ltd., 516 F.3d 853, 857 & n. 1 (10th Cir.2008); Donchez v. Coors Brewing Co., 392 F.3d 1211, 1215 (10th Cir.2004). Section 32 of the Act allows the owner of a registered mark to bring an action for infringement against any person who use[s] in commerce any reproduction, counterfeit, copy, or colorable imitation of [the] registered mark in connection with the sale, offering for sale, distribution, or advertising of any goods or services on or in connection with which such use is likely to cause confusion, or to cause mistake, or to deceive.... 15 U.S.C. § 1114(l)(a). Similarly, under § 43(a) the owner of any valid mark, registered or not, may sue any person who uses in commerce any word, term, name, symbol, or device, or any combination thereof, or any false designation of origin, false or misleading description of fact, or false or misleading representation of fact, which ... is likely to cause confusion, or to cause mistake, or to deceive as to the affiliation, connection, or association of such person with another person, or as to the origin, sponsorship, or approval of his or her goods, services, or commercial activities by another person.... Id. § 1125(a). The elements of an infringement claim under § 43(a) are (1) that the plaintiff has a protectable interest in the mark; (2) that the defendant has used “an identical or similar mark” in commerce, Donchez, 392 F.3d at 1215 (brackets and internal quotation marks omitted); and (3) that the defendant’s use is likely to confuse consumers. See Utah Lighthouse Ministry v. Found. for Apologetic Info. & Research, 527 F.3d 1045, 1050 (10th Cir.2008). An infringement claim under § 32 has nearly identical elements, see Jordache Enters., Inc. v. Hogg Wyld, Ltd., 828 F.2d 1482, 1484 (10th Cir.1987), except that the registration of a mark serves as prima facie evidence of both the mark’s validity and the registrant’s exclusive right to use it in commerce, see 15 U.S.C. § 1115(a) (2002). The central question in a typical infringement action under either § 32 or § 43(a) is whether the defendant’s use of the plaintiffs mark is likely to cause consumer confusion. Confusion can be of several sorts. For example, consumers may experience direct confusion of source when they develop the mistaken belief that the plaintiff is the origin of the defendant’s goods or services — so that the defendant capitalizes on the plaintiffs good name. See Australian Gold, Inc. v. Hatfield, 436 F.3d 1228, 1238 (10th Cir.2006). The classic case of direct confusion occurs when “[c]ustomers want to buy the [plaintiffs] product and because of the similarity of the marks, mistakenly buy the [defendant’s] product instead.” 4 J. Thomas McCarthy, McCarthy on Trademarks and Unfair Competition § 23:10 at 23-70 (4th ed.2013) (4 McCarthy). Or consumers may experience reverse confusion of source when they mistakenly believe that the defendant is the origin of the plaintiffs goods or services. See Australian Gold, 436 F.3d at 1238. Reverse confusion typically occurs “when the [defendant’s] advertising and promotion so swamps the [plaintiffs] reputation in the market” that “customers purchase the [plaintiffs] goods under the mistaken impression that they are getting the goods of the [defendant].” 4 McCarthy § 23:10 at 23-70 to 71. In that circumstance the defendant would not be trying to take a free ride on the plaintiffs reputation but would drown out the value of the plaintiffs mark. This can arise when a national firm adopts a mark that was already being used by a small business operating in only one locality. See id. at 23-71 to 75 (setting forth examples). Confusion need not be limited to the incorrect perception that one party was the source of the other party’s product or service; it may also arise from “a mistaken belief in common sponsorship or affiliation.” Amoco Oil Co. v. Rainbow Snow, 748 F.2d 556, 558 (10th Cir.1984). Nor must the confusion occur at the point of sale; postsale confusion may propagate among potential consumers who see the relevant product after the original buyer has purchased it. See Gen. Motors Corp. v. Urban Gorilla, LLC, 500 F.3d 1222, 1227 (10th Cir.2007). The type of confusion alleged by 1-800 is an additional variety — namely, initial-interest confusion, a distinct theory that we recognized in Australian Gold. Initial-interest confusion “results when a consumer seeks a particular trademark holder’s product and instead is lured to the product of a competitor by the competitor’s use of the same or a similar mark.” Australian Gold, 436 F.3d at 1238. As the name implies, the improper confusion occurs even if the consumer becomes aware of the defendant’s actual identity before purchasing the product. See id. at 1238-39. In Australian Gold the defendants (1) used Australian Gold’s trademarks on their own websites; (2) placed Australian Gold’s marks in hidden codes associated with their websites (metatags), so that an Internet search for those trademarks would return links to the defendants; and (3) paid a website to list the defendants in a preferred position whenever a user searched for Australian Gold’s marks. See id. at 1233 n. 3, 1239. We affirmed the denial of the defendants’ motion for judgment as a matter of law because we agreed with the district court that a genuine issue of fact existed regarding the likelihood of initial-interest confusion. See id. at 1240. We have identified six factors (the King of the Mountain factors) as relevant to whether a likelihood of confusion exists: (a) the degree of similarity between the marks; (b) the intent of the alleged infringer in adopting its mark; (c) evidence of actual confusion; (d) the relation in use and the manner of marketing between the goods or services marketed by the competing parties; (e) the degree of care likely to be- exercised by purchasers; and (f) the strength or weakness of the marks. King of the Mountain Sports, Inc. v. Chrysler Corp., 185 F.3d 1084, 1089-90 (10th Cir.1999). These factors are not exhaustive. See id. at 1090. And they should not be applied mechanically; some factors may carry far more weight than others depending on the circumstances. See id. (“[T]he weight afforded to some of the factors differs when applied in ... separate contexts.”); cf. Vail Assocs., 516 F.3d at 863-66 (treating actual confusion as the most important factor); Universal Money Ctrs., Inc. v. AT & T Co., 22 F.3d 1527, 1536 (10th Cir.1994) (indicating that the lack of actual confusion and the dissimilarity of the marks were the paramount considerations). A defendant may be held liable for service-mark infringement even though it has not directly infringed on the plaintiffs mark through its own acts. Two theories of secondary liability are pertinent here. First, we have joined the Third Circuit in recognizing that the Lanham Act incorporates common-law agency principles: a principal may be held vicariously liable for the infringing acts of an agent. See Procter & Gamble Co. v. Haugen, 317 F.3d 1121, 1127-28 (10th Cir.2003); AT & T Co. v. Winback & Conserve Program, Inc., 42 F.3d 1421, 1433-34 (3d Cir.1994); 4 McCarthy § 25:21.25. Second, in Inwood Laboratories, Inc. v. Ives Laboratories, Inc., 456 U.S. 844, 853-54, 102 S.Ct. 2182, 72 L.Ed.2d 606 (1982), the Supreme Court ruled that contributory infringement can violate the Lanham Act. Akin to aiding and abetting, contributory infringement generally consists of either intentionally causing or knowingly facilitating the infringement of the plaintiffs mark by a third party. The Inwood Court formulated the theory as follows: [Liability for trademark infringement can extend beyond those who actually mislabel goods with the mark of another. Even if a manufacturer does not directly control others in the chain of distribution, it can be held responsible for their infringing activities under certain circumstances. Thus, if a manufacturer or distributor intentionally induces another to infringe a trademark, or if it continues to supply its product to one whom it knows or has reason to know is engaging in trademark infringement, the manufacturer or distributor is contributorially responsible for any harm done as a result of the deceit. Id. at 853-54, 102 S.Ct. 2182 (footnote omitted). C. Proceedings Before the District Court 1-800 moved for partial summary judgment on the issues of direct and secondary liability for service-mark infringement. Except for the few ads that used the mark in their text, 1-800’s only clearly expressed theory of infringement was initial-interest confusion. Although it asserts on appeal that Lens.com’s acts of direct infringement included purchasing merely generic keywords and then failing to designate the 1800CONTACTS mark as a negative keyword, that theory was not raised in district court. Its brief in opposition to summary judgment disclaimed such a position, stating: “On the Internet, a competitor is free to purchase keywords of the product category (i.e., contact lenses) but the competitor is not free to purchase a competitor’s trademark as a keyword.” Aplt.App., Vol. 10 at 2281. The brief made plain that its direct-infringement claim was limited to the nine Challenged Keywords, and that its principal secondary-infringement claim was limited to the purchase by Goggans and McCoy of keywords that closely resembled its mark. In an effort to show actual confusion (the third King of the Mountain factor), 1-800 offered an example of one confused consumer and the results of a consumer survey conducted by its expert, Carl Degen. Lens.com moved to strike the survey as unreliable. It also moved for summary judgment on all claims. The district court granted Lens.com’s motion to strike the survey. See id., Vol. 3 at 5456 (Memorandum Decision & Order on Carl Degen Evidence at 1, 1-800 Contacts, Inc. v. Lens.com, Inc., Case No. 2:07-cv-591 CW (D. Utah Dec. 15, 2010)) (Survey Order). And it awarded summary judgment to Lens.com. See 1-800 Contacts, Inc. v. Lens.com, Inc., 755 F.Supp.2d 1151, 1191 (D.Utah 2010). In granting summary judgment the court first ruled that a defendant’s purchase of search-engine keywords — in Lens, corn’s case, the nine Challenged Keywords; and in Goggans and McCoy’s case, the term “1800Contacts” and slight variations thereof — can amount to a use in commerce under the Lanham Act. See id. at 1169-70. But it also ruled that merely purchasing such a keyword cannot, on its own, give rise to liability for infringement. See id. at 1171-74. Observing that an impression for Lens.com might result just as easily from a Google search for an unprotected, generic term like “contacts” as from a search for the registered mark 1800CON-TACTS, the court explained that a Google user confronted with a screenshot of search results would be unable to tell from those results alone which keyword had been purchased. See id. at 1173. It reasoned that as a matter of law, a defendant’s purchase of a. search-engine keyword cannot, by itself, create the likelihood of confusion that is necessary for infringement liability; rather, the court ruled that keyword use can generate a likelihood of confusion only in combination with the specific language of the resulting impressions. See id. at 1173-74. It thus ruled that insofar as the keyword use of Lens.com and its affiliates generated ads that did not feature 1-800’s mark or any variation in their text, no likelihood of confusion existed. See id. at 1181-82. Turning to the few ads (all placed by one affiliate, McCoy) whose text did feature some variation of the mark, the district court disposed of 1-800’s secondary-liability claims by rejecting its theories of vicarious infringement and contributory infringement. With respect to vicarious infringement it ruled that the evidence did not support a principal-agent relationship between Lens.com and any of its affiliates. See id. at 1182-84. And with respect to contributory infringement it ruled that 1-800 had failed to provide any evidence from which a reasonable jury could find that Lens.com either (1) intentionally induced its affiliates to use the 1800CON-TACTS mark in the text of their impressions or (2) knew or had reason to know that they were doing so yet failed to take appropriate action. See id. at 1185-87. On appeal 1-80Q argues (1) that there were disputed facts regarding likelihood of confusion and (2) that the evidence would support findings of secondary liability under theories of both vicarious liability and contributory infringement. In turn, Lens, com cross-appeals from an order sanctioning it for discovery abuses. And in a separate appeal Lens.com challenges the district court’s denial of its motion for attorney fees under both the Lanham Act, see 15 U.S.C. § 1117(a) (2008), and Utah law, see Utah Code Ann. § 78B-5-825 (2008). We affirm on all issues but one: contributory infringement. We disagree with the district court’s ruling that there was insufficient evidence that Lens.com had the necessary actual or constructive knowledge to be held contributorially liable for the conduct of its affiliates. We also reject Lens.com’s unclean-hands defense to 1-800’s claims. Therefore, we reverse and remand for further proceedings on the contributory-infringement claim. II. DISCUSSION We first resolve the issues presented by 1-800’s appeal, We then address Lens, corn’s cross-appeal and its appeal of the denial of attorney fees. A. Direct Liability for Ads Placed by Lens.com The district court awarded summary judgment to Lens.com on 1-800’s claim that Lens.com was directly liable for infringing on its service mark. It ruled that 1-800 had created no genuine factual issue regarding whether Lens.com’s keyword use was likely to cause confusion. See 1-800 Contacts, 755 F.Supp.2d at 1181-82. 1-800 asserts that this ruling was error. It argues generally about likelihood of confusion, not distinguishing its § 32 infringement claims from its § 43(a) claims. We, too, need not differentiate between the two provisions, as the tests for likelihood of confusion under § 32 and § 43(a) do not differ materially. See Jordache, 828 F.2d at 1484. We review the district court’s grant of summary judgment de novo. See Sally Beauty Co., Inc. v. Beautyco, Inc., 304 F.3d 964, 971 (10th Cir.2002). “Summary judgment is appropriate if the pleadings, depositions, other discovery materials, and affidavits demonstrate the absence of a genuine issue of material fact and that the moving party is entitled to judgment as a matter of law.” Id. “An issue is genuine if the evidence is such that a reasonable jury could return a verdict for the nonmoving party.” Id. at 972 (internal quotation marks omitted). Although the party moving for summary judgment “bears the initial burden of demonstrating an absence of a genuine issue of material fact,” it can satisfy that burden with respect to an issue on which it does not bear the burden of persuasion at trial “simply by indicating to the court a lack of evidence for the nonmovant on an essential element of the nonmovant’s claim.” Id. at 971. Once the moving party has done so, “the burden shifts to the nonmoving party to go beyond the pleadings and set forth specific facts showing that there is a genuine issue for trial.” Id. As the party alleging service-mark infringement, 1-800 has the burden of proving likelihood of confusion. See John Allan Co. v. Craig Allen Co. L.L.C., 540 F.3d 1133, 1138 (10th Cir.2008). Likelihood of confusion is ordinarily a question of fact for the jury, but summary judgment is appropriate if no reasonable juror could find that such a likelihood exists. See Sally Beauty Co., 304 F.3d at 972; cf. King of the Mountain, 185 F.3d at 1089 (“Courts retain an important authority to monitor the outer limits of substantial similarity within which a jury is permitted to make the factual determination whether. there is a likelihood of confusion.” (brackets and internal quotation marks omitted)). Again, the elements of an infringement claim under the Lanhanr Act are (1) that the plaintiff has a protectable interest in the mark, (2) that the defendant has used an identical or similar mark in commerce, and (3) that the defendant’s use is likely to confuse consumers. That 1-800 has a protectable interest in its mark is not in dispute. And the district court ruled that purchasing the Challenged Keywords satisfied the use-in-commerce requirement, see 1-800 Contacts, 755 F.Supp.2d at 1169-70, a premise that we will assume without deciding. Thus, the only contested issue is likelihood of confusion. 1-800’s theory of confusion is initial-interest confusion. Its essential contention is that although Lens, com never published any ads with 1-800’s mark in their text, its bidding on the nine Challenged Keywords caused its ads to appear in response. to searches for the mark, thereby diverting customer interest away from 1-800’s website and toward Lens.com’s websites. The district court ruled that use of the Challenged Keywords, divorced from the text of the resulting ads, could not result in a likelihood of confusion. It pointed out that because Google users view only the results of their searches and cannot tell exactly which keywords an advertiser has purchased, a user who searches for “1-800 Contacts” and then sees an ad published by Lens.com has no way of knowing whether Lens.com has reserved 1-800’s mark as a keyword or instead has reserved simply the term contacts. See id. at 1173. “Given that fact,” the court reasoned, “it would be anomalous to hold a competitor liable simply because it purchased a trademarked keyword when the advertisement generated by the keyword is the exact same from a consumer’s perspective as one generated by a generic keyword.” Id. at 1174. This argument has some attraction, although if confusion does indeed arise, the advertiser’s choice of keyword may make a difference to the infringement analysis even if the consumer cannot discern that choice. Cf. Holiday Inns, Inc. v. 800 Reservation, Inc., 86 F.3d 619, 625-26 (6th Cir.1996) (defendants did not use plaintiffs mark in commerce when they reserved a common misdialing of the plaintiffs telephone number, thereby merely exploiting preexisting confusion rather than creating it). In any event, we need not resolve the matter because 1-800’s direct-infringement claim fails for lack of adequate, evidence of initial-interest confusion. We have already set forth a list of six helpful, nonexhaustive factors for determining likelihood of confusion: (1) similarity of the marks, (2) intent of the alleged infringer, (3) evidence of actual confusion, (4) similarity of the competing parties’ services and manner of marketing, (5) degree of consumer care, and (6) strength of the marks. See King of the Mountain, 185 F.3d at 1089-90. As we and other courts have emphasized, however, other factors may be considered, and the weight of any given factor can depend very much on context. See Team Tires Plus, Ltd. v. Tires Plus, Ltd., 394 F.3d 831, 833 (10th Cir.2005) (“As with so many of our multi-factor tests, we have emphasized that this list of factors is not exhaustive, that no single factor is dispositive, and that all factors must be considered as an interrelated whole.”); King of the Mountain, 185 F.3d at 1090 (noting that the similarity of the marks constituted “the heart of our analysis” in a eonfusionof-sponsorship case, whereas the parties’ manner of marketing and the degree of consumer care had “little importance”); Nabisco, Inc. v. Warner-Lambert Co., 220 F.3d 43, 48 (2d Cir.2000) (dissimilarity of the marks outweighed all other factors); Rosetta Stone Ltd. v. Google, Inc., 676 F.3d 144, 153-154 (4th Cir.2012) (in some contexts “the application of the traditional multi-factor test is difficult because often many of the factors are either unworkable or not suited or helpful as indicators of confusion” (internal quotation marks omitted)); CareFirst of Md., Inc. v. First Care, P.C., 434 F.3d 263, 267-68 (4th Cir.2006) (ruling that actual confusion was the paramount factor); Levi Strauss & Co. v. Blue Bell, Inc., 778 F.2d 1352, 1360 (9th Cir.1985) (although “certain aspects of the multifactor test describe the circumstances to which a trier of fact would refer in making an educated guess as to what was going on in the minds of consumers,” in many cases direct evidence may “outweigh whatever circumstantial evidence has been introduced”); Kellogg Co. v. Pack’em Enters., Inc., 951 F.2d 330, 333 (Fed.Cir.1991) (“We know of no reason why, in a particular case, a single ... factor may not be dispositive.”). The upshot of these cases is that when certain facts are more probative than others on the ultimate issue of likelihood of confusion, those facts may dominate the analysis. In this case, one item of evidence particularly suggests an absence of initial-interest confusion, the variety of consumer confusion on which 1-800 relies. As we explained in Australian Gold, initial-interest confusion occurs when a consumer in search of the plaintiffs product “is lured to the product of a competitor.” 436 F.3d at 1238 (emphasis added); see Vail Assocs., 516 F.3d at 872 (“Initial interest confusion is a ‘bait and switch’ tactic that permits a competitor to lure consumers away from a service provider by passing off services as those of the provider, notwithstanding that the confusion is dispelled by the time of sale.” (emphasis added)). Applying that description to this case, initial-interest confusion would arise as follows: a consumer enters a query for “1-800 Contacts” on Google; sees a screen with an ad for Lens.com that is generated because of Lens.com’s purchase of one of the nine Challenged Keywords; becomes confused about whether Lens, com is the same source as, or is affiliated with, 1-800; and therefore clicks on the Lens.com ad to view the site. Lens.com has exploited its use of 1-800’s mark to lure the confused consumer to its website. Ordinarily, the likelihood of such luring would need to be estimated by what we can call “informed judgment,” which is assisted by analyzing the six King of the Mountain factors. Here, however, we have AdWords data setting an upper limit on how often consumers really were lured in such fashion. A report by Lens.com’s expert explained that Lens.com’s use of the nine Challenged Keywords yielded 1,626 impressions for Lens.com or its associated websites over eight months. In only 25 (1.5%) of these 1,626 instances did the user click on the ad for Lens.com. (We do not know how many of the 25 made a purchase from Lens.com.) The users in those 25 instances may have been confused into thinking that Lens.com was affiliated with 1-800, or they may simply have wished to look at the offerings of those whom they knew to be 1-800’s competitors. What we can say, though, is that initial-interest confusion occurred at most 1.5% of the time that a Lens.com ad was generated by a Challenged Keyword in those eight months. This number cannot support an inference that Lens.com’s keyword activity was likely to “lure[j” consumers away from 1-800. Australian Gold, 486 F.3d at 1238. It is thus insufficient to justify relief. See Universal Money Ctrs., 22 F.3d at 1534, 1537 (characterizing a 2.6% confusion rate as de minimis); cf. CareFirst, 434 F.3d at 268 (survey reporting a confusion rate of 2% was “hardly a sufficient showing of actual confusion”); Henri’s Food Prods. Co., Inc. v. Kraft, Inc., 717 F.2d 352, 358-59 (7th Cir.1983) (survey reporting a confusion rate of 7.6% weighed against a finding of infringement). Moreover, 1-800’s arguments based on other King of the Mountain factors does not suggest a contrary conclusion. It points to the district court’s determination that the likelihood of confusion is supported by factors (4) and (5): the parties offer the same services in the same channels of trade (retail sales of replacement contact lenses over the Internet) and “it is unlikely that consumers exercise a high degree of care in selecting this service.” 1-800 Contacts, 755 F.Supp.2d at 1177. In addition, it challenges the district court’s determination on factor (6) that 1-800’s mark is “only moderately strong,” id. at 1181; and on factor (1), it argues that the relevant marks were identical or nearly identical, because the consumer was using the 1800CONTACTS mark as a search term and Lens.com had triggered the ad by using a nearly identical mark. This analysis by 1-800 illustrates the danger of applying the factors mechanically without attention to context. The specific issue before us is the likelihood that a consumer who conducts an Internet search for 1-800 Contacts and then sees an ad for Lens.com on the results page will be confused into thinking that Lens.com has a business association with 1-800. To begin with, even if consumers in general may not much care what retailer supplies their contact lenses, the consumers relevant to this suit are looking for a particular retailer. Presumably they have narrowed their search because they have already selected 1-800 as the preferred retailer and are searching for its website or perhaps commentary on its performance. Given the purpose of the search, the shoppers will be attentive to click on those results that will connect them with sites relating to 1-800. In addition, once the consumers see the results page, the substantial dissimilarity between “1-800 Contacts” and “Lens.com” (or its other websites) can be expected to greatly reduce the chance that the consumers will think that the parties are related enterprises; the similarity of the search term and 1-800’s mark is of minor relevance. Perhaps in the abstract, one who searches for a particular business with a strong mark and sees an entry on the results page will naturally infer that the entry is for that business. But that inference is an unnatural one when the entry is clearly labeled as an advertisement and clearly identifies the source, which has a name quite different from the business being searched for. It is for this reason that the Ninth Circuit considered “the labeling and appearance of the advertisements and the surrounding context on the screen displaying the results page” to be a critical factor in finding no likelihood of confusion in a case in which the alleged infringer used a competitor’s mark as a keyword. Network Automation v. Advanced Sys. Concepts, 638 F.3d 1137, 1154 (9th Cir.2011). We conclude that the factors other than evidence of actual confusion (even if we assume that 1-800’s mark is a strong one) firmly support the unlikelihood of confusion. This case is readily distinguishable from Australian Gold, in which the alleged infringer used its competitor’s trademarks on its websites. See 436 F.3d at 1239. We now turn to 1-800’s arguments regarding actual confusion. First, it cites what it claims to be anecdotal evidence of actual confusion in the marketplace: a customer-service record disclosed by Lens, com reported that a customer called Lens, com in July 2006 to cancel her order, apparently because she had just realized that Lens.com was not 1-800. Lens.com counters that the customer-service record cannot be probative of the relevant confusion in this case because, among other reasons, it gives no indication how the customer found Lens.com to place her order initially. We agree. It would be speculation to assume that she had clicked on a Lens.com ad after specifically searching for 1-800. Moreover, a single customer-service record is entitled to little weight. See King of the Mountain, 185 F.3d at 1092 (“[Ijsolated instances of actual confusion may be de minimis.” (brackets and internal quotation marks omitted)); Universal Money Ctrs., 22 F.3d at 1535-36 (characterizing limited evidence of actual confusion as de minimis). 1-800 insists that we must infer that additional, undisclosed customer-service records contained similar evidence of actual confusion because the district court, in sanctioning Lens.com for discovery abuses, forbade Lens.com from characterizing the July 2006 record as de minimis. But this argument misreads the district court’s order. At the magistrate judge’s recommendation, the district court in 20.09 ordered that “Lens.com shall be precluded from relying upon any business records of Lens, com or its contractors/subcontractors that were not produced by December 9, 2008,” and that “Lens.com shall also be precluded from testifying in a manner that characterizes the contents of such documents.” Aplt.App., Vol. 2 at 3463 (Order at 2,1-800 Contacts, No. 2:07-cv-591 CW (D. Utah Feb. 27, 2009)). This order simply forbids Lens.com from testifying as to the meaning of documents that it never produced; it does not require any inference on the court’s part as to the meaning of any documents, and it does not relieve 1-800 of its burden of producing evidence of actual confusion. 1-800 still had to bring evidence of actual confusion to the district court’s attention. 1-800 cites no such evidence on appeal apart from the lone customer-service record from July 2006. Next, 1-800 argues that its consumer-confusion survey was wrongly excluded and that it, too, demonstrated actual confusion. Respondents to this survey were recruited through an online questionnaire and were limited to consumers who said that they either had bought contact lenses in the previous 12 months or were considering buying them in the next 12 months. During the survey they were told to imagine that they had just conducted a Google search for “1800contacts,” and then they viewed screenshots of search results in which an ad for Lens.com appeared among the sponsored links. After studying the screenshots, they were asked whether they thought that the Lens.com ad either “originate[d] from 1-800-CONTACTS,” id, Vol. 12 at 3307, 3315, or “ha[d] sponsorship or approval from 1-800-CONTACTS,” id. at 3308, 3316. The district court excluded the survey results under Fed.R.Evid. 702 on the ground that methodological flaws undermined the survey’s reliability. It focused on two perceived flaws. First, it ruled that the population of respondents was too broad, as it was not limited to prospective Internet consumers of contact lenses. Second, it ruled that the questions were ambiguous and leading. The ambiguity arose from the first question’s failure to clarify whether “1-800-CONTACTS” referred to a search term or a company. And in the court’s view the questions were leading because they suggested the possibility of a connection between Lens.com and 1-800 when the respondents might not have considered such a connection on their own. The court found it unnecessary to address Lens.com’s arguments concerning other alleged flaws because the survey would have been inadmissible regardless. “Surveys can be used as evidence of actual confusion, but their evidentiary value depends on the relevance of the questions asked and the technical adequacy of the survey procedures.” Universal Money Ctrs., 22 F.3d at 1534 n. 3 (internal quotation marks omitted). Although methodological flaws in a confusion survey will typically affect only the survey’s weight and not its admissibility, see Brunswick Corp. v. Spinit Reel Co., 832 F.2d 513, 523 (10th Cir.1987), these flaws may justify exclusion under Rule 702 if they are serious and pervasive enough. See Vail Assocs., 516 F.3d at 864 n. 8. We apply abuse-of-discretion review to the manner in which a district court performs its gatekeeping function under Rule 702, recognizing the latitude that it has in determining whether expert testimony is reliable enough to be admitted. See Bitler v. A.O. Smith Corp., 400 F.3d 1227, 1232 (10th Cir.2004). We, too, are concerned about the reliability of the survey. We note only the ambiguity of a key question. Respondents were told that they had entered “1800con-tacts” into a Google search and were then asked whether they thought that the ad for Lens.com on the results screen “originates from 1-800-CONTACTS.” Aplt. App., Vol. 12 at 3307, 3315. As the district court noted, respondents may have believed that they were being asked whether the ad had resulted from use of the search term “1-800-CONTACTS.” See Aplt. App., Vol. 3 at 5474-75 (Survey Order at 19-20). An affirmative answer based on this belief would not have been at all probative of the likelihood of confusion that 1-800 has alleged. In presenting the survey responses, 1-800’s expert lumped together the affirmative responses to the ambiguous question with the affirmative responses to the question whether the respondent believed that the Lens.com ad “ha[d] sponsorship or approval from 1-800-CON-TACTS.” Id., Vol. 12 at 3308, 3316. As a result, the court had no way of accurately discounting the survey data for any misunderstandings that might have arisen from the ambiguity of the first question’s language. In any event, even assuming that the survey should have been admitted, it does not warrant reversal of summary judgment because it was insufficiently probative of confusion to overcome the factors discussed above. The survey revealed that the relevant confusion was fairly low. To isolate confusion arising specifically from the use of 1-800’s mark as a search term and keyword, the survey used a control group; respondents in this group were told to imagine that they had searched for the term contact lenses rather than the term 1800contacts. When these control-group respondents were asked whether they thought that the Lens.com ad either originated from 1-800 or had sponsorship or approval from 1-800,11.9% answered in the affirmative. By comparison, 19.4% of respondents in the first noncontrol group and 19.2% of respondents in the second noncontrol group answered likewise. Subtracting the control group’s 11.9% rate of confusion, one is left with net confusion rates of only 7.5% and 7.3% for the two noncontrol groups, or an average net confusion rate of only 7.4%. See 6 J. Thomas McCarthy, McCarthy on Trademarks and Unfair Competition § 32:187 at 32-432 (4th ed.2013) (6 McCarthy) (noting use of such corrections to eliminate the “general background noise” of confusion in predicting the likelihood of confusion (internal quotation marks omitted)). The 7.4% figure is at (or below) the lowest confusion rate that, together with other evidence supporting confusion, could justify a conclusion that consumer confusion was likely. 1-800 relies on Grotrian, Helfferich, Schulz, Th. Steinweg Nachf. v. Steinway & Sons, 365 F.Supp. 707, 709 (S.D.N.Y.1973), in which the plaintiff, a piano maker, sought a declaratory judgment that its trademark Grotrian-Steinweg was not likely to cause confusion with the defendant’s trademarks, Steinway and Steinway & Sons. Following a bench trial, the district court found that confusion was likely. See id. at 719-20. This finding was based on a welter of evidence in the defendant’s favor: The defendants had diligently promoted and continuously used the Steinway name in the United States for many years, making it a strong trademark. See id. at 712-13. The plaintiffs had “candidly adopted the name Steinweg for the sole purpose of exploiting the Steinweg name in exporting pianos to English-speaking countries ... despite knowledge of the defendant’s trademark and its objections.” Id. at 714. The parties’ marks, as well as their products and the manner in which they marketed them, were closely similar. See id. at 714-15. And the defendants submitted one survey that “consisted of a series of 23 tape-recorded personal' interviews with recent purchasers of Grotrian-Steinweg pianos,” id. at 716, many of whom displayed confusion between the two brand names, see id. n. 33 (reporting “[e]xemplitive responses”' that Steinweg was “a Steinway made in Germany,” “the Steinway Company in Germany,” “the parent company to Steinway,” and “the original Steinway, the German Steinway” (internal quotation marks omitted)). Against this backdrop, the district court considered a second survey in which 7.7% of respondents “p[er]ceived a business .connection between the two companies and 8.5% confused the names.” Id. at 716. The court found that the two surveys; taken together, were “strong evidence of the likelihood of confusion which the Lanham Act was designed to prohibit.” Id. On appeal the Second Circuit ruled only that the district court did not err “in giving weight to the surveys as evidence of actual confusion.” Grotrian, Helfferich, Schulz, Th. Steinweg Nachf. v. Steinway & Sons, 523 F.2d 1331, 1341 (2d Cir.1975). In contrast, the Seventh Circuit has ruled that a district court was correct to hold that a survey showing a 7.6% confusion rate weighed against infringement. See Henri’s, 717 F.2d at 358-59. The court surveyed relevant cases, including Grotrian, and remarked that it had found “no case in which a 7.6% figure constituted likelihood of confusion.” Id. at 358. It further noted that the district court’s opinion in Grotrian did not make clear what overlap, if any, existed between the 7.7% of respondents who perceived a business connection between the two piano makers and the 8.5% who confused the names. See id. Indeed, a later decision by the Southern District of New York apparently assumed that no such overlap existed, explaining that in Grotrian “at least 15 percent of consumers were confused as to source or endorsement.” Weight Watchers Int’l, Inc. v. Stouffer Corp., 744 F.Supp. 1259, 1274 (S.D.N.Y.1990) (emphasis added). The Weight Watchers court assigned little value to a survey that showed a 9.2% rate of confusion. See id. 1-800 also cites Goya Foods, Inc. v. Condal Distributors, Inc., 732 F.Supp. 453, 457 n. 7 (S.D.N.Y.1990), for the proposition that any rate of confusion greater than 7% is meaningful. In that case the plaintiff, Goya, contended that the packaging of the defendant’s “Condal” five- and ten-pound bags of rice imitated the packaging of Goya’s “Canilla” rice. See id. at 453. The court found confusion, based on the strength of the Canilla mark, the similarity in the packaging, the close competitive proximity of the products, the lack of sophistication of rice consumers, and three surveys. See id. at 454-58. The court dismissed one survey as not designed to measure the relevant confusion. See id. at 456. A second survey reported that when respondents were shown a package of Condal rice, 44.9% said that the first company or brand that came to mind was Goya or Canilla; and of those, 27.5% said that the packaging caused them to say that and 20.5% said that the Condal and Canilla packages were very similar or identical. See id. In the third survey, respondents were handed an empty Condal bag and were asked what brand it was; 9% identified the bag as Canilla or Goya. See id. Goya’s expert said that “any figure greater than 7 percent in this sort of study is meaningful and represents a real confusion.” Id. at 457 (emphasis added) (internal quotation marks omitted). The court said that the expert’s statement was supported by case law, citing only Grotrian. See id. n. 7. We are not persuaded that Grotrian and Goya support the proposition that surveys showing confusion as low as 7% can by themselves sustain a finding of likelihood of confusion. Both eases seem rightly decided, but primarily because of the strength of other factors supporting confusion and other persuasive survey results. The strongest statement that can be made based on those opinions is that “surveys without obvious defects indicating confusion of seven percent to 15 percent of the sample have been held adequate, when supported by other evidence, to prove a likelihood of confusion.” Restatement (Third) of Unfair Competition § 20 cmt. g. at 216-17 (1995) (emphases' added); cf id. at 217 (“The weight to be accorded a specific survey depends on the facts and circumstances of each case. The fact that a particular percentage is held sufficient to establish infringement in one case thus does not necessarily indicate that it is sufficient to establish infringement in other cases.”). The great weight of authority appears to be that “[w]hen the percentage results of a confusion survey dip below 10%, they can become evidence which will indicate that confusion is not likely.” 6 McCarthy § 32:189 at 32-440 (emphasis added). Thus, 1-800’s survey is entitled to no more than minimal weight. And that minimal weight cannot sustain a finding of likelihood of confusion in the circumstances presented here. The other factors, including the hard data noted above, overwhelmingly indicate the unlikelihood of confusion. Even if the survey was admissible evidence, summary judgment for Lens.com was required. B. Secondary Liability for Ads Placed by Lens.com Affiliates 1-800 claims that Lens.com should have been denied summary judgment on the claims of secondary liability for infringement allegedly committed by affiliates who published ads on its behalf. 1-800’s arguments focus exclusively on the conduct of two affiliates, Goggans and McCoy. Both Goggans and McCoy purchased keywords that were either identical or closely similar to 1-800’s service mark. In addition, McCoy published at least one ad for www. JustLenses.com that featured a close variation of the mark in its text. Again, 1-800’s theories of secondary liability are vicarious liability and contributory infringement. Vicarious liability arises when common-law principles of agency impose liability on the defendant for the infringing acts of its agent. See Procter & Gamble, 317 F.3d at 1127-28. Contributory infringement occurs when the defendant either (1) intentionally induces a third party to infringe on the plaintiffs mark or (2) enables a third party to infringe on the mark while knowing or having reason to know that the third party is infringing, yet failing to take reasonable remedial measures. See Inwood, 456 U.S. at 853-54, 102 S.Ct. 2182; Procter & Gamble, 317 F.3d at 1128 (“An action for contributory liability is not limited to a manufacturer, but may also extend to licensors, franchisers, or to similarly situated third parties.”); Coach, Inc. v. Goodfellow, 717 F.3d 498, 505 (6th Cir.2013) (defendant was liable for contributory infringement “because he knew or had reason to know of the infringing activities and yet continued to facilitate those activities ... without undertaking a reasonable investigation or taking other appropriate remedial measures”). Vicarious and contributory liability must be predicated on some direct infringement by the third party. See 4 McCarthy § 25:17 (“By definition, there can be no liability for contributory infringement unless there is direct infringement.”); cf. La Resolana Architects, PA v. Reno, Inc., 555 F.3d 1171, 1181 (10th Cir.2009) (“[B]oth contributory and vicarious infringements require someone to have directly infringed the copyright.”). Lens, com therefore cannot incur secondary liability unless one of the affiliates in question directly violated the Lanham Act. As noted, the low ratio of clicks to impressions associated with Lens.com’s own keyword use and the other King of the Mountain factors convince us that summary judgment was appropriate on 1-800’s direct-infringement claim. The same factors and similar data convince us that insofar as Goggans and McCoy used keywords that resulted in ads for Lens.com entities that did not display 1-800’s mark in their text, no genuine factual issue exists regarding likelihood of confusion. As for hard data, the record reveals that McCoy’s use of “1800Contacts” or some variation thereof as a keyword generated more than 448,000 impressions whose text did not display the mark. Of these impressions, at most 3,163 — or about .7%— resulted in clicks. Likewise, one of 1-800’s own exhibits revealed that Goggans’s use of “1800Contacts” as a keyword generated 242,864 impressions for wwwJust Lenses.com that did not display the mark in their text, and only 1,445 of the impressions — also fewer than 1% — resulted in clicks. 1-800 does not dispute these numbers, which are even more in Lens.com’s favor than the 1.5% clicks-to-impressions rate for 1-800’s direct-liability claim. Thus, to the extent that 1-800’s secondary-liability claim derives from keyword use by Goggans and McCoy that did not generate ads containing the 1800CONTACTS mark, there is insufficient evidence of direct infringement. And absent any evidence of direct infringement, Lens.com cannot be secondarily liable. The district court properly granted summary judgment to Lens.com on this keyword use. 1-800’s only remaining claim is that Lens.com is secondarily liable for McCoy’s publication of ads that featured variations of the 1-800 mark in their text. We examine vicarious and contributory infringement on this claim. 1. Vicarious Liability. The district court granted summary judgment to Lens.com on 1-800’s vicarious-liability theory, ruling that the evidence would not support a reasonable inference that the affiliates were Lens.com’s agents. See 1-800 Contacts, 755 F.Supp.2d at 1182-84. We have some concerns with the district court’s analysis and Lens.com’s arguments that there was no agency relationship. First, one need not show a fiduciary relationship to establish that an agency relationship exists; rather, fiduciary duties arise as a result of circumstances establishing the agency relationship. See Restatement (Third) of Agency § 1.01 cmt. e. at 23 (2006) (“To establish that a relationship is one of agency, it is not necessary to prove its fiduciary character as an element.”). Second, that certain affiliates may have w