Dz Reserve v. Meta Platforms, Inc.

FOR PUBLICATION UNITED STATES COURT OF APPEALS FOR THE NINTH CIRCUIT DZ RESERVE; CAIN MAXWELL, No. 22-15916 DBA Max Martialis, D.C. No. Plaintiffs-Appellees, 3:18-cv-04978-JD v. OPINION META PLATFORMS, INC., FKA Facebook, Inc., Defendant-Appellant. Appeal from the United States District Court for the Northern District of California James Donato, District Judge, Presiding Argued and Submitted September 12, 2023 San Francisco, California Filed March 21, 2024 Before: J. Clifford Wallace, Sidney R. Thomas, and Danielle J. Forrest, Circuit Judges. Opinion by Judge Sidney R. Thomas; Partial Dissent by Judge Forrest 2 DZ RESERVE V. META PLATFORMS, INC. SUMMARY * Class Certification The panel affirmed the district court’s order certifying one class of advertisers who paid Meta Platforms, Inc. (Meta) to place advertisements on its social media platforms—the damages class, and vacated the district court’s order certifying another class of advertisers—the injunction class. The advertisers alleged that Meta fraudulently misrepresented the “Potential Reach” of advertisements on its platforms by stating that Potential Reach was an estimate of people, although it was actually an estimate of accounts. The panel affirmed the district court’s certification under Fed. R. Civ. P. 23(b)(3) of the damages class. The misrepresentation constituted a “common course of conduct” under the test for determining whether common issues predominate among the class. Given that all class members encountered the same misrepresentation about Potential Reach—the nucleus of the fraud—the slight variations in the other information available on the Ads Manager did not defeat the commonality of the misrepresentation. The district court properly determined that the element of justifiable reliance was capable of classwide resolution. The panel affirmed the district court’s holding that the requirements of typicality and adequacy were satisfied. Accordingly, the district court did not abuse * This summary constitutes no part of the opinion of the court. It has been prepared by court staff for the convenience of the reader.  DZ RESERVE V. META PLATFORMS, INC. 3 its discretion in determining that Fed. R. Civ. P. 23(b)(3) was satisfied. The panel vacated the certification of the Rule 23(b)(2) injunction class for the district court to reconsider whether the named Plaintiff Cain Maxwell had Article III standing to seek an injunction. The district court had no occasion to consider the record or to analyze Meta’s argument against Maxwell’s standing to seek injunctive relief. Dissenting in part, Judge Forrest agreed that the district court’s certification of the injunction class must be vacated and remanded for the district court to reconsider whether Plaintiff Cain Maxwell had standing to pursue that claim. She disagreed that the district court properly certified the damages class because Plaintiffs cannot satisfy the predominance requirement where there were individual questions that must be answered related to multiple elements of Plaintiffs’ fraud-based claims. COUNSEL Geoffrey Graber (argued), Andrew N. Friedman, Karina G. Puttieva, and Madelyn Petersen, Cohen Milstein Sellers & Toll PLLC, Washington, D.C.; Eric Kafka, Cohen Milstein Sellers & Toll PLLC, New York, New York; Charles Reichmann, Law Offices of Charles Reichmann, Kensington, California; for Plaintiffs-Appellees. Andrew B. Clubok (argued), Susan E. Engel, and Margaret A. Upshaw, Latham & Watkins LLP, Washington, D.C.; Elizabeth L. Deeley, Melanie M. Blunschi, Nicholas Rosellini, and Nicole Valco, Latham & Watkins LLP, San 4 DZ RESERVE V. META PLATFORMS, INC. Francisco, California; Samir Deger-Sen, Latham & Watkins LLP, New York, New York; for Defendant-Appellant. Jennifer B. Dickey and Jordan L. Von Bokern, U.S. Chamber Litigation Center, Washington, D.C.; Erik R. Zimmerman, Jazzmin M. Romero, and Jordan T. DeJaco, Robinson Bradshaw & Hinson PA, Chapel Hill, North Carolina; for Amicus Curiae Chamber of Commerce of the United States of America. David M. Berger, Gibbs Law Group LLP, Oakland, California, for Amicus Curiae Digital Content Next. OPINION S.R. THOMAS, Circuit Judge: Meta Platforms, Inc. (Meta), formerly known as Facebook, appeals the district court’s order certifying two classes of advertisers who paid Meta to place advertisements on its social media platforms—a damages class and an injunction class. The advertisers allege that Meta fraudulently misrepresented the “Potential Reach” of advertisements on its platforms by stating that Potential Reach was an estimate of people, although it was actually an estimate of accounts. As to the damages class, the primary issue on appeal is whether that misrepresentation constitutes a “common course of conduct” under our test for determining whether common issues predominate among the class. We conclude that it does. Because the district court did not abuse its discretion in determining that Federal Rule of Civil Procedure 23(b)(3) was satisfied, we affirm the certification of the damages class. However, we vacate the  DZ RESERVE V. META PLATFORMS, INC. 5 certification of the Rule 23(b)(2) injunction class for the district court to reconsider whether the named Plaintiffs have standing to seek an injunction. I Meta owns and operates several online social media and messaging platforms and applications, including Facebook, Instagram, and WhatsApp. As with many social media companies, Meta “generates substantially all of its revenue from advertising.” In 2018, a nationwide class of advertisers (“Plaintiffs”) filed this action against Meta, alleging that Meta had misrepresented the Potential Reach of advertisements on its platforms. Meta tells advertisers that “Potential Reach estimates how many people your ad could potentially reach depending on the targeting and ad placement options you select while creating an ad.” Each time that an advertiser designs a Meta advertising campaign, Meta’s self-service advertisement creation interface, known as the Ads Manager, displays the campaign’s Potential Reach. Plaintiffs assert that Potential Reach is misleading because it actually measures social media accounts, not living humans. Meta has taken steps to increase the accuracy of Potential Reach by working to remove fake and duplicate accounts, as well as by updating the calculation of Potential Reach to include only accounts that were shown an advertisement in the last thirty days. Nevertheless, throughout the class period, the number of accounts was always larger than the number of people because non-human entities like businesses and clubs have accounts, some people have multiple accounts, and some people and bots create fake accounts. 6 DZ RESERVE V. META PLATFORMS, INC. Each advertiser views a different Potential Reach for each campaign dependent on that campaign’s unique targeting criteria, so the discrepancy between people and accounts varies by campaign. The parties disagree as to the size of this discrepancy. The district court noted this evidentiary dispute but concluded that Meta’s criticism of Plaintiffs’ expert evidence “does not foreclose classwide proof of injury.” Plaintiffs allege that because of the misrepresentation of Potential Reach, they purchased more Meta advertisements and paid more for those advertisements than they would have with accurate information. The named Plaintiffs are two former Meta advertisers, DZ Reserve and Cain Maxwell. DZ Reserve was an e- commerce business that spent over $1 million on 740 Meta advertising campaigns. Maxwell operated an online firearm mount store and spent approximately $379 on 11 Meta advertising campaigns. DZ Reserve has ceased operations since the filing of the complaint, and it is unclear from the record whether Maxwell’s business is still operating. Following motion practice and the filing of several amended complaints, the district court sustained three of Plaintiffs’ claims under California state law: fraudulent misrepresentation, fraudulent concealment, and violation of California’s Unfair Competition Law (“UCL”). Plaintiffs then moved to certify the following class under Federal Rule of Civil Procedure 23: United States residents who purchased at least one advertisement on Meta’s platforms from August 15, 2014 to the present, excluding advertisers who used certain specialized purchasing methods or who were shown a Potential Reach lower than 1,000. The district court certified the class under Rule 23(b)(3) seeking damages for fraudulent misrepresentation and concealment,  DZ RESERVE V. META PLATFORMS, INC. 7 and under Rule 23(b)(2) seeking injunctive relief under the UCL. II We have jurisdiction pursuant to 28 U.S.C. § 1292(e) and Rule 23(f) of the Federal Rules of Civil Procedure. We review a district court’s decision to certify a class for abuse of discretion. Olean Wholesale Grocery Coop., Inc. v. Bumble Bee Foods LLC, 31 F.4th 651, 663 (9th Cir. 2022) (en banc). “A class certification order is an abuse of discretion if the district court applied an incorrect legal rule or if its application of the correct legal rule was based on a factual finding that was illogical, implausible, or without support in inferences that may be drawn from the facts in the record.” Van v. LLR, Inc., 61 F.4th 1053, 1062 (9th Cir. 2023) (internal quotation marks and citation omitted). “When reviewing an order granting class certification, we accord the district court noticeably more deference than when we review a denial.” Jabbari v. Farmer, 965 F.3d 1001, 1005 (9th Cir. 2020). “We review the district court’s determination of underlying legal questions de novo, and its determination of underlying factual questions for clear error.” Olean, 31 F.4th at 663 (citations omitted). III A Before certifying a class, the district court must ensure that the plaintiffs have made two showings, one under Rule 23(a) and one under Rule 23(b). Olean, 31 F.4th at 663. 8 DZ RESERVE V. META PLATFORMS, INC. First, the proposed class action must satisfy four prerequisites under Rule 23(a): (1) the class is so numerous that joinder of all members is impracticable; (2) there are questions of law or fact common to the class; (3) the claims or defenses of the representative parties are typical of the claims or defenses of the class; and (4) the representative parties will fairly and adequately protect the interests of the class. Fed. R. Civ. P. 23(a). The district court must perform a “rigorous analysis” of these prerequisites, which frequently “will entail some overlap with the merits of the plaintiff’s underlying claim.” Wal-Mart Stores, Inc. v. Dukes, 564 U.S. 338, 351 (2011). That being said, “[m]erits questions may be considered to the extent—but only to the extent—that they are relevant to determining whether the Rule 23 prerequisites for class certification are satisfied.” Amgen Inc. v. Connecticut Ret. Plans & Tr. Funds, 568 U.S. 455, 466 (2013). Second, the class must fit into at least one of three categories outlined in Rule 23(b). Olean, 31 F.4th at 663. Here, the district court certified the class under Rule 23(b)(3), which enables the potential recovery of damages and requires both that “questions of law or fact common to class members predominate over any questions affecting only individual members,” and that a class action be “superior to other available methods for fairly and efficiently adjudicating the controversy.” Fed. R. Civ. P. 23(b)(3). The  DZ RESERVE V. META PLATFORMS, INC. 9 district court also certified the class under Rule 23(b)(2), which requires “the party opposing the class has acted or refused to act on grounds that apply generally to the class, so that final injunctive relief or corresponding declaratory relief is appropriate respecting the class as a whole.” Fed. R. Civ. P. 23(b)(2). We address certification of the damages class under Rule 23(b)(3) and certification of the injunction class under Rule 23(b)(2) in turn. B We need not analyze all of the criteria required for certification of a damages class, because Meta challenges only the district court’s findings regarding the predominance of common factual or legal issues under Rule 23(b)(3) and typicality and adequacy of representation under Rule 23(a)(3) and (4). The district court did not abuse its discretion in concluding that Plaintiffs have sufficiently demonstrated predominance, typicality, and adequacy, and so we affirm certification of the damages class under Rule 23(b)(3). 1 The requirement under Rule 23(b)(3) that common questions predominate over individual ones “tests whether proposed classes are sufficiently cohesive to warrant adjudication by representation.” Amchem Prods., Inc. v. Windsor, 521 U.S. 591, 623 (1997). The predominance inquiry is “more demanding” than the commonality inquiry. Id. at 624. Contrary to Meta’s contentions, predominance is not more demanding because the common issues must in some way be “more common” than would be required under Rule 23(a)(2). Rather, predominance is more demanding because not only must 10 DZ RESERVE V. META PLATFORMS, INC. there be common issues, but the common issues must predominate. “The requirements of Rule 23(b)(3) overlap with the requirements of Rule 23(a): the plaintiffs must prove that there are questions of law or fact common to class members that can be determined in one stroke, in order to prove that such common questions predominate over individualized ones.” Olean, 31 F.4th at 664 (cleaned up). To clarify the inquiry, we proceed with the predominance analysis in three steps. First, we identify which questions are central to the plaintiffs’ claim. Second, we determine which of these questions are common to the class and which present individualized issues. Third, we analyze whether the common questions predominate over the individual questions. Under step one, we must identify which questions are central to the plaintiffs’ claim, which “begins, of course, with the elements of the underlying cause of action.” Erica P. John Fund, Inc. v. Halliburton Co., 563 U.S. 804, 809 (2011). The proposed class under Rule 23(b)(3) seeks damages for fraudulent concealment and fraudulent misrepresentation under California law, both of which require a showing of five elements: “(a) misrepresentation (false representation, concealment, or nondisclosure); (b) knowledge of falsity (or ‘scienter’); (c) intent to defraud, i.e. to induce reliance; (d) justifiable reliance; and (e) resulting damage.” Engalla v. Permanente Med. Grp., Inc., 15 Cal. 4th 951, 974 (1997), as modified (July 30, 1997) (internal quotation marks and citation omitted). Under step two, we determine which of those elements are “common”—which means they are “capable of being established through a common body of evidence, applicable to the whole class.” Olean, 31 F.4th at 666. Because this  DZ RESERVE V. META PLATFORMS, INC. 11 standard is identical to the analysis under Rule 23(a)(2)’s commonality requirement, “courts must consider cases examining both subsections in performing a Rule 23(b)(3) analysis.” Id. at 664. The district court properly determined that each of the five elements of fraud under California law is capable of classwide resolution. Meta has only legitimately challenged the district court’s findings regarding misrepresentation and justifiable reliance. On appeal, Meta does not dispute the district court’s conclusion that the knowledge and intent elements present common issues. Although Meta does appeal the district court’s damages finding, we decline to consider Meta’s damages argument because it was not raised before the district court. 1 Accordingly, we concentrate our analysis on the elements of misrepresentation and justifiable reliance. i Where, as in this case, a defendant has uniformly represented that a certain metric means something that it does not, the element of misrepresentation presents a common question. See In re Hyundai & Kia Fuel Econ. Litig., 926 F.3d 539, 557–65 (9th Cir. 2019) (en banc); In re First All. Mortg. Co. (First Alliance), 471 F.3d 977, 990–91 1 “[A]n issue will generally be deemed waived on appeal if the argument was not raised sufficiently for the trial court to rule on it.” Armstrong v. Brown, 768 F.3d 975, 981 (9th Cir. 2014) (internal quotation marks and citation omitted). Before the district court, Meta relied exclusively on criticisms of Plaintiffs’ experts’ damages modeling techniques and inputs. Meta’s argument on appeal is altogether different, as Meta now contends not that the model itself is deficient, but that it is not possible to use such a model at all. Because Meta did not raise this argument before the district court, we consider it waived. 12 DZ RESERVE V. META PLATFORMS, INC. (9th Cir. 2006); Blackie v. Barrack, 524 F.2d 891, 902–05 (9th Cir. 1975). Class action fraud claims often involve similar misrepresentations that cause a large number of victims to each suffer a small financial loss. Fraud claims are thus particularly well suited to class treatment under Rule 23(b)(3), which was designed “to overcome the problem that small recoveries do not provide the incentive for any individual to bring a solo action prosecuting his or her rights.” Amchem, 521 U.S. at 617 (quoting Mace v. Van Ru Credit Corp., 109 F.3d 338, 344 (7th Cir. 1997)). We have “consistently upheld” the availability of the class action to address mass frauds perpetrated through similar misrepresentations in the securities context “in large part because of the substantial role that the deterrent effect of class actions plays in accomplishing the objectives of the securities laws.” Blackie, 524 F.2d at 903. That reasoning applies equally well to consumer protection laws, and we have explained that consumer fraud victims often present a “cohesive group” because “[i]n many consumer fraud cases, the crux of each consumer’s claim is that a company’s mass marketing efforts, common to all consumers, misrepresented the company’s product . . . .” Hyundai, 926 F.3d at 559. In sum, “[p]redominance is a test readily met in certain cases alleging consumer or securities fraud . . . .” Amchem, 521 U.S. at 625. In determining whether a misrepresentation presents a common question, we generally categorize the misrepresentation as falling into one of two groups. On the one hand, a “fraud perpetrated on numerous persons by the use of similar misrepresentations may be an appealing situation for a class action . . . .” First Alliance, 471 F.3d at 990 (quoting Fed. R. Civ. P. 23, Advisory Committee Notes  DZ RESERVE V. META PLATFORMS, INC. 13 to 1966 Amendments, Subdivision (b)(3)). Accordingly, “this court has followed an approach that favors class treatment of fraud claims stemming from a ‘common course of conduct.’” Id. A “common course of conduct” refers to a defendant’s “centrally orchestrated strategy” to defraud, whereby “[e]ach plaintiff is similarly situated with respect to” that scheme. Id. at 991 (internal quotation marks and citation omitted). On the other hand, “a case may be unsuited for class treatment ‘if there was material variation in the representations made or in the kinds or degrees of reliance by the persons to whom they were addressed . . . .’” Id. at 990 (quoting Fed. R. Civ. P. 23, Advisory Committee Notes to 1966 Amendments, Subdivision (b)(3)). In this case, the claimed misrepresentation is the one that the district court described in its certification order: “[T]he ability of Potential Reach to reach ‘people,’ namely unique individuals” when the metric was “actually . . . an estimate of ‘accounts’ reached.” Meta misstates the misrepresentation at issue, insisting that the misrepresentation is the numerical discrepancy between people and accounts, rather than the fact that Meta substituted people for accounts. Under its theory, Meta contends the misrepresentations materially varied because the numerical value of the discrepancy differed for each individual advertiser based on its advertising budget and targeting, and thus there was no common misrepresentation among the class. We disagree. In Blackie, we rejected a similar strategy to create the illusion of variation in a claimed misrepresentation by mischaracterizing the nature of the misrepresentation at issue. See Blackie, 524 F.2d at n.20. There, a class of stockholders alleged that the Ampex Corporation uniformly 14 DZ RESERVE V. META PLATFORMS, INC. misapplied an accounting principle, which resulted in overstatements of various financial estimates. Id. at 902–05. Like Meta, Ampex argued that the misrepresentation was the numerical discrepancy in each financial estimate, such that there was material variation in the exact numerical discrepancies. Id. at 904 n.20. We rejected that argument and affirmed class certification, stating that “plaintiffs are complaining of abuses of accounting principles, not estimates.” Id. Likewise, we will not opine on the viability of Meta’s alternative misrepresentation theory—the numerical discrepancy between people and accounts— because it is not the theory presented to us. Meta’s insistence that the misrepresentation must be the numerical discrepancy between people and accounts is based partly on its suggestion that the substitution of people for accounts is not itself material. However, we have previously affirmed both class certification and ultimate liability based on similar facts. In First Alliance, we affirmed class certification and a finding of class-wide fraud where a bank induced borrowers to agree to unconscionable loan terms by having loan officers “point to the ‘amount financed’ and represent it as the ‘loan amount.’” See 471 F.3d at 985, 990– 92. We did not focus on the numerical difference between the amount financed and the loan amount for each individual borrower, but instead concluded that the overall scheme was fraudulent. Id. More importantly, proof of materiality “is not a prerequisite to class certification.” Amgen, 568 U.S. at 459. As the Supreme Court has instructed: Rule 23(b)(3) requires a showing that questions common to the class predominate, not that those questions will be answered, on  DZ RESERVE V. META PLATFORMS, INC. 15 the merits, in favor of the class. Because materiality is judged according to an objective standard, the materiality of [defendant’s] alleged misrepresentations and omissions is a question common to all members of the class [named plaintiffs] would represent. . . . As to materiality, therefore, the class is entirely cohesive: It will prevail or fail in unison. Id. at 459–60. Because materiality is an objective inquiry, differences in the size and sophistication of the advertisers in the class are irrelevant. Here, the question is the same for every class member: Would substituting people for accounts in Potential Reach be material to the reasonable consumer? At the class certification stage, identification of a common question is all that is required. The district court properly concluded that issue was a matter for trial. Given the claimed misrepresentation to be the substitution of people for accounts, Plaintiffs have clearly satisfied our “common course of conduct” test. It is undisputed that Potential Reach was shown to every advertiser on Meta’s Ads Manager, Potential Reach was always expressed as a number of people, and Potential Reach always estimated a number of accounts. Class members were thus exposed to uniform misrepresentations about the potential reach of their advertisements. Meta raises two additional arguments against a finding of Potential Reach estimates being a common misrepresentation. First, Meta disputes that the misrepresentation was uniform because Plaintiffs viewed 16 DZ RESERVE V. META PLATFORMS, INC. Potential Reach alongside other metrics, namely “Estimated Daily Reach.” While Potential Reach represents how many people meet a campaign’s targeting criteria, Estimated Daily Reach factors in an advertiser’s budget and past performance. These slight differences do not defeat commonality under our “common course of conduct” test. As we have previously explained, “[t]he class action mechanism would be impotent if a defendant could escape much of his potential liability for fraud by simply altering the wording or format of his misrepresentations across the class of victims.” First Alliance, 471 F.3d at 992. Consequently, “[c]onfronted with a class of purchasers allegedly defrauded over a period of time by similar misrepresentations, courts have taken the common sense approach that the class is united by a common interest in determining whether a defendant’s course of conduct is in its broad outlines actionable, which is not defeated by slight differences in class members’ positions . . . .” Blackie, 524 F.2d at 902 (collecting cases). We have consistently held that similar contextual differences do not constitute material variations. In Blackie, we held that there was commonality where defendants uniformly misapplied an accounting principle in some forty- five different documents, even though the resulting financial estimates fluctuated over time. Id. In First Alliance, we applied Blackie to hold that borrowers exposed to similarly misleading sales presentations represented a cohesive class, even though the exact wording of the sales presentations and individual loan specifics varied. First Alliance, 471 F.3d at 990–91. Most recently, we affirmed a class of car purchasers exposed to uniform fuel economy misrepresentations, even though some purchasers viewed the misrepresentations on stickers placed on the vehicles, while others were only  DZ RESERVE V. META PLATFORMS, INC. 17 exposed to the misrepresentations through nationwide marketing. Hyundai, 926 F.3d at 560–61. Here, the variations in Estimated Daily Reach and disclosures accompanying Potential Reach are no more material than the fluctuating estimates, differently worded sales pitches, and disparate modes of exposure considered in our prior cases. Second, Meta contends that any misrepresentations differed among class members because it updated its disclosures about Potential Reach twice during the class period. In September 2017, Meta disclosed that Potential Reach “[e]stimates are based on the placements and targeting criteria you select,” and are “not designed to match population or census estimates.” In June 2020, Meta disclosed that “[t]hese metrics are considered estimated and sampled, and depend on factors such as how many accounts are used by each person on Facebook Company Products.” We have determined that there were individualized questions where “explicit signs or explicit verbal advice would negate the claimed misrepresentation” for some class members. Berger v. Home Depot USA, Inc., 741 F.3d 1061, 1070 (9th Cir. 2014), abrogated on other grounds by Microsoft Corp. v. Baker, 582 U.S. 23 (2017). However, unlike the situation in Berger, none of the disclosures here negated the misrepresentation, which would have required a clear statement that Potential Reach measures accounts. Instead, Meta essentially argues that Plaintiffs should have known better than to rely on Potential Reach. But as the district court found, several documents offered by Plaintiffs show that Meta intended for advertisers to rely on its Potential Reach numbers. Thus, “[w]e find unpersuasive in this case the defense that plaintiffs should not have relied on 18 DZ RESERVE V. META PLATFORMS, INC. statements that were made with the fraudulent intent of inducing reliance.” First Alliance, 471 F.3d at 992. In support of its disclosure argument, Meta also relies on Mazza v. Am. Honda Motor Co., 666 F.3d 581 (9th Cir. 2012), overruled on other grounds by Olean, 31 F.4th 651. Disclosures were not at issue in Mazza. Instead, Mazza held that an inference of reliance was inappropriate because “it is likely that many class members were never exposed to the allegedly misleading advertisements.” Id. at 595. Unlike Mazza, here it is undisputed that all class members were exposed to Potential Reach. Given that all class members encountered the same misrepresentation about Potential Reach—the nucleus of the fraud—the slight variations in the other information available on the Ads Manager do not defeat the commonality of the misrepresentation. ii The district court properly determined that the element of justifiable reliance is capable of classwide resolution. Under California law, “when the same material misrepresentations have actually been communicated to each member of a class, an inference of reliance arises as to the entire class.” Mirkin v. Wasserman, 5 Cal. 4th 1082, 1095 (1993). Because Meta communicated the same misrepresentation to all class members—that Potential Reach measures people when it really measures accounts— the class is entitled to an inference of reliance. Meta’s argument to the contrary rests on its theory that Plaintiffs were not exposed to a uniform misrepresentation, which we have rejected.  DZ RESERVE V. META PLATFORMS, INC. 19 Despite California’s presumption of reliance, Meta argues that reliance is always an individualized inquiry because defendants have a right to rebut the presumption of reliance. As a practical matter, Meta’s argument that reliance can never be a common question is incompatible with the voluminous caselaw from both the United States and California Supreme Court certifying fraud class actions. See Basic Inc. v. Levinson, 485 U.S. 224, 242 (1988) (explaining the utility of the presumption of reliance in the federal security fraud context and stating that “[r]equiring proof of individualized reliance . . . effectively would have prevented respondents from proceeding with a class action . . . .”); see also Vasquez v. Superior Ct., 4 Cal. 3d 800, 814– 15 (1971) (discussing California’s presumption of reliance for common law fraud and analogizing to the presumption in federal securities fraud cases). The purpose of the presumption of reliance is to avoid precluding all fraud class actions. See Basic, 485 U.S. at 242. Accordingly, if the availability of rebuttal defeated commonality, the presumption would be pointless. While rebuttal “has the effect of leaving individualized questions of reliance in the case, there is no reason to think that these questions will overwhelm common ones and render class certification inappropriate under Rule 23(b)(3).” Halliburton Co. v. Erica P. John Fund, Inc., 573 U.S. 258, 276 (2014) (internal quotation marks and citation omitted). Meta finally argues that the Rules Enabling Act prohibits application of California’s presumption of reliance here. The Rules Enabling Act instructs that rules of procedure “shall not abridge, enlarge or modify any substantive right.” 28 U.S.C. § 2072(b). Meta argues that application of the presumption of reliance amounts to lessening a plaintiff’s burden of proving reliance in a class action case. However, 20 DZ RESERVE V. META PLATFORMS, INC. California’s presumption of reliance also applies in individual fraud actions. Engalla, 15 Cal. 4th at 977. Failing to apply the presumption of reliance would thus amount to abridging a substantive right, as the presumption would apply in individual cases but not in federal class actions. Contrary to Meta’s contention, the Rules Enabling Act requires application of California’s presumption of reliance. Because the presumption of reliance applies to each member of the class, reliance presents a common question provable by common evidence. See Vasquez, 4 Cal. 3d at 814 (“If [Plaintiffs] can establish without individual testimony that the representations were made to each plaintiff and that they were false, it should not be unduly complicated to sustain their burden of proving reliance thereon as a common element.”). iii Having arrived at step three, our analysis in this case is a simple one. “The predominance inquiry asks whether the common, aggregation-enabling, issues in the case are more prevalent or important than the non-common, aggregation- defeating, individual issues.” Tyson Foods, Inc. v. Bouaphakeo, 577 U.S. 442, 453 (2016) (internal quotation marks and citation omitted). Although predominance does not require that all questions be common, Hyundai, 926 F.3d at 557, predominance is necessarily satisfied if all questions are common. Because the district court properly concluded that each of the five elements of fraud presents a common question, the district court did not abuse its discretion in holding that common issues predominated.  DZ RESERVE V. META PLATFORMS, INC. 21 2 Meta argues the named Plaintiffs are not typical or adequate because they suffer from credibility problems that expose them to individualized defenses related to reliance. The district court did not clearly err in finding that the named Plaintiffs’ credibility was not vulnerable to attack. Accordingly, we affirm the district court’s holding that the requirements of typicality and adequacy are satisfied. Although Meta names both typicality and adequacy in its argument, its contention that Plaintiffs will be preoccupied with unique defenses falls within our typicality caselaw. Under Federal Rule of Civil Procedure Rule 23(a), class plaintiffs must demonstrate, among other things, that the named plaintiffs are typical class representatives. See Olean, 31 F.4th at 663 (citing Fed. R. Civ. P. 23(a)). “Under the rule’s permissive standards, representative claims are ‘typical’ if they are reasonably co-extensive with those of absent class members; they need not be substantially identical.” Hanlon v. Chrysler Corp., 150 F.3d 1011, 1020 (9th Cir. 1998), overruled on other grounds by Wal-Mart, 564 U.S. at 338 (quoting Fed. R. Civ. P. 23(a)(3)). A named plaintiff is not typical if “there is a danger that absent class members will suffer if their representative is preoccupied with defenses unique to it.” Hanon v. Dataproducts Corp., 976 F.2d 497, 508 (9th Cir. 1992) (quoting Gary Plastic Packaging Corp. v. Merrill Lynch, Pierce, Fenner & Smith, Inc., 903 F.2d 176, 180 (2d Cir. 1990)). We will affirm a district court’s typicality determination if “[t]he district court did not commit a clear error of judgment in concluding that . . . [the named plaintiff] would not be subject to unique defenses such that typicality would be defeated . . . .” Just Film, Inc. v. Buono, 847 F.3d 1108, 1120 (9th Cir. 2017). 22 DZ RESERVE V. META PLATFORMS, INC. The district court did not clearly err in finding no danger that the named Plaintiffs would be preoccupied with unique defenses. Meta insists that the named Plaintiffs are not typical because, unlike other class members, neither named Plaintiff actually relied on the Potential Reach estimates. We have “emphasize[d] that the defense of non-reliance is not a basis for denial of class certification” and reliance is more appropriately considered at the merits stage. Hanon, 976 F.2d at 509. Even so, the record supports the district court’s finding at the certification stage that the named Plaintiffs relied on Meta’s misrepresentations. Meta argues that DZ Reserve’s owner dishonestly testified that the Potential Reach misrepresentation deterred him from buying Meta advertisements, and that Maxwell dishonestly claimed to have relied on Potential Reach. The district court rejected these contentions by pointing to evidence that DZ Reserve had been deterred from using Meta advertisements, Maxwell relied on Potential Reach, and both named Plaintiffs would have spent less money on Meta advertisements had they known that Potential Reach was a misrepresentation. The record supports the district court’s conclusion that the named Plaintiffs have no credibility issues that would destroy their typicality. Even if DZ Reserve and Maxwell faced credibility questions, those issues would not destroy typicality. Credibility issues only destroy typicality in “unique situation[s]” where “it is predictable that a major focus of the litigation will be on a defense unique” to the named plaintiff. Id. at 509. We have found such unique situations where a named plaintiff in a securities action was a serial litigant who purchased stock solely to facilitate litigation, id. at 508, or where the named plaintiff insisted that he was not really deceived by the alleged misrepresentation. Stearns v.  DZ RESERVE V. META PLATFORMS, INC. 23 Ticketmaster Corp., 655 F.3d 1013, 1019 (9th Cir. 2011), abrogated on other grounds by Comcast Corp. v. Behrend, 569 U.S. 27 (2013). Neither of those situations apply here, where the named Plaintiffs are not serial litigants and presented evidence that they both actually received and relied upon the alleged misrepresentation. 3 In sum, for the foregoing reasons, we conclude that the district court did not abuse its discretion in certifying the damages class under Rule 23(b)(3). C Meta appeals the district court’s order certifying an injunction class under Rule 23(b)(2) on the basis that the named Plaintiffs lack Article III standing to seek injunctive relief under California’s UCL. Meta did not present this theory before the district court. However, an objection that a federal court lacks subject-matter jurisdiction “may be raised by a party, or by a court on its own initiative, at any stage in the litigation, even after trial and the entry of judgment.” Arbaugh v. Y&H Corp., 546 U.S. 500, 506 (2006). As we explain below, DZ Reserve did not submit any evidence that would support its standing to seek injunctive relief. However, Maxwell’s standing is a closer call and may require additional factual development. Therefore, we remand the question of Maxwell’s standing to seek injunctive relief to the district court for its consideration in the first instance. “In a class action, standing is satisfied if at least one named plaintiff meets the requirements.” Bates v. United Parcel Serv., Inc., 511 F.3d 974, 985 (9th Cir. 2007). In order to establish Article III standing, “the plaintiff must 24 DZ RESERVE V. META PLATFORMS, INC. have suffered an injury in fact—a concrete and imminent harm to a legally protected interest, like property or money—that is fairly traceable to the challenged conduct and likely to be redressed by the lawsuit.” Biden v. Nebraska, 600 U.S. __, 143 S.Ct. 2355, 2365 (2023) (citing Lujan v. Defenders of Wildlife, 504 U.S. 555, 560–61 (1992)). “[A] plaintiff must demonstrate standing separately for each form of relief sought.” Friends of the Earth, Inc. v. Laidlaw Env’t Servs. (TOC), Inc., 528 U.S. 167, 185 (2000). Thus, the fact that the named Plaintiffs have standing to seek damages does not mean that they automatically have standing to seek injunctive relief. See City of Los Angeles v. Lyons, 461 U.S. 95, 105–06 (1983); see also TransUnion LLC v. Ramirez, 594 U.S. 413, 436 (2021) (“[A] plaintiff’s standing to seek injunctive relief does not necessarily mean that the plaintiff has standing to seek retrospective damages.”). In order to establish standing for injunctive relief, “a plaintiff must show that he is under threat of suffering ‘injury in fact’ that is concrete and particularized; the threat must be actual and imminent, not conjectural or hypothetical; it must be fairly traceable to the challenged action of the defendant; and it must be likely that a favorable judicial decision will prevent or redress the injury.” Summers v. Earth Island Inst., 555 U.S. 488, 493 (2009) (citing Friends of Earth, 528 U.S. at 180–81). “The plaintiff must demonstrate that he has suffered or is threatened with a concrete and particularized legal harm, coupled with a sufficient likelihood that he will again be wronged in a similar way.” Bates, 511 F.3d at 985 (citations and quotation marks omitted). “Past exposure to harmful or illegal conduct does not necessarily confer standing to seek injunctive relief if the plaintiff does not continue to suffer adverse effects.” Mayfield v. United  DZ RESERVE V. META PLATFORMS, INC. 25 States, 599 F.3d 964, 970 (9th Cir. 2010). “Nor does speculation or ‘subjective apprehension’ about future harm support standing.” Id. (quoting Friends of the Earth, 528 U.S. at 184 and citing Lujan, 504 U.S. at 564). Consumer fraud plaintiffs can satisfy the imminent injury requirement by showing they “will be unable to rely on the product’s advertising or labeling in the future, and so will not purchase the product although [they] would like to.” Davidson v. Kimberly-Clark Corp., 889 F.3d 956, 970 (9th Cir. 2018). The plaintiff bears the burden of establishing the elements of standing. See Lujan, 504 U.S. at 561. A plaintiff must also demonstrate Article III standing at each stage of the litigation, including on appeal. Bain v. Cal. Teachers Ass’n, 891 F.3d 1206, 1211–12 (9th Cir. 2018). Standing must be proven, “with the manner and degree of evidence required at the successive stages of the litigation.” Lujan, 504 U.S. at 561. Thus, although standing may be established at the pleading stage through allegations in the complaint, the plaintiff must prove the elements of standing at each successive stage. Id. Because the preponderance of the evidence standard applies at the class certification stage, standing at the time of class certification must be established by a preponderance of the evidence. See Olean, 31 F.4th at 664–65. With these general principles in mind, we examine the standing of the named Plaintiffs to assert claims for injunctive relief. 1 DZ Reserve does not have standing to seek injunctive relief. DZ Reserve did not submit any evidence of a threat 26 DZ RESERVE V. META PLATFORMS, INC. of suffering “actual and imminent” future injury that was concrete and particularized, and that could be redressed by injunctive relief. Nor did DZ Reserve demonstrate a sufficient likelihood that it would again be wronged in a similar way. Rather, the owner of DZ Reserve simply testified that he would have spent less on Meta advertisements in the past had he known the truth about Potential Reach. He did not testify about his desire to purchase Meta advertisements in the future. Further, as we have noted, DZ Reserve is no longer operating as a business. Thus, DZ Reserve lacks standing to assert a claim of injunctive relief. 2 We remand the question of whether Maxwell has adequately pled an injury sufficient to confer standing to seek injunctive relief. In so doing, we note that there are two issues for the district court to consider. The first question is whether Maxwell’s testimony that he “think[s] [he] would” purchase Meta advertisements in the future satisfies Davidson, which relied on a plaintiff’s more direct assertion that she “desires to purchase” and “would purchase” a product if she was able to trust the product’s advertising. 889 F.3d at 970–71. The second question is how to square Maxwell’s testimony with the evidence suggesting that Maxwell no longer has a business to advertise. A plaintiff typically loses standing to challenge a policy affecting businesses when the plaintiff has ceased operating an affected business, unless the challenged policy caused the business’s closure. See City News & Novelty, Inc. v. City of Waukesha, 531 U.S. 278, 283 (2001); see also San Lazaro Ass’n, Inc. v. Connell, 286 F.3d 1088, 1096 (9th Cir. 2002); Clark v. City of Lakewood, 259  DZ RESERVE V. META PLATFORMS, INC. 27 F.3d 996, 1007–08 (9th Cir. 2001). Maxwell’s business ceased operations sometime in 2019. He testified that he stopped operations because he “ran out of inventory.” The record does indicate that Maxwell has not officially dissolved the business and that his associated tax ID remains active. The record does not indicate whether Maxwell has continued to pay taxes associated with the business. It will be difficult for Maxwell to establish an imminent injury if he has no business to advertise, or in the alternative, if he does not offer a compelling explanation for why he would purchase advertisements without a business. The district court has had no occasion to consider the record or to analyze Meta’s argument against Maxwell’s standing to seek injunctive relief. Moreover, Maxwell did not have the opportunity to present arguments concerning his standing to seek injunctive relief directly to the district court. Therefore, we remand the question of standing to seek injunctive relief to the district court for its consideration in the first instance. D In sum, we affirm the district court’s certification of the damages class. We vacate the district court’s certification of the injunction class and remand for further proceedings consistent with this opinion. Each party should bear its own costs on appeal. AFFIRMED in part; VACATED AND REMANDED in part. 28 DZ RESERVE V. META PLATFORMS, INC. FORREST, J., dissenting in part: I agree that the district court’s certification of Plaintiffs’ Rule 23(b)(2) injunction class must be vacated and remanded for the district court to reconsider whether Plaintiff Cain Maxwell has standing to pursue that claim. I disagree, however, that the district court properly certified Plaintiffs’ Rule 23(b)(3) damages class because Plaintiffs cannot satisfy the predominance requirement where there are individual questions that must be answered related to multiple elements of Plaintiffs’ fraud-based claims. Therefore, I respectfully dissent in part. I. BACKGROUND Defendant-Appellant Meta Platforms, Inc. (Meta), one of the world’s largest social media companies, owns and operates Facebook and Instagram, among other platforms. Meta claims that more than two billion people use Facebook every month, with over 200 million monthly active users in the United States alone. Because of its large user base, Meta’s platforms are attractive to prospective advertisers, ranging from Fortune 500 companies and government agencies to small businesses and individual proprietors. And Meta “generates substantially all of its revenue from advertising.” A. Meta’s Advertising System Most advertisers purchase ads from Meta through its online self-service ad creation interface, known as “Ads Manager.” Advertisers “have a wide range of different advertising objectives, which influences how they set up their ads and assess ad performance.” When developing an ad campaign in Ads Manager, advertisers specify their objective. Advertisers who want to generate awareness of  DZ RESERVE V. META PLATFORMS, INC. 29 their product or service, and who want Meta to show their ad “the largest number of times to the largest number of people in a given audience,” may choose “brand awareness” or “reach” as their advertising objective. Other advertisers “are interested in ‘performance advertising,’ or driving specific actions with their ads, such as clicks and conversions”—i.e., prompting users to visit a website or purchase a product. These advertisers “are typically focused on trying to identify or have their ads delivered to specific users likely to take a desired action,” and a large audience size is less important. Ads Manager provides several planning tools to help advertisers design their ad campaigns and target their desired audience. First (and relevant here) is Potential Reach, which is defined as “an estimation of how many people are in an ad set’s target audience” based on statistical sampling and modeling. A default Potential Reach automatically displays in Ads Manager, and it updates dynamically in real time as an advertiser tailors its ad campaign using numerous targeting and placement criteria, such a demographics (e.g., age, gender, location, education), interests (e.g., sports teams, dogs), and the platform where the ads will be shown (Facebook, Instagram, etc.). The “default” Potential Reach displayed to each advertiser during the class period was between 200 to 250 million, purportedly reflecting the number of people in the United States between 18 and 65 years old who use Meta’s platforms. As an advertiser selects targeting criteria, the Potential Reach recalculates, and a color-coded dial shows whether the target audience is “fairly broad,” “defined,” or “too specific.” Each advertiser sees a different Potential Reach estimate for each ad campaign they run because the non-default—or targeted—Potential Reach estimate is calculated based on the advertiser’s selected criteria. 30 DZ RESERVE V. META PLATFORMS, INC. Potential Reach is “not an estimate of how many people will actually see [an advertiser’s] ad” or how many people may click on an ad or take any other action with respect to an ad. That data is provided in separate Estimated Daily Results metrics, which are displayed adjacent to Potential Reach in Ads Manager. Estimated Daily Reach is part of the Estimated Daily Results and is the estimated number of people that an ad actually will reach per day based on the advertiser’s selected criteria, budget, and past ad performance. Advertisers are not charged based on the Potential Reach calculation. Once an ad launches, advertisers can track their results in real time. Based on detailed performance data, such as the number of times an ad was shown and clicked on, advertisers can assess the success of their campaign and return on investment and adjust their campaign and budget as they see fit. Advertisers are not shown Potential Reach as part of the post-ad purchase results. B. Changes to Potential Reach Calculation Potential Reach has always been displayed to advertisers as an estimate, but during the class period Meta changed how it calculates Potential Reach and updated its disclosures in Ads Manager accordingly. In September 2017, Meta introduced an “information” icon in Ads Manager explaining that Potential Reach “[e]stimates are based on the placements and targeting criteria you select,” and are “not designed to match population or census estimates.” A year and a half later in March 2019, Meta changed its calculation methodology to count only those people who had actually seen an ad on Meta’s platforms in the last 30 days, rather than those who were active on a Meta platform and could have seen an ad. Lastly, in June 2020, Meta “added  DZ RESERVE V. META PLATFORMS, INC. 31 disclosures to explain that ‘people’ is an ‘estimated and sampled’ metric, which depends on ‘factors such as how many accounts are used by each person on [Meta’s products].” Throughout the class period, Meta also undertook efforts to remove fake accounts and de-duplicate accounts across platforms—i.e., counting separate Instagram and Facebook accounts belonging to the same person as only one person in Potential Reach estimates. C. Plaintiffs’ Lawsuit In 2018, Plaintiffs sued Meta alleging the Potential Reach calculation is materially misleading because it exceeds the actual number of people in an ad’s target audience, causing advertisers to purchase more ads and pay higher prices for ads than they otherwise would have. Named Plaintiffs DZ Reserve, an e-commerce business, and Cain Maxwell are former Meta advertisers. Between December 2017 and December 2018, DZ Reserve spent over $1 million on 740 ad campaigns comprising approximately 26,000 ads. Maxwell (d/b/a Max Martialis) operated an online store and spent approximately $400 on 11 ad campaigns comprising 28 ads between September 2018 and May 2019. Named Plaintiffs alleged that they viewed and relied on Potential Reach in purchasing Meta ads. Plaintiffs proceeded on three California state-law claims: (1) fraudulent misrepresentation, (2) fraudulent concealment, and (3) injunctive relief under California’s Unfair Competition Law (UCL). And they sought to certify a class related to each claim encompassing the millions of advertisers (persons or entities) in the United States who paid to place at least one ad on Meta’s platforms from August 2014 to the present. Plaintiffs asserted that Potential Reach is a material misrepresentation because Meta 32 DZ RESERVE V. META PLATFORMS, INC. characterizes it as a calculation of “people,” which Meta knows is inaccurate because it is a calculation of accounts, and because Potential Reach is always significantly more than the number of people. Plaintiffs further claimed that the inflation of the Potential Reach calculation is susceptible to proof through common evidence because their statistics expert, Dr. Charles Cowan, established that the default Potential Reach shown to advertisers is always inflated by at least 33% and the targeted Potential Reach is always inflated by at least 10%. Through a conjoint survey, Plaintiffs’ expert, Dr. Greg Allenby, further determined that Potential Reach inflation, as found by Dr. Cowan, has “a statistically significant impact on consumer demand for [Meta] advertisements.” Meta opposed class certification, arguing, among other things, that Plaintiffs could not satisfy Rule 23(b)(3)’s predominance requirement because each class member received a fundamentally different Potential Reach estimate and the class members varied in multiple ways that are material to whether the elements of Plaintiffs’ claims can be met—including the varying disclosures that advertisers may have viewed, the advertisers’ objectives for their ad campaign, and the mix of information each advertiser had access to or relied on in purchasing ads. Over Meta’s objection, the district court certified two classes: a Rule 23(b)(3) damages class for Plaintiffs’ common law fraud claims and a Rule 23(b)(2) class for Plaintiffs’ UCL injunction claim. 1 The district court evaluated Rule 23(a)’s threshold commonality requirement 1 The class includes only those who purchased ads through Ads Manager under Meta’s standard contract, and for which Meta provided a Potential Reach of 1,000 of greater.  DZ RESERVE V. META PLATFORMS, INC. 33 and Rule 23(b)(3) predominance requirement “in tandem.” It determined that all class members were exposed to a similar misrepresentation and that “whether Meta made misrepresentations to all class members [could] be shown through common evidence” because Potential Reach was represented as an estimate of “people” when it really was “an estimate of ‘accounts,’” and “the number of unique accounts and unique people were different.” It further reasoned that materiality and reliance do “not necessarily undermine predominance” in fraud cases because, under California law, a “presumption, or at least an inference, of reliance arises wherever there is a showing that a misrepresentation was material.” Rather, materiality and reliance could be established “through common evidence” because “Potential Reach metrics were shown to all advertisers,” it was “an important number for advertisers,” and “[a] majority of advertisers rely on Potential Reach as a metric for their advertisements.” This court granted review of the district court’s class certification decision under Federal Rule of Civil Procedure 23(f). II. ANALYSIS We review the district court’s class certification decision for abuse of discretion. Olean Wholesale Grocery Coop., Inc. v. Bumble Bee Foods LLC (Olean), 31 F.4th 651, 663 (9th Cir. 2022) (en banc). The district court “abuses its discretion only if it (1) relies on an improper factor, (2) omits a substantial factor, or (3) commits a clear error of judgment in weighing the correct mix of factors.” B.K. ex rel. Tinsley v. Snyder, 922 F.3d 957, 965 (9th Cir. 2019). The district court’s determination of underlying legal questions is reviewed de novo, and its determination of underlying 34 DZ RESERVE V. META PLATFORMS, INC. factual questions is reviewed for clear error. Olean, 31 F.4th at 663. “An error of law is a per se abuse of discretion.” B.K., 922 F.3d at 965. A factual finding is clearly erroneous if it is illogical, implausible, or “without support in inferences that may be drawn from the record.” Id. at 965–66. Federal Rule of Civil Procedure 23 governs class certification. Plaintiffs, as the party seeking class certification, bear the burden of demonstrating that the requirements of Rule 23 are satisfied. Stromberg v. Qualcomm Inc., 14 F.4th 1059, 1066 (9th Cir. 2021). “As a threshold matter, a class must first meet the four requirements of Rule 23(a): (1) numerosity, (2) commonality, (3) typicality, and (4) adequacy of representation.” Id. Additionally, “the class must meet the requirements of at least one of the ‘three different types of classes’ set forth in Rule 23(b).” Id. (citation omitted). Relevant here, Plaintiffs must satisfy the requirements of Rule 23(b)(3) for a damages class. Certification under this provision is appropriate only where “questions of law or fact common to class members predominate over any questions affecting only individual members.” Fed. R. Civ. P. 23(b)(3). The goal of Rule 23(b)(3) is well-established—by adding the predominance (and superiority) requirements, the Advisory Committee intended to “achieve economies of time, effort, and expense, and promote . . . uniformity of decision as to persons similarly situated, without sacrificing procedural fairness.” Amchem Prod., Inc. v. Windsor, 521 U.S. 591, 615 (1997) (emphasis added) (quoting Advisory Committee Notes). Thus, predominance is established where the proposed class is “sufficiently cohesive” to justify class- wide adjudication. Id. at 623. This required cohesion exists where there are common questions capable of class-wide resolution. Olean, 31 F.4th at 663. Or, stated another way,  DZ RESERVE V. META PLATFORMS, INC. 35 where common questions of law or fact “can be determined in one stroke.” Id. at 664. Conversely, individual questions dominate where evidence will inevitably vary from class member to class member. Id. Plaintiffs must establish that the preponderance requirement is met by a preponderance of the evidence. Id. at 665. Rule 23 “does not set forth a mere pleading standard,” but instead requires that the district court conduct “a rigorous analysis” to ensure that the party seeking certification has satisfied its burden “through evidentiary proof.” Comcast Corp. v. Behrend, 569 U.S. 27, 33 (2013). “Such an analysis will frequently entail overlap with the merits of the plaintiff’s underlying claim . . . because the class determination generally involves considerations that are enmeshed in the factual and legal issues comprising the plaintiff’s cause of action.” Id. at 33–34 (citations and internal quotation marks omitted). And the Supreme Court has instructed that “[i]f anything, Rule 23(b)(3)’s predominance criterion is even more demanding than Rule 23(a).” Comcast, 569 U.S. at 34. In considering predominance, the court begins “with the elements of the underlying cause of action.” Olean, 31 F.4th at 665. Plaintiffs must show that a common question relating to an essential element predominates. Id. at 666. A class may fail to establish predominance where even one essential element requires individualized determination and this individualized issue outweighs “common, aggregation- enabling issues.” See Lara v. First Nat’l Ins. Co. of Am., 25 F.4th 1134, 1138 (9th Cir. 2022). The district court certified a Rule 23(b)(3) class for Plaintiffs’ fraudulent misrepresentation and fraudulent concealment claims. Under California law, the elements of these claims are: “(1) a misrepresentation (false representation, concealment, or 36 DZ RESERVE V. META PLATFORMS, INC. nondisclosure); (2) knowledge of falsity (or scienter); (3) intent to defraud, i.e., to induce reliance; (4) justifiable reliance; and (5) resulting damage.” Robinson Helicopter Co. v. Dana Corp., 34 Cal. 4th 979, 990 (2004). A plaintiff may rely on a presumption of reliance, but “only [by making] a showing that the misrepresentations were material.” Engalla v. Permanente Med. Group, Inc., 15 Cal 4th 951, 977 (1997), as modified (July 30, 1997). The plaintiffs’ ability to prove each element of their claim must be considered in light of the class-action mechanism, which often is ill-suited to fraud claims. As the 1966 Advisory Committee on Rule 23 notes, even where a “common core” exists in a fraud case, it nonetheless “may be unsuited for treatment as a class action if there was material variation in the representation made or in the kinds or degrees of reliance by the persons to whom they were addressed.” 2 Fed. R. Civ. P. 23, Advisory Committee Notes to 1966 Amendments, Subdivision (b)(3). Here, Plaintiffs failed to establish predominance because there are three 2 The majority asserts that fraud claims are “particularly well suited to class treatment under Rule 23(b)(3).” Maj. Op. 12. The majority references a passing statement from Amchem, which states that predominance may be “readily met in certain cases” of consumer fraud. See 521 U.S. at 625 (emphasis added). The majority ignores, however, the rest of the paragraph from which it quotes, which specifically cautioned courts to heed the “[Rule 23 Advisory] Committee’s warning, [which] continues to call for caution when individual stakes are high and disparities among class members great.” Id. The majority’s statement that fraud claims are “particularly well suited” for class treatment runs in the face of the Committee’s cautionary understanding that our sister circuits have consistently recognized. See, e.g., In re St. Jude Med., Inc., 522 F.3d 836, 838 (8th Cir. 2008) (noting the “difficulty with class treatment of cases alleging fraud or misrepresentation”); Moore v. PaineWebber, Inc., 306 F.3d 1247, 1253 (2d Cir. 2002) (same).  DZ RESERVE V. META PLATFORMS, INC. 37 issues that involve individualized questions: (1) whether each advertiser in the class was subject to a misrepresentation, (2) whether any misrepresentation was material, and (3) whether each advertiser relied on a material misrepresentation. I address each in turn. A. Misrepresentation To assess whether predominance is satisfied regarding the misrepresentation element, we must first be specific in identifying Plaintiffs’ claimed misrepresentation. On appeal, Plaintiffs argue that Meta categorically misrepresented its Potential Reach metric presented to advertisers by characterizing it as a metric of “people” rather than “accounts.” The majority accepts this characterization of Plaintiffs’ claims. But Plaintiffs’ complaint alleged that Meta failed to provide “accurate Potential Reach” because this calculation “is inflated.” Core to Plaintiffs’ claims is the degree of discrepancy between the number of people and the number of accounts (not just the characterization of Potential Reach as a calculation of people), which the Plaintiffs explicitly attempt to prove. 3 Meta does not dispute that Potential Reach calculated accounts as a proxy for people. But contrary to the majority’s suggestion, this proxy is not inherently misleading like the 3 If Plaintiffs’ claimed misrepresentation rested solely on the description of Potential Reach as a calculation of people, there would have been no need for Plaintiffs to submit statistical evidence regarding the degree of inflation of the Potential Reach calculation. The analysis could have been merely definitional—especially given that Meta does not dispute it used accounts as a proxy for people. And the district court recognized that Plaintiffs’ theory was more than merely definitional, stating: “Potential Reach was always expressed as a number of ‘people,’ and the discrepancy between people and accounts made the number inaccurate.”. 38 DZ RESERVE V. META PLATFORMS, INC. accounting practices challenged in Blackie v. Barrack, 524 F.2d 891 (9th Cir. 1975), discussed more below. Potential Reach as described is misleading only if there is a significant deviation between the number of accounts and the number of people that may see ads. 4 If these two populations neatly correlate, characterizing Potential Reach as a calculation of people is accurate. Moreover, whether a deviation between the number of accounts and the number of people is a misrepresentation must consider Meta’s express disclosure that Potential Reach is an estimate. Cf. Estimate, Merriam- Webster (defining “estimate” as “a rough or approximate calculation”), https://www.merriam- webster.com/dictionary/estimate?utm_campaign=sd&utm_ medium=serp&utm_source=jsonld; Estimate, Oxford English Dictionary (defining “estimate” as “an approximate notion of (the amount, number, magnitude, or position of anything) without actual enumeration or measurement”), https://www.oed.com/dictionary/estimate_v?tab=meaning_ and_use#5272337. On this point, the district court erred by reasoning that any variation between accounts and people was a misrepresentation. Properly framed, Plaintiffs’ assertion that there is a cohesive class for which common questions predominate begins to unravel. Consistent with the Rule 23 Advisory Committee’s admonishment that fraud claims are not well suited for class treatment, we have upheld class certification of these kinds of claims in limited circumstances where the 4 The relevant metric to whether a misrepresentation occurred is the Potential Reach after targeting. Only 1.2% of U.S. ads were purchased with numbers near the default Potential Reach of 200-250 million. Additionally, while the district court mentioned default Potential Reach in analyzing typicality, it did not rely on default Potential Reach in analyzing predominance for the misrepresentation element.  DZ RESERVE V. META PLATFORMS, INC. 39 misrepresentations stemmed from a “common course of conduct” or “centrally-orchestrated scheme.” In re First All. Mortg. Co., 471 F.3d 977, 990–91 (9th Cir. 2006). The substance of the misrepresentation must be sufficiently uniform to prove fraud on a class-wide basis. See In re Hyundai & Kia Fuel Econ. Litig., 926 F.3d 539, 560 (9th Cir. 2019) (approving a class certification where class members were uniformly exposed to a nationwide advertising campaign that gave “uniform fuel-economy misrepresentations”); see also In re First All., 471 F.3d at 990 (“The required degree of uniformity among misrepresentations in a class action for fraud is a question of law . . . .”). If the challenged communication is not sufficiently uniform, then whether a material misrepresentation occurred depends on individual questions specific to each class member. See Berger v. Home Depot USA, Inc., 741 F.3d 1061, 1069 (9th Cir. 2014), abrogated on other grounds by Microsoft Corp. v. Baker, 582 U.S. 23 (2017). Plaintiffs assert that Meta made a common misrepresentation that fits “comfortably” within the “common course of conduct” principle, first established in Blackie, and applied again in First Alliance, because Meta uniformly represented that Potential Reach was a measurement of people. This argument is unavailing. At issue in First Alliance was a challenge to certification based on Rule 23(a)(2)’s commonality requirement, not Rule 23(b)(3)’s predominance requirement. For commonality, we declined to adopt a “talismanic rule” that requires “representations [to be] all but identical.” In re First All., 471 F.3d at 991 (quoting In re Am. Cont’l Corp./Lincoln Sav. & Loan Sec. Litig., 140 F.R.D. 425, 430 (D. Ariz. 1992)). Rather, we held that in a common-course-of-conduct 40 DZ RESERVE V. META PLATFORMS, INC. analysis, courts must determine whether the “center of gravity” of the fraud overshadows any variations in individual misrepresentations across the class. Id. at 991. Where there are immaterial variations, a common course of conduct can compensate for reduced uniformity. See id. at 990. The center of gravity in First Alliance was a standardized protocol to induce fraud: sales agents were “carefully trained” in a “standardized training program” requiring memorization of a specific sales pitch and “strict adherence to a specific method of hiding information.” Id. In Blackie, we analyzed whether financial reports that “uniformly misrepresent a particular item” presented a common question, again for purposes of the commonality requirement. 524 F.2d at 903. Plaintiffs cite Blackie’s commonality analysis to support their predominance argument. 5 But the difference between the commonality and predominance analyses and the factual differences between Blackie and this case are key. The Blackie plaintiffs argued that 45 financial documents fraudulently inflated a stock price. Id. at 902. The court found a common misrepresentation based on the “unique situation of the accounting and legal principles” at play and that “financial reports throughout the period uniformly and fraudulently” failed to adhere to “accepted accounting principles” in a manner that “injur[ed] all purchasers.” 6 Id. at 904. The court 5 The Blackie court did not analyze predominance for the misrepresentation element, only for reliance, causation, and damages. 524 F.2d at 905–08. 6 The commonality and predominance analyses in Blackie were informed by the “flexibl[e]” context of securities fraud laws. 524 F.2d at 907; see also id. at 903 n.19; cf. Sullivan v. Chase Inv. Servs. of Bos., Inc., 79 F.R.D. 246, 259 (N.D. Cal. 1978) (citing Blackie to support the proposition that “common questions generally predominate in securities  DZ RESERVE V. META PLATFORMS, INC. 41 further noted that “plaintiffs are complaining of abuses of accounting principles, not estimates.” Id. at 904 n.20. But this case is about estimates. Meta represented to the advertiser class that Potential Reach is an estimate of people who could potentially view a given ad based on the advertiser’s targeting criteria. That Meta provided a common description of Potential Reach does not automatically establish that this description was a misrepresentation as to all class members. Cf. Blackie, 524 F.2d at 903 n.19 (noting that even where a common course of conduct exists in a fraud class that satisfies Rule 23(a)(2) commonality, the predominance requirement may not be satisfied). Predominance requires more than a common but superficial thread connecting class members—this may be shown where class claims “prevail or fail in unison.” Amgen Inc. v. Conn. Ret. Plans & Tr. Funds, 568 U.S. 455, 460 (2013). And given the context here, whether Meta’s characterization of Potential Reach was misleading turns on how much deviation there was between the Potential Reach estimate and the number of people that fell within the advertiser’s target criteria. As discussed, targeted Potential Reach estimates are tailored to each advertiser’s choices. Advertisers can narrow their estimates with standard demographics (age, education, gender, etc.) and by location and interests. Altogether, the available targeting criteria provide thousands of options. How this targeting criteria impacts the accuracy of each estimate is apparent considering, for example, duplicate counts. One Facebook user may have two or more accounts—take, for example, one professional and one fraud cases involving standardized written representations to a class of investors”). 42 DZ RESERVE V. META PLATFORMS, INC. personal. So, if an advertiser selects only geographic criteria, both accounts may be counted. But if the advertiser selects both geography and interests criteria, the work account may be excluded and only the personal account counted or vice versa. Given the variability at play in targeted Potential Reach calculations, their degree of accuracy relative to the number of people is not uniform—one Potential Reach calculation may be an accurate “estimate” of the people who may see an advertisement based on the selected criteria while another is not. To make up for the lack of uniformity in the millions of Potential Reach calculations that Meta provided, Plaintiffs first assert that the district court found that Potential Reach estimates were always “significantly inflated.” This misconstrues the record. The district court acknowledged that Plaintiffs’ expert opined the Potential Reach calculation was always significantly inflated and that Meta’s expert did not eliminate the possibility that some inflation occurred. 7 What underpins the district court’s decision is the latter 7 The majority seems to agree that the district court’s acknowledgement that Meta failed to show that no inflation occurred is not the same as crediting Plaintiffs’ expert that significant inflation always occurred, but it ignores the significance of this point in the class context. Resolving conflicts in the evidence is within the district court’s purview. See Ellis v. Costco Wholesale Corp., 657 F.3d 970, 983 (9th Cir. 2011) (“[T]he district court was required to resolve any factual disputes necessary to determine whether there was a common pattern and practice that could affect the class as a whole.”); Olean, 31 F.4th at 666 (“The determination whether expert evidence is capable of resolving a class-wide question in one stroke may include weighing conflicting expert testimony and resolving expert disputes, where necessary to ensure that Rule 23(b)(3)’s requirements are met . . . .” (cleaned up)). And without a finding regarding the rate of inflation, the common pattern begins to unravel because there is no set inflation range binding class members together.  DZ RESERVE V. META PLATFORMS, INC. 43 point—that because Meta’s expert failed to establish that no inflation occurred, characterizing Potential Reach as a calculation of people was inaccurate, regardless of the degree of inaccuracy. This means the class includes advertisers who received targeted Potential Reach estimates with a discrepancy between people and accounts that could range from 1% to 50%. Cf. Thrifty Payless, Inc. v. The Americana at Brand, LLC, 218 Cal. App. 4th 1230, 1242 & n.7 (2013) (holding fraud action may be based on an estimate after considering the disparity and finding “the huge disparity between the estimates and the ultimate costs supports an inference of misrepresentation” (emphasis added)). Plaintiffs’ reliance on In re U.S. Foodservice Inc. Pricing Litig., 729 F.3d 108 (2d Cir. 2013), in arguing that a uniform misrepresentation was made regardless of any variation in Potential Reach inflation, is unavailing. In that case, the degree of difference between what the plaintiffs were charged and the “cost-plus” pricing they were entitled to pay was irrelevant to liability because any difference— one cent or a thousand dollars—was proof that plaintiffs were harmed. 729 F.3d at 118, 123. The same is not true here. Meta did not charge advertisers based on its Potential Reach estimates. And the degree of inflation in the Potential Reach calculation is the crux of whether Meta misrepresented the estimated number of people who could potentially see a given ad. Determining whether the Potential Reach calculations were misrepresentations is further challenged by Meta’s evolving disclosures over the class period. Early on, Meta’s disclosures stated that Potential Reach was “not designed to match population or census estimates.” Then in 2019, Meta changed its disclosure to state that Potential Reach depends 44 DZ RESERVE V. META PLATFORMS, INC. on “[h]ow many accounts are used per person.” Meta changed the disclosure again in 2021 to state that “the presence of fake accounts” could impact the Potential Reach calculation. I disagree that the impact of these changes goes only to class-wide merits issues. The court must determine whether individual or common questions will predominate in assessing whether Meta’s Potential Reach calculations were fraudulent misrepresentations. The disclosures that Meta provided regarding the nature of its calculated estimate are important to this analysis. The reasoning in Berger v. Home Depot USA, Inc. is particularly persuasive. 741 F.3d at 1067–69. There, the district court denied certification of a fraud claim brought against Home Depot related to its tool- rental contracts over a multi-year period. Id. at 1066. We affirmed on predominance grounds because the class period covered five different versions of the contract, each with different language requiring an “independent legal analysis.” Id. at 1069. The varying disclosures that Meta provided about the limitations of Potential Reach estimates likewise present individualized issues in determining whether Meta made fraudulent misrepresentations. Cf. Mazza v. Am. Honda Motor Co., 666 F.3d 581, 596 (9th Cir. 2012) (holding a class definition as fatally overbroad where many class members learned that the advertising was misleading before purchase). In the cases where we have upheld certification of a fraud class based on misrepresentation of an estimate, the class members were given the same estimate. See In re Hyundai, 926 F.3d at 553, 559 (upholding certification based on “inflated fuel economy standards” that were uniformly disseminated). That is not what happened here, and the evidence does not establish that the millions of unique Potential Reach calculations that Meta provided to the class  DZ RESERVE V. META PLATFORMS, INC. 45 had the same degree of inflation. Is a Potential Reach calculation with a 2% deviation a misrepresentation where the targeted population includes millions of people? What about a Potential Reach calculation with an 8% deviation where the targeted population includes only 1,000 people? Where a class claim “prevail[s] or fail[s] in unison,” it satisfies predominance. Amgen, 568 U.S. at 460. That standard is not met here because the factfinder could conclude that some, but not all, Potential Reach calculations presented to the class members were fraudulently misleading. See Lara, 25 F.4th at 1139 (affirming denial of class certification because “figuring out whether each individual putative class member was harmed would involve an inquiry specific to that person”). 8 B. Materiality Because this case does not involve a uniform misrepresentation, many of the problems discussed in relation to the misrepresentation element of Plaintiffs’ claim also apply to the materiality-of-the-misrepresentation element. Under California law, a misrepresentation is material if “a reasonable man would attach importance to its existence or nonexistence in determining his choice of action 8 The reasoning in Reitman v. Champion Petfoods USA, Inc., 830 F. App’x 880 (9th Cir. 2020), though unpublished, is similarly persuasive. There, this court affirmed denial of certification on predominance grounds because whether a representation was false depended on comparing each individual product. Id. at 881. The products, dog food, had packaging that contained different information, and the court would need to conduct a bag-to-bag comparison for each representation. This led to individual questions predominating. Id. Here, each Potential Reach estimate is akin to an individual product that would require an individualized assessment to determine if each “product” was indeed false. 46 DZ RESERVE V. META PLATFORMS, INC. in the transaction in question.” Engalla, 15 Cal. 4th at 977 (quoting Restatement (Second) of Torts § 538 (1977)). Materiality is generally a fact question unless the “fact misrepresented is so obviously unimportant that the jury could not reasonably find that a reasonable man would have been influenced by it.” Id. Our focus here is whether common or individual issues will predominate in determining whether a misrepresentation is material, not whether Plaintiffs can prove materiality. See Amgen, 568 U.S. at 469; Olean, 31 F.4th at 667. In the majority’s view, Amgen established that materiality always satisfies predominance because it is governed by an objective standard. I disagree. In Amgen, the Court concluded that the class had a “fatal similarity.” 568 U.S. at 470. If materiality failed for one, it failed for all. Id. at 468 (“A failure of proof on the common question of materiality ends the litigation and thus will never cause individual questions of reliance or anything else to overwhelm questions common to the class.”). The Court reached this conclusion because “[i]n no event will the individual circumstances of particular class members bear on the inquiry” of the materiality of the allegedly fraudulent statements Amgen made about its products that inflated its stock price. 568 U.S. at 460. This makes sense in securities- fraud cases that address fraudulent statements that are released to and impact the market. See id. at 466 (“[I]mmaterial information, by definition, does not affect market price . . . .”). But nothing in Amgen commands that materiality, no matter the context, necessarily is provable with class-wide evidence and, therefore, satisfies the predominance requirement. Additionally, while Amgen addressed a claim arising under federal law, the Plaintiffs’ claims here are governed  DZ RESERVE V. META PLATFORMS, INC. 47 by California law. California courts applying that state’s law have recognized that materiality cannot be resolved on a class-wide basis where this issue inevitably depended on individualized questions. See, e.g., In re Vioxx Class Cases, 180 Cal. App. 4th 116, 129 (2009) (stating that “if the issue of materiality . . . is a matter that would vary from consumer to consumer, the issue is not subject to common proof, and the action is properly not certified as a class action”). 9 And federal courts applying California law likewise have found Rule 23 predominance lacking when plaintiffs fail to proffer class-wide evidence of how a reasonable consumer would interpret the allegedly misrepresented fact or when consumers are interested in a product for a variety of reasons. See, e.g., Townsend v. Monster Beverage Corp., 303 F. Supp. 3d 1010, 1045, 1047–48 (C.D. Cal. 2018) (collecting cases). Here, plaintiffs primarily rely on two pieces of evidence in arguing that materiality is susceptible to class-wide proof: 9 In an unpublished decision, the California Court of Appeal upheld denial of class certification in part because individualized questions predominated regarding the materiality of “online fuel calculator” estimates provided to encourage consumers to buy the Toyota Prius. Reynante v. Toyota Motor Sales USA, Inc., No. B275937, 2018 WL 329569 (Cal. Ct. App. Jan. 9, 2018). The online calculations were accompanied by a message stating that “results are based on estimates.” Id. at *4. The court reasoned that “[w]hether a consumer was actually misled by the fuel calculator prior to purchasing a [car] necessarily would vary by customer” because “[s]ome customers, for example, could have viewed the fuel calculator and have been adequately informed—whether by their experience with vehicle EPA estimates or by the disclaimer—that their actual fuel efficiency would vary based on driving conditions.” Id. Similarly here, many advertisers are repeat players and Meta provides historical data from previous ad campaigns that can further alter their understanding of Potential Reach. 48 DZ RESERVE V. META PLATFORMS, INC. Dr. Cowan’s statistical analysis and Dr. Allenby’s conjoint survey. Plaintiffs assert that Dr. Cowan can establish that all Potential Reach estimates were inflated by at least 10%. The Supreme Court has held that “proving classwide liability” through statistical sampling is appropriate if “each class member could have relied on that sample to establish liability” in an individual action. Tyson Foods, Inc. v. Bouaphakeo, 577 U.S. 442, 455 (2016). But here, it is unclear that materiality can be established based on just the percentage of deviation. As discussed above, the degree of inflation in the Potential Reach estimate informs whether a misrepresentation has occurred, let alone a material misrepresentation. The degree of inflation relative to the total number of people within the targeted audience may also be relevant. A 10% deviation may have different import as relates to a reach of millions than to a reach of thousands or hundreds. But even assuming plaintiffs could establish that a 10% inflation rate, or some other threshold, is always material, that does not resolve the claims of class members who received Potential Reach estimates with less than the threshold. Thus, again, while Plaintiffs’ statistical evidence may prove or disprove some claims, they have not shown it can resolve all claims within the far-reaching class. 10 10 This is true even if the misrepresentation at issue is merely Meta’s statement that Potential Reach is an estimate of people instead of accounts. Plaintiffs point to no evidence that could establish on a class- wide basis that reasonable advertisers view the account-as-proxy-for- people itself as a material misrepresentation regardless of the degree of deviation between those two metrics. Cf. In re NJOY, Inc. Consumer Class Action Litig., 120 F. Supp. 3d 1050, 1117 (C.D. Cal. 2015) (rejecting a survey as dispositive of materiality, in a predominance analysis, where it “did not ask respondents questions relevant in assessing the materiality of information omitted from the packaging”).  DZ RESERVE V. META PLATFORMS, INC. 49 Turning to Dr. Allenby’s conjoint survey, the district court assessed this evidence in analyzing damages, not materiality. This survey included only small-to-medium businesses, not the full breadth of entities that compose the class. It also did not mirror Meta’s varying disclosures during the class period. And lastly, this survey is representative of only 7% of the class. 11 While the survey shows that some respondents would increase their spending if an audience size was increased 10% in the abstract, Plaintiffs have not pointed to any evidence that the reasonable ad purchaser in this class would understand the estimated Potential Reach to not have any inflation or deviation. Nor is there any evidence addressing how reasonable advertisers would understand Potential Reach in light of Meta’s evolving disclosures. In sum, there are two primary reasons why predominance is not satisfied as to materiality. First, determining the objective perspective of a reasonable advertiser is made difficult by the breadth of Plaintiffs’ proposed class, which includes millions of advertisers of all types conducting advertising campaigns ranging from millions of dollars to tens of dollars. Cf. Webb v. Carter’s Inc., 272 F.R.D. 489, 502 (C.D. Cal. 2011) (declining to apply the objective “reasonable consumer standard” where materiality would “vary from consumer to consumer”). Second, Meta told advertisers that Potential Reach is an estimate, and Meta provided evolving disclosures about the limitations of this estimate. A false estimate undoubtedly can 11 The study only included respondents that spent $1,000 to $25,000 per year on advertising through their employment. The majority of Meta’s advertisers spend less than $50 per year. And one of the Named Plaintiffs spent upwards of $1 million on Meta advertising. 50 DZ RESERVE V. META PLATFORMS, INC. be the basis for a fraud claim, see Aloe Vera of Am., Inc. v. United States, 699 F.3d 1153, 1164 (9th Cir. 2012), but what a reasonable purchaser believes about the precision of information necessarily is impacted by what they are told about precision. This case is a far cry from the objective class-wide materiality analysis that was appropriate in Amgen. Because securities fraud impacts the market, “fantastic scenarios in which an individual investor might rely on immaterial information (think of the superstitious investor who sells her securities based on a CEO’s statement that a black cat crossed the CEO’s path that morning)” do not establish that materiality is an individualized issue. Amgen, 568 U.S. at 469. But here, the ability to establish materiality based on class-wide proof is not undermined by “fantastic scenarios.” Id. The Named Plaintiffs’ own actions help demonstrate the point. Taking a “peek at the merits,” Dancel v. Groupon, Inc., 949 F.3d 999, 1005 (7th Cir. 2019) (citation omitted), the owner of DZ Reserve made statements online that inflation in the Potential Reach calculation “should . . . not deter anyone from doing [Facebook] ads for [e-commerce].” Cf. Johnson v. Harley-Davidson Motor Co. Grp., LLC, 285 F.R.D. 573, 581 (E.D. Cal. 2012) (finding materiality lacking, in part, because the former named plaintiffs, even with full knowledge of a product’s defect, “would still buy and recommend the [product]”). And Maxwell set an advertising budget of $20 regardless of whether the Potential Reach estimate was one million or 50 million. Contrary to the district court’s assertion otherwise, this evidence suggests that ad buyers as a group may not have “attach[ed] importance” to Potential Reach in choosing to buy Facebook ads. Engalla, 15 Cal. 4th at 977.  DZ RESERVE V. META PLATFORMS, INC. 51 The district court’s assertion that materiality is provable on a class-wide basis because “Potential Reach is an important number for advertisers,” improperly conflates the importance of the subject matter with the importance of the claimed misrepresentation and also fails to meet the rigors of Rule 23(b)(3). Cf. In re ConAgra Foods, Inc., 90 F. Supp. 3d 919, 1019 (C.D. Cal. 2015), aff’d sub nom. Briseno v. ConAgra Foods, Inc., 674 F. App’x 654 (9th Cir. 2017), and aff’d sub nom. Briseno v. ConAgra Foods, Inc., 844 F.3d 1121 (9th Cir. 2017) (analyzing the materiality of food labels not for their importance generally but for how consumers understand them). ConAgra concerned whether a “100% Natural” food label on cooking oil was a misrepresentation. Id. at 1018. Food labels are shown to all consumers, but the district court did not consider the importance of food labels in the abstract, it considered the content of the challenged label and how reasonable consumers would understand that content. Id. at 1019. Here, the district court’s and the majority’s framing of Plaintiffs’ case derails their analyses. Cf. Gonzalez v. Corning, 885 F.3d 186, 201 (3d Cir. 2018), as amended (Apr. 4, 2018) (“[T]he ‘question of defect’ they propose is only superficially a ‘common question,’ just as any question becomes universal when it includes the word ‘all.’”); In re Vioxx, 180 Cal. App. 4th at 133–34 (rejecting an argument, as an “oversimplification,” where plaintiffs argued there was nothing more material than “risk of death” because some patients and doctors would still use the medicine regardless of the risk). The proper focus is on how advertisers in the class would view the Potential Reach estimates they received in specific transactions, based on the total mix of information available at the time of purchase. See Engalla, 15 Cal. 4th at 977–78 (assessing materiality based on the explicit and 52 DZ RESERVE V. META PLATFORMS, INC. implicit representations made in the context of the transaction). Facebook provided advertisers with individualized information beyond Potential Reach. For example, the “Estimated Daily Reach” calculation—viewed alongside Potential Reach—estimated how many people might see an ad each day based on the buyer’s advertising budget. The Estimated Daily Reach was part of the calculus informing the buyers’ reasonable expectations in purchasing ads. Cf. Algarin v. Maybelline, LLC, 300 F.R.D. 444, 457 (S.D. Cal. 2014) (looking at consumer expectations for a product in determining that materiality was not susceptible to common proof). For all these reasons, the materiality analysis required in this case centers on individualized questions of what advertisers understood about the information they were given at the time they purchased Facebook ads, and, therefore, Plaintiffs have not satisfied the predominance requirement. C. Reliance Finally, actual reliance is an essential element of fraud under California law. Conroy v. Regents of Univ. of Cal., 45 Cal. 4th 1244, 1256 (2009). Actual reliance does not require proving the alleged misrepresentation was the “sole” or “decisive” cause of the plaintiff entering into the transaction. Id. A plaintiff need only prove the misrepresentation was an “immediate cause” or “played a substantial part” in entering the transaction. Id. A plaintiff meets this burden by showing that, absent the misrepresentation, the plaintiff “would not, in all reasonable probability, have entered into the . . . transaction.” Id. (internal quotation marks and citation omitted). California law recognizes that “when the same material misrepresentations have actually been  DZ RESERVE V. META PLATFORMS, INC. 53 communicated to each member of a class, an inference of reliance arises as to the entire class.” Kaldenbach v. Mutual of Omaha Life Insurance Co., 178 Cal. App. 4th 830, 851 (2009), as modified (Oct. 26, 2009) (citation and emphasis omitted). But the presumption of reliance does not apply where uniformity of representation is lacking, or at least does not predominate. Id. The seminal California case applying this presumption involved salesmen that “memorized a standard statement” that was “recited by rote to every member of the class.” Vasquez v. Superior Ct., 4 Cal. 3d 800, 812 (1971); see also Occidental Land, Inc. v. Superior Ct., 18 Cal. 3d 355, 358– 59, 363 (1976) (applying presumption where the class read the same document containing the misrepresentation and was required to state in writing that they had read it). On the other hand, the Kaldenbach court did not apply the presumption of reliance where the case involved individualized sales presentations because the plaintiff had not overcome the “significant individual issues” of whether misrepresentations were made to each class member. 178 Cal. App. 4th at 851. Here, the district court erred by applying the presumption of reliance as a basis for granting class certification of Plaintiffs’ Rule 23(b)(3) damages class because Plaintiffs did not establish that Meta made a uniform misrepresentation. 12 12 Plaintiffs rely on Tobacco II to discount alternative information Meta provided to advertisers. Tobacco II stated that “an allegation of reliance is not defeated merely because there was alternative information available to the consumer-plaintiff, even regarding an issue as prominent as whether cigarette smoking causes cancer.” 46 Cal. 4th 298, 328 (2009). This language, however, concerns alternative information 54 DZ RESERVE V. META PLATFORMS, INC. For these reasons, I would reverse the district court’s certification of Plaintiffs’ Rule 23(b)(3) damages class. I do not reach Meta’s additional challenges regarding the district court’s typicality and adequacy analyses. external to a defendant’s representation, such as medical studies by third parties. See id. (citing to a case discussing “common knowledge”). It does not concern information provided by the defendant that is directly relevant in determining whether a misrepresentation occurred at all.