Concurrence by Judge WATFORD
OPINION
TALLMAN, Circuit Judge:
I.
Between February 29, 2012, and March 10, 2012, Chris Gillespie used a domain name registrar to acquire 763 domain names that included the word “google.” Each of these domain names paired the word “google” with some other term identifying a specific brand, person, or product—for example, “googledisney.com,” “googlebarackobama.net,” and “google-newtvs.com.”
Google, Inc. (“Google”) objected to these registrations and promptly filed a complaint with the National Arbitration Forum (“NAF”), which has authority to decide certain domain name disputes under the registrar’s terms of use. Google argued that the registrations violate the Uniform Domain Name Dispute Resolution Policy, which is included in the registrar’s terms of use, and amount to domain name infringement, colloquially known as “cybers-quatting.” Specifically, Google argued that the domain names are confusingly similar to the GOOGLE trademark1 and were registered in bad faith. The NAF agreed, and transferred the domain names to Goo-gle on May 10, 2012.
Shortly thereafter, David Elliott filed, and Gillespie later joined,2 an action in the [1155]*1155Arizona District Court. Elliott petitioned for cancellation of the GOOGLE trademark under the Lanham Act, which allows cancellation of a registered trademark if it is primarily understood as a “generic name for the goods or services, or a portion thereof, for which it is registered.” 15 U.S.C. § 1064(3). Elliott petitioned for cancellation on the ground that the word “goo-gle” is primarily understood as “a generic term universally used to describe the act[ ] of internet searching.”
On September 28, 2013, the parties filed cross-motions for summary judgment on the issue of genericness. Elliott requested summary judgment because (1) it is an indisputable fact that a majority of the relevant public uses the word “google” as a verb—i.e., by saying “I googled it,” and (2) verb use constitutes generic use as a matter of law. Google maintained that verb use does not automatically constitute generic use, and that Elliott failed to create even a triable issue of fact as to whether the GOOGLE trademark is generic. Specifically, Google argued that Elliott failed to present sufficient evidence to support a jury finding that the relevant public primarily understands the word “google” as a generic name for internet search engines. The district court agreed with Google and its framing of the relevant inquiry, and granted summary judgment in its favor.
Elliott raises two arguments on appeal. First, he argues that the district court misapplied the primary significance test and failed to recognize the importance of verb use. Second, he argues that the district court impermissibly weighed the evidence when it granted summary judgment for Google. We review the district court’s grant of summary judgment de novo and ask, viewing the evidence in the light most favorable to Elliott, “whether there are any genuine issues of material fact and whether the district court correctly applied the relevant substantive law.” KP Permanent Make-Up, Inc. v. Lasting Impression I, Inc., 408 F.3d 596, 602 (9th Cir. 2005) (citing Clicks Billiards, Inc. v. Sixshooters, Inc., 251 F.3d 1252, 1257 (9th Cir. 2001)). For the reasons described below, we reject both of Elliott’s arguments and affirm summary judgment for Google.
II.
We recognize four categories of terms with regard to potential trademark protection: (1) generic, (2) descriptive, (3) suggestive, and (4) arbitrary or fanciful terms. Filipino Yellow Pages, Inc. v. Asian Journal Publ’ns, Inc., 198 F.3d 1143, 1146 (9th Cir. 1999) (quoting Surgicenters of Am., Inc. v. Med. Dental Surgeries, Co., 601 F.2d 1011, 1014 (9th Cir. 1979)). This case involves the first and fourth categories, which lie at opposite ends of the spectrum with regard to pro-tectability. At one extreme, generic terms are “common descriptive” names which identify only the type of good “of which the particular product or service is a species.” Park ‘N Fly, Inc. v. Dollar Park & Fly, Inc., 718 F.2d 327, 329 (9th Cir. 1983), rev’d on other grounds, 469 U.S. 189, 105 S.Ct. 658, 83 L.Ed.2d 582 (1985). Generic terms are not protectable because they do not identify the source of a product. Id. At the other extreme, arbitrary or fanciful marks “employ words and phrases with no commonly understood connection to the product.” JL Beverage Co. v. Jim Beam Brands Co., 828 F.3d 1098, 1107 (9th Cir. 2016). Arbitrary or fanciful marks are “automatically entitled to protection because they naturally serve to identify a particular source of a product.” KP Permanent Make-Up, Inc., 408 F.3d at 602 (alterations omitted) (quoting Two Pesos, Inc. v. Taco Cabana, Inc., 505 U.S. 763, 768, 112 S.Ct. 2753, 120 L.Ed.2d 615 (1992)).
Over time, the holder of a valid trademark may become a “victim of ‘gen-[1156]*1156ericide.’ ” Freecycle Network, Inc. v. Oey, 505 F.3d 898, 905 (9th Cir. 2007) (quoting J. Thomas McCarthy, McCarthy on Trademarks and Unfair Competition § 12:1 (4th ed. 1998) [hereinafter McCarthy]). Generi-eide occurs when the public appropriates a trademark and uses it as a generic name for particular types of goods or services irrespective of its source. For example, ASPIRIN, CELLOPHANE, and ESCALATOR were once protectable as arbitrary or fanciful marks because they were primarily understood as identifying the source of certain goods. But the public appropriated those marks and now primarily understands aspirin, cellophane, and escalator as generic names for those same goods. See Bayer Co. v. United Drug Co., 272 F. 505, 510 (S.D.N.Y. 1921); DuPont Cellophane Co. v. Waxed Prods. Co., 85 F.2d 75, 82 (2d Cir. 1936); Freecycle Network, Inc., 505 F.3d at 905. The original holders of the ASPIRIN, CELLOPHANE, and ESCALATOR marks are thus victims of genericide.
The question in any case alleging generi-cide is whether a trademark has taken the “fateful step” along the path to genericness. Ty Inc. v. Softbelly’s Inc., 353 F.3d 528, 531 (7th Cir. 2003). The mere fact that the public sometimes uses a trademark as the name for a unique product does not immediately render the mark generic. See 15 U.S.C. § 1064(3). Instead, a trademark only becomes generic when the “primary significance of the registered mark to the relevant public” is as the name for a particular type of good or service irrespective of its source. Id.
We have often described this as a “who-are-you/what-are-you” test. See Yellow Cab Co. of Sacramento v. Yellow Cab of Elk Grove, Inc., 419 F.3d 925, 929 (9th Cir. 2005) (quoting Filipino Yellow Pages, Inc., 198 F.3d at 1147). If the relevant public primarily understands a mark as describing “who” a particular good or service is, or where it comes from, then the mark is still valid. But if the relevant public primarily understands a mark as describing “what” the particular good or service is, then the mark has become generic. In sum, we ask whether “the primary significance of the term in the minds of the consuming public is [now] the product [and not] the producer.” Kellogg Co. v. Nat’l Biscuit Co., 305 U.S. 111, 118, 59 S.Ct. 109, 83 L.Ed. 73 (1938).
A.
On appeal, Elliott claims that he has presented sufficient evidence to create a triable issue of fact as to whether the GOOGLE trademark is generic, and that the district court erred when it granted summary judgment for Google. First, he argues that the district court erred because it misapplied the primary significance test and failed to recognize the importance of verb use. Specifically, he argues that the district court erroneously framed the inquiry as whether the primary significance of the word “google” to the relevant public is as a generic name for internet search engines, or as a mark identifying the Google search engine in particular. Instead, Elliott argues that the court should have framed the inquiry as whether the relevant public primarily uses the word “google” as a verb.
We conclude that Elliott’s proposed inquiry is fundamentally flawed for two reasons. First, Elliott fails to recognize that a claim of genericide must always relate to a particular type of good or service. Second, he erroneously assumes that verb use automatically constitutes generic use. For similar reasons, we conclude that the district court did not err in its formulation of the relevant inquiry under the primary significance test.
[1157]*1157First, we take this opportunity to clarify that a claim of genericide or gener-icness must be made with regard to a particular type of good or service. We have not yet had occasion to articulate this requirement because parties usually present their claims in this manner sua sponte. See, e.g., KP Permanent Make-Up, Inc., 408 F.3d at 605 (claiming that “micro colors” is generic for micropigmentation services); Filipino Yellow Pages, Inc., 198 F.3d at 1146 (claiming that “Filipino Yellow Pages” is generic for “telephone directories targeted at the Filipino-American community”); Park ‘N Fly, Inc., 718 F.2d at 330 (claiming that “Park ‘N Fly” is generic for airport parking lots). But here, Elliott claims that the word “google” has become a generic name for “the act” of searching the internet, and argues that the district court erred when it focused on internet search engines. We reject Elliott’s criticism and conclude that the district court properly recognized the necessary and inherent link between a claim of gen-ericide and a particular type of good or service.
This requirement is clear from the text of the Lanham Act, which allows a party to apply for cancellation of a trademark when it “becomes the generic name for the goods or services ... for which it is registered.” 15 U.S.C. § 1064(3) (emphasis added). The Lanham Act further provides that “[i]f the registered mark becomes the generic name for less than all of the goods or services for which it is registered, a petition to cancel the registration for only those goods or services may be filed.” Id. (emphasis added). Finally, the Lanham Act specifies that the relevant question under the primary significance test is “whether the registered mark has become the generic name of [certain] goods or services.” Id. (emphasis added). In this way, the Lanham Act plainly requires that a claim of genericide relate to a particular type of good or service.
We also note that such a requirement is necessary to maintain the viability of arbitrary marks as a protectable trademark category. By definition, an arbitrary mark is an existing word that is used to identify the source of a good with which the word otherwise has no logical connection. See JL Beverage Co., 828 F.3d at 1107. If there were no requirement that a claim of gener-icide relate to a particular type of good, then a mark like IVORY, which is “arbitrary as applied to soap,” could be can-celled outright because it is “generic when used to describe a product made from the tusks of elephants.” Abercrombie & Fitch Co. v. Hunting World, Inc., 537 F.2d 4, 9 n.6 (2d Cir. 1976). This is not how trademark law operates: Trademark law recognizes that a term may be unprotectable with regard to one type of good, and pro-tectable with regard to another type of good. In this way, the very existence of arbitrary marks as a valid trademark category supports our conclusion that a claim of genericide must relate to a particular type of good or service.
Second, Elliott’s alternative inquiry fails because verb use does not automatically constitute generic use. Elliott claims that a word can only be used in a trademark sense when it is used as an adjective. He supports this claim by comparing the definitions of adjectives and trademarks, noting that both adjectives and trademarks serve descriptive functions.
Once again, Elliott’s semantic argument contradicts fundamental principles underlying the proteetability of trademarks. When Congress amended the Lanham Act to specify that the primary significance test applies to claims of genericide, it specifically acknowledged that a speaker might use a trademark as the name for a product, i.e., as a noun, and yet use the [1158]*1158mark with a particular source in mind, i.e., as a trademark. It further explained that:
A trademark can serve a dual function— that of [naming] a product while at the same time indicating its source. Admittedly, if a product is unique, it is more likely that the trademark adopted and used to identify that product will be used as if it were the identifying name of that product. But this is not conclusive of whether the mark is generic.
S. Rep. No. 98-627, at 5 (1984). In this way, Congress has instructed us that a speaker might use a trademark as a noun and still use the term in a source-identifying trademark sense.
Moreover, we have already implicitly rejected Elliott’s theory that only adjective use constitutes trademark use. In Coca-Cola Co. v. Overland, Inc., 692 F.2d 1250 (9th Cir. 1982), the Coca-Cola Company sued a local restaurant for trademark infringement because its servers regularly and surreptitiously replaced customer orders for “a coke” with a non-Coca-Cola beverage. Id. at 1252. The restaurant defended on the basis of genericide, arguing that the COKE trademark had become a generic name for all cola beverages. Id. at 1254. To support its claim, the restaurant presented employee affidavits stating that the employees believed that Customers who ordered “a coke” were using the term in a generic sense. Id. We rejected these affidavits because they were not based on personal knowledge. More significant to the issue at hand, we also noted that the mere fact that customers ordered “a coke,” i.e., used the mark as a noun, failed to show '“what ... customers [were] thinking,” or whether they had a particular source in mind. Id. at 1255.
If Elliott were correct that a trademark can only perform its source-identifying function when it is used as an adjective, then we would not have cited a need for evidence regarding the customers’ inner thought processes. Instead, the fact that the customers used the trademark as a noun and asked for “a coke” would prove that they had no particular source in mind. In this way, we have implicitly rejected Elliott’s theory that a trademark can only serve a source-identifying function when it is used as an adjective.
For these reasons, the district court correctly rejected Elliott’s theory that verb use automatically constitutes generic use.3 Moreover, the district court aptly coined the terms “discriminate verb” and “indiscriminate verb” in order to evaluate Elliott’s proffered examples of verb use and determine whether they were also examples of generic use. Although novel, these terms properly frame the relevant inquiry as whether a speaker has a particular source in mind. We have already acknowledged that a customer might use the noun “coke” in an indiscriminate sense, with no particular cola beverage in mind; or in a discriminate sense, with a Coca-Cola beverage in mind. In the same way, we now recognize that an internet user might use the verb “google” in an indiscriminate sense, with no particular search engine in mind; or in a discriminate sense, with the Google search engine in mind.
Because a claim of genericide must relate to a particular type of good or service [1159]*1159and because verb use does not necessarily constitute generic use, the district court did not err when it refused to frame its inquiry as whether the relevant public primarily uses the word “google” as a verb. Moreover, the district court correctly framed its inquiry as whether the primary significance of the word “google” to the relevant public is as a generic name for internet search engines or as a mark identifying the Google search engine in particular. We therefore evaluate Elliott’s claim of genericide and the sufficiency of his proffered evidence under the proper inquiry.
B.
Elliott next argues that the district court must have impermissibly weighed the evidence when it granted summary judgment for Google in light of the “sheer quantity” of evidence that Elliott produced to support his claim of gen-ericide. See Jesinger v. Nev. Fed. Credit Union, 24 F.3d 1127, 1131 (9th Cir. 1994) (noting that a court “must not weigh the evidence” at summary judgment). We disagree. Instead, we conclude that Elliott’s admissible evidence is largely inapposite to the relevant inquiry under the primary significance test because Elliott ignores the fact that a claim of genericide must relate to a particular type of good or service.
A party applying for cancellation of a registered trademark bears the burden of proving genericide by a preponderance of the evidence. Anti-Monopoly, Inc. v. Gen. Mills Fun Grp., 684 F.2d 1316, 1319 (9th Cir. 1982). Moreover, the holder of a registered trademark benefits from a presumption of validity and has “met its [initial] burden of demonstrating” the lack of “a genuine issue of material fact” regarding genericide. Coca-Cola Co., 692 F.2d at 1254. Therefore, in light of the relevant inquiry under the primary significance test, Elliott was required to identify sufficient evidence to support a jury finding that the primary significance of the word “google” to the relevant public is as a name for internet search engines generally and not as a mark identifying the Google search engine in particular.
At summary judgment, the district court assumed that a majority of the public uses the verb “google” to refer to the act of “searching on the internet without regard to [the] search engine used.”4 In other words, it assumed that a majority of the public uses the verb “google” in a generic and indiscriminate sense. The district court then concluded that this fact, on its own, cannot support a jury finding of gen-ericide under the primary significance test. We agree.
As explained above, a claim of genericide must relate to a particular type of good. Even if we assume that the public uses the verb “google” in a generic and indiscriminate sense, this tells us nothing about how the public primarily understands the word itself, irrespective of its grammatical function, with regard to internet search engines. As explained below, we also agree that Elliott’s admissible evidence only supports the favorable but insufficient inference already drawn by the district court—that a majority of the public uses the verb “google” in a generic sense. Standing in isolation,5 this fact is [1160]*1160insufficient to support a jury finding of genericide. The district court therefore properly granted summary judgment for Google.
We begin with Elliott’s three consumer surveys. Consumer surveys may be used to support a claim of genericide “so long as they are conducted according to accepted principles.” Stuhlbarg Int’l Sales Co. v. John D. Brush & Co., 240 F.3d 832, 840 (9th Cir. 2001). Here, the district court properly excluded two of Elliott’s consumer surveys because they were not conducted according to accepted principles. Specifically, these surveys were designed and conducted by Elliott’s counsel, who is not qualified to design or interpret surveys. See Federal Judicial Center, Reference Manual on Scientific Evidence 864 (3d ed. 2011) (explaining that valid survey design typically requires graduate training or professional experience in survey research).6
The district court properly considered only Elliott’s third survey, which was conducted by James Berger—a qualified survey expert. Elliott’s third survey is a “Thermos” survey, which generally “puts the respondent in an imaginary situation ... and asks how the respondent would ask” for the type of good for which the trademark is alleged to be generic. McCarthy § 12:15 (citing Am. Thermos Prods. Co. v. Aladdin Indus., 207 F.Supp. 9, 21-22 (D. Conn. 1962), aff'd, 321 F.2d 577 (2d Cir. 1963)). Here, Berger asked 251 respondents: “If you were going to ask a friend to search for something on the Internet, what word or phrase would you use to tell him/her what you want him/her to do?” Over half of the 251 respondents answered this question by using the word “google” as a verb.
Although verb use does not automatically constitute generic use, the district court allowed Berger to rely on the third survey to offer his expert “opinion that a majority of the public uses the word google as a [generic and indiscriminate] verb to mean search on the internet.” In this way, Elliott’s admissible consumer survey evidence goes no further than supporting the favorable inference already drawn by the district court.7
We next consider Elliott’s examples of alleged generic use by the media and by consumers. Documented examples of generic use might support a claim of genericide if they reveal a prevailing public consensus regarding the primary significance of a registered trademark. See McCarthy § 12:13 (explaining that generic [1161]*1161use by the media is a “strong indication of the general public’s perception”) (quoting Murphy Door Bed Co. v. Interior Sleep Sys., Inc., 874 F.2d 95, 101 (2d Cir. 1989)). However, if the parties offer competing examples of both generic and trademark use, this source of evidence is typically insufficient to prove genericide. See id.
Initially, we note that Elliott’s admissible examples are only examples of verb use. To repeat, verb use does not automatically constitute generic use. For instance, Elliott purports to offer an example of generic use by T-Pain, a popular rap music artist. But we will not assume that T-Pain is using the word “google” in a generic sense simply because he tells listeners to “google [his] name.” T-Pain, Bottlez, on rEVOLVEr (RCA Records 2011). Without further evidence regarding T-Pain’s inner thought process, we cannot tell whether he is using “google” in a discriminate or indiscriminate sense. In this way, many of Elliott’s admissible examples do not even support the favorable inference that a majority of the relevant public uses the verb “google” in a generic sense.
Elliott also attempted to offer clear examples of indiscriminate verb use by the media and by consumers. For example, in response to Google’s motion for summary judgment, he produced a transcript from an episode of a German television show in which a character claims to have “googled at Wikipedia.” Elliott also produced examples in which the media uses phrases like “googled on ebay,” “goo-gled on facebook,” and “googled on pinter-est.” Finally, Elliott produced evidence suggesting that certain consumers claimed that they accessed a website by “googling” it, even though those consumers actually accessed the website through a non-Google search engine.
The district court properly excluded these examples of indiscriminate verb use because they were not disclosed during discovery and because Elliott failed to show that his delay was “substantially justified or ... harmless.” Fed. R. Civ. P. 37(c)(1); see also Yeti by Molly, Ltd. v. Deckers Outdoor Corp., 259 F.3d 1101, 1106 (9th Cir. 2001) (“[W]e give particularly wide latitude to the district court’s discretion to issue [discovery] sanctions. ...”). Moreover, even if these examples had been timely disclosed, they are largely irrelevant because they only support the favorable inference already drawn by the district court.
We next consider Elliott’s proffered expert testimony. Each of Elliott’s experts, including Dr. Berger, Dr. Patrick Farrell, and Dr. Allan Metcalf, opine that the word “google” is used in a generic sense when it is used as a verb.8 On its face, this testimony simply supports the favorable inference already drawn by the district court.
Next, we consider Elliott’s proffered dictionary evidence. See McCarthy § 12:13 (noting that dictionary definitions are “sometimes persuasive in determining public usage”). Elliott does not present any examples where “google” is defined as a generic name for internet search engines. Instead, Elliott presents secondary definitions where google is defined as a verb. See, e.g., Google, CollinsEnglishDiction-ary.com, https://www.collinsdictionary.com/ dictionary/english/google (last visited'Apr. 15, 2017) (defining google primarily as a “trademark” but secondarily as a verb meaning “to search for (something on the [1162]*1162internet) using a search engine”); Google, Dictionary.com, http://dictionary.reference. com/browse/google (last visited Apr. 15, 2017) (defining google primarily as the “brand name of a leading Internet search engine” but secondarily as a verb meaning “to search the Internet for information about [something]”). Once again, Elliott’s proffered dictionary evidence only supports the favorable inference already drawn by the district court.9
Next, we consider Elliott’s claim that Google has used its own trademark in a generic sense. Generic use of a mark by the holder of that mark can support a finding of genericide. See McCarthy § 12:13. However, Elliott has not presented an example of generic use by Google. Instead, Elliott has presented an email from Google cofounder Larry Page, which encourages recipients to “[h]ave fun and keep googling!” Once again, Elliott relies on an example of verb use. Elliott has not shown, nor is it likely that he could show, that the cofounder of Google had no particular search engine in mind when he told recipients of the “Google Friends Newsletter” to “keep googling.” 10
Finally, we consider Elliott’s claim that there is no efficient alternative for the word “google” as a name for “the act” of searching the internet regardless of the search engine used. Once again, a claim of genericide must relate to a particular type of good or service. In order to show that there is no efficient alternative for the word “google” as a generic term, Elliott must show that there is no way to describe “internet search engines” without calling them “googles.” Because not a single competitor calls its search engine “a google,” and because members of the consuming public recognize and refer to different “internet search engines,” Elliott has not shown that there is no available substitute for the word “google” as a generic term. Compare, e.g., Q-Tips, Inc. v. Johnson & Johnson, 108 F.Supp. 845, 863 (D.N.J. 1952) (concluding that “medical swab” and “cotton-tipped applicator” are efficient alternatives for Q-Tips); with Bayer Co., 272 F. at 505 (concluding that there is no efficient substitute for the generic term “aspirin” because consumers do not know the term “acetyl salicylic acid”); see also Softbelly’s Inc., 353 F.3d at 531 (explaining that genericide does not typically occur “until the trademark has gone so far toward becoming the exclusive descriptor of the product that sellers of competing brands cannot compete effectively without using the name”).
Elliott cannot survive summary judgment based on “sheer quantity” of irrelevant evidence. We agree with the district court that, at best, Elliott has presented admissible evidence to support the inference that a majority of the relevant public uses the verb “google” in a generic sense. Because this fact alone cannot support a claim of genericide, the district court properly granted summary judgment for Goo-gle.
[1163]*1163HI.
The district court did not misapply the primary significance test, nor did it weigh the evidence when it granted summary judgment for Google. We agree that Elliott has failed to present sufficient evidence to support a jury finding that the relevant public primarily understands the word “google” as a generic name for internet search engines and not as a mark identifying the Google search engine in particular. We therefore affirm the district court’s grant of summary judgment.
Costs shall be taxed against Elliott. See Fed. R. App. P. 39(a)(2).
AFFIRMED.