Opinion for the Court filed by Circuit Judge BROWN.
Dissenting opinion filed by Circuit Judge ROGERS.
BROWN, Circuit Judge:The Family Smoking Prevention and Tobacco Control Act (“the Act”), Pub.L. No. 111-31, 123 Stat. 1776 (2009), directed the Secretary of the U.S. Department of Health and Human Services to issue regulations requiring all cigarette packages manufactured or sold in the United States to bear one of nine new textual warnings, as well as “color graphics depicting the negative health consequences of smoking.” See id. § 201(a). Pursuant to this authority, the Food and Drug Administration (“FDA”) initiated a rulemaking proceeding through which it selected the nine images that would accompany the statutorily-prescribed warnings. Five tobacco companies (“the Companies”) challenged the rule, alleging that FDA’s proposed graphic warnings violated the First Amendment. See Compl. at 35-36.1 The district court granted the Companies’ motion for summary judgment on February 29, 2012.2 FDA appeals, and we affirm.
1. Background
The Act gives FDA the authority to regulate the manufacture and sale of tobacco products, including cigarettes. In addition to requiring cigarette packages and advertisements to bear one of nine new warning statements, the Act mandates that the new warning labels comprise the top 50 percent of the front and rear panels of cigarette packages and 20 percent of the area of each cigarette advertisement. Act § 201(a), 123 Stat. at 1842-45. The Act directs the Secretary to issue final regula*1209tions identifying the graphic component of the warnings by June 22, 2011, and provides that the revised health warnings will take effect by September 22, 2012. See 15 U.S.C. § 1333 note.
Pursuant to the statutory directive, FDA issued a Proposed Rule seeking comment on thirty-six potential images for the new graphic warning labels. Required Warnings for Cigarette Packages and Advertisements, 75 FecLReg. 69,524, 69,534 (Nov. 12, 2010) (hereinafter Proposed Rule). At the outset of the Proposed Rule, FDA asserted the government’s “substantial interest in reducing the number of Americans, particularly children and adolescents, who use cigarettes and other tobacco products in order to prevent the life-threatening health consequences associated with tobacco use.” Id. at 69,525. In ' accordance with the requirements of the Act, FDA proposed a dramatic expansion of the existing health warnings, which it justified based on scientific literature and a “strong worldwide consensus”3 regarding the relative effectiveness of graphic warnings compared to the text-only warnings the United States currently requires. Id. The agency explained that by “clearly and effectively conveying] the negative health consequences of smoking,” the new warnings would discourage nonsmokers, particularly minors, from “initiating cigarette use,” and encourage current smokers to quit. Id. at 69,526.
FDA promulgated the final set of nine images — one for each warning statement— by regulations issued on June 22, 2011. See Required Warnings for Cigarette Packages and Advertisements, 76 Fed. Reg. 36,628 (June 22, 2011) (hereinafter Final Rule). FDA also required each graphic image to bear the phone number of the National Cancer Institute’s “Network of Tobacco Cessation Quitlines,” which uses the telephone portal “1-800-QUIT-NOW.” Id. at 36,681.
FDA based its selection of the final images on an 18,000-person internet-based consumer study it commissioned. The study divided respondents into two groups: a control group that was shown the new text in the format of the current warnings (located on the side of cigarette packages), and a separate treatment group that was shown the proposed graphic warnings, which included the new text, the accompanying graphic image, and the 1-800-QUIT-NOW number. Id. at 36,638. Each group then answered questions designed to assess, among other things, whether the graphic warnings, relative to the text-only control, (1) increased viewers’ intention to quit or refrain from smoking; (2) increased viewers’ knowledge of the health risks of smoking or secondhand smoke; and (3) were “salient,” which FDA defined in part as causing viewers to feel “depressed,” “discouraged,” or “afraid.” Id.
In selecting these nine images, FDA reviewed and responded to over a thousand public comments, including joint comments submitted by plaintiffs-appellees RJ Reynolds, Lorillard, and Commonwealth Brands. See id. at 36,629. Several eom*1210ments — including comments from cancer researchers, nonprofits, and academics— criticized the single exposure study design, noting it prevented the government from assessing the long-term or actual effects of the proposed warnings. Two of these comments recommended FDA conduct longitudinal research or post-market surveillance to assess actual long-term effects. Id. at 36,639. FDA conceded the study did not permit it to reach “firm” conclusions about the “long-term, real-world effects” of the proposed warnings, but claimed the existing scientific literature “provides a substantial basis for our conclusion that the required warnings will effectively communicate the health risks of smoking, thereby encouraging smoking cessation and discouraging smoking initiation.” Id. Still other comments asserted that FDA’s research study failed to provide evidence that the proposed warnings would actually affect smoking rates, significantly affect consumers knowledge of the risks of smoking, or bring about actual behavior change. See id. at 36,640. But FDA disagreed, again relying on the “substantial research” showing the effectiveness of similar graphic health warnings in other countries. Id. (citing Proposed Rule at 69,531-34).4 Another comment asserted that the study’s selection bias constituted a serious methodological flaw. Namely, participants were recruited from an internet panel and offered the opportunity to participate in an FDA-sponsored research study. Id. at 36,643. FDA avoided the substance of this argument by conceding that its study “provides insight on the relative effectiveness of the various warnings under consideration,” not on the “absolute effects of the warnings in general.” Id.
Some comments also criticized the lack of statistical evidence supporting FDA’s belief that requiring cigarette packages to bear the graphic warnings would reduce smoking rates. See id. For example, the Companies noted that the Canadian data revealed no statistically significant decline in smoking rates for adolescents and adults after the introduction of similar graphic warnings, which implied that the warnings were ineffective and that FDA’s warnings would be ineffective as well. Id. FDA summarily disagreed, stating that the images it selected would satisfy its “primary goal, which is to effectively convey the negative health consequences of smoking on cigarette packages and in advertisements,” which can help “both to discourage nonsmokers ... from initiating cigarette use and to encourage current smokers to consider cessation.” Final Rule at 36,633. FDA also explained that the data from Canada did not indicate that the warnings had been ineffective, because other studies showed that the warnings had been “effective at providing ... smokers with health information, making *1211consumers think about the health effects of smoking, and increasing smokers’ motivations to quit smoking.” Id. at 36,634.
After FDA finalized the Rule, the Companies filed suit in the district court, claiming the cigarette warnings required under the Act and FDA’s implementing regulations violated the First Amendment. The district court granted the Companies’ motion for a preliminary injunction on November 7, 2011, and subsequently granted their motion for summary judgment. FDA appeals, and we review de novo the district court’s decision to grant summary judgment. Davis v. Pension Benefit Guar. Corp., 571 F.3d 1288, 1291 (D.C.Cir.2009).
II. Level of Scrutiny
The Companies do not dispute Congress’s authority to require health warnings on cigarette packages, nor do they challenge the substance of any of the nine textual statements mandated by the Act. The only question before us is whether FDA’s promulgation of the graphic warning labels — which incorporate the textual warnings, a corresponding graphic image, and the “1-800-QUIT-NOW” cessation hotline number — violates the First Amendment. We begin our analysis by determining the applicable level of scrutiny.
Both the right to speak and the right to refrain from speaking are “complementary components of the broader concept of individual freedom of mind” protected by the First Amendment. Wooley v. Maynard, 430 U.S. 705, 714, 97 S.Ct. 1428, 51 L.Ed.2d 752 (1977). Any attempt by the government either to compel individuals to express certain views, see id. at 714-15, 97 S.Ct. 1428, or to subsidize speech to which they object, see United States v. United Foods, Inc., 533 U.S. 405, 410-11, 121 S.Ct. 2334, 150 L.Ed.2d 438 (2001), is subject to strict scrutiny. The general rule “that the speaker has the right to tailor the speech[] applies not only to expressions of value, opinion, or endorsement, but equally to statements of fact the speaker would rather avoid.” Hurley v. Irish-Am. Gay, Lesbian & Bisexual Grp. of Bos., 515 U.S. 557, 573-74, 115 S.Ct. 2338, 132 L.Ed.2d 487 (1995). This holds true whether individuals, see W. Va. State Bd. of Educ. v. Barnette, 319 U.S. 624, 642, 63 S.Ct. 1178, 87 L.Ed. 1628 (1943), or corporations, see Pac. Gas & Elec. Co. v. Pub. Utils. Comm’n, 475 U.S. 1, 16, 106 S.Ct. 903, 89 L.Ed.2d 1 (1986) (plurality opinion), are being compelled to speak.
This case contains elements of compulsion and forced subsidization. The Companies contend that, to the extent the graphic warnings go beyond the textual warnings to shame and repulse smokers and denigrate smoking as an antisocial act, the message is ideological and not informational. “[B]y effectively shouting well-understood information to consumers,” they explain, “FDA is communicating an ideological message, a point of view on how people should live their lives: that the risks from smoking outweigh the pleasure that smokers derive from it, and that smokers make bad personal decisions, and should stop smoking.” In effect, the graphic images are not warnings, but admonitions: “[D]on’t buy or use this product”5 No one doubts the government can *1212promote smoking cessation programs; can use shock, shame, and moral opprobrium to discourage people from becoming smokers; and can use its taxing and regulatory authority to make smoking economically prohibitive and socially onerous. And the government can certainly require that consumers be fully informed about the dangers of hazardous products. But this case raises novel questions about the scope of the government’s authority to force the manufacturer of a product to go beyond making purely factual and accurate commercial disclosures and undermine its own economic interest — in this case, by making “every single pack of cigarettes in the country [a] mini billboard” for the government’s anti-smoking message.6
Even assuming the Companies’ marketing efforts (packaging, branding, and other advertisements) can be properly classified as commercial speech, and thus subject to less robust First Amendment protections, a thorny question remains: how much leeway should this Court grant the government when it seeks to compel a product’s manufacturer to convey the state’s subjective — and perhaps even ideological — view that consumers should reject this otherwise legal, but disfavored, product? Neither the Act nor the agency’s regulation squarely addresses this question. However, for present purposes, we can assume, without deciding, that if such compulsion is constitutionally permissible, the state’s actions must still withstand the applicable level of scrutiny.
Courts have recognized a handful of “narrow and well-understood exceptions” to the general rule that content-based speech regulations — including compelled speech — are subject to strict scrutiny. See Turner Broad. Sys., Inc. v. FCC, 512 U.S. 622, 641, 114 S.Ct. 2445, 129 L.Ed.2d 497 (1994). There are two primary exceptions in the commercial speech context. First, “purely factual and uncontroversial” disclosures are permissible if they are “reasonably related to the State’s interest in preventing deception of consumers,” provided the requirements are not “unjustified or unduly burdensome.” Zauderer; 471 U.S. at 651, 105 S.Ct. 2265. Second, restrictions on commercial speech are subject to less stringent review than restrictions on other types of speech. For a statute burdening commercial speech to survive, the government must affirmatively prove that (1) its asserted interest is substantial, (2) the restriction directly and materially advances that interest, and (3) the restriction is narrowly tailored. See Cent. Hudson Gas & Elec. Corp. v. Pub. Serv. Comm’n, 447 U.S. 557, 566, 100 S.Ct. 2343, 65 L.Ed.2d 341 (1980). While this test is not quite as demanding as strict scrutiny, it is significantly more stringent than Zauderer’s standard, which is akin to rational-basis review.
The district court concluded the graphic warnings were “not the type of purely factual and uncontroversial” disclosures reviewable under the less stringent Zauderer standard. R.J. Reynolds Tobacco Co. v. *1213FDA, 845 F.Supp.2d 266, 272-73 (D.D.C.2012) (hereinafter “Merits Opinion”). Applying strict scrutiny, the court held that FDA failed to satisfy its burden of demonstrating that the Rule is narrowly tailored to achieve a compelling government interest. See id. at 275-77. FDA argues that the district court erred in finding the Zauderer standard inapplicable. Alternatively, it contends that the district court erred by failing to apply the intermediate-level scrutiny generally afforded to commercial speech, and that the graphic warnings pass constitutional muster under Central Hudson. We address each argument in turn.
a. Applicability of the Zauderer Standard
In Zauderer, the Court applied a lower level of scrutiny to regulations requiring attorneys to fully disclose information about the actual cost and consequences of services. 471 U.S. at 651-52, 105 S.Ct. 2265. Noting that the First Amendment’s protection of commercial speech is premised on its informational value to consumers, the Court reasoned that an advertiser’s constitutional interest in not providing additional factual information was “minimal.” Id. at 651, 105 S.Ct. 2265. Although the Court acknowledged that “unjustified or unduly burdensome disclosure requirements might offend the First Amendment by chilling protected commercial speech,” it “h[e]ld that an advertiser’s rights are adequately protected as long as disclosure requirements are reasonably related to the State’s interest in preventing deception of consumers.” Id.; see also Milavetz, Gallop & Milavetz, P.A. v. United States, — U.S. -, 130 S.Ct. 1324, 1340, 176 L.Ed.2d 79 (2010) (applying the Zauderer standard to disclosure requirements “intended to combat the problem of inherently misleading commercial advertisements”).
The Supreme Court has never applied Zauderer to disclosure requirements not designed to correct misleading commercial speech. FDA argues that Zauderer’s lenient standard of scrutiny applies to regulations that serve a different governmental interest: disclosure of the health and safety risks associated with commercial products. See Appellant’s Br. at 26.
But by its own terms, Zauderer’s holding is limited to cases in which disclosure requirements are “reasonably related to the State’s interest in preventing deception of consumers.” 471 U.S. at 651, 105 S.Ct. 2265. Zauderer “carries no authority for a mandate unrelated to the interest in avoiding misleading or incomplete commercial messages.” Glickman v. Wileman Bros. & Elliott, Inc., 521 U.S. 457, 491, 117 S.Ct. 2130, 138 L.Ed.2d 585 (1997) (Souter, J., dissenting, joined by Rehnquist, C.J., and Scalia and Thomas, JJ.) (explaining why Zauderer was inapplicable in that case).7 In United States v. United Foods, for example, the Court declined to apply the Zauderer standard when evaluating a federal law requiring mushroom producers to pay an assessment to support generic advertising. The Court distinguished Zauderer because there was no suggestion “that the mandatory assessments imposed to require one group of private persons to pay for speech by others are somehow necessary to make voluntary advertisements non-misleading for consumers.” 533 U.S. at 416, 121 S.Ct. 2334. And as the Court explained in Pacific Gas, “[njothing in Zauderer suggests ... that *1214the State is equally free to require [entities] to carry the messages of third parties, where the messages themselves are biased against or are expressly contrary to the [entity’s] views.” 475 U.S. at 15 n. 12, 106 S.Ct. 903 (plurality opinion).
Ibanez v. Florida Department of Business and Professional Regulation also suggests that Zauderer should be construed to apply only when the government affirmatively demonstrates that an advertisement threatens to deceive consumers. In that case, the state Board of Accountancy contended that an attorney’s use of her Certified Financial Planner designation in an advertisement was “potentially misleading,” and thus entitled the Board to require her to include a disclaimer. 512 U.S. 136, 146, 114 S.Ct. 2084, 129 L.Ed.2d 118 (1994). But the Court declined to apply Zauderer, finding that “given the state of this record,” the Board failed “to point to any harm that is potentially real, not purely hypothetical.” Id. Put simply, the government could not seek review under the lenient Zauderer standard absent a showing that the advertisement at issue would likely mislead consumers.
In fact, the Court’s only recent application of the Zauderer standard involved a disclosure requirement that “share[d] the essential features of the rule at issue in Zauderer.” Milavetz, 130 S.Ct. at 1340. In Milavetz, a law firm challenged a provision of the Bankruptcy Abuse Prevention and Consumer Protection Act of 2005 (“BAPCPA”) that required professionals qualifying as debt relief agencies to “clearly and conspicuously disclose in any advertisement of bankruptcy assistance services ... that the services or benefits are with respect to bankruptcy relief under this title.” 11 U.S.C. § 528(a)(3). BAPCPA also required qualifying professionals to state that “[w]e are a debt relief agency. We help people file for bankruptcy relief under the Bankruptcy Code.” Id. § 528(a)(4). The Court upheld the statute’s disclosure requirement because, as in Zauderer, the law firm’s advertisements were “inherently misleading” — in this case, because they “promis[ed] ... debt relief without any reference to the possibility of filing for bankruptcy, which has inherent costs.” Milavetz, 130 S.Ct. at 1340. One Justice even cautioned against interpreting the Court’s holding as a “presumptive[ ] endorsement of] laws requiring the use of government-scripted disclaimers in commercial advertising,” noting that Zauderer does not stand for the proposition that government “can constitutionally compel the use of a scripted disclaimer in any circumstance in which its interest in preventing consumer deception might plausibly be at stake.” Id. at 1343-44 (Thomas, J., concurring in part and concurring in the judgment).
Zauderer, Ibanez, and Milavetz thus establish that a disclosure requirement is only appropriate if the government shows that, absent a warning, there is a self-evident — or at least “potentially real” — danger that an advertisement will mislead consumers. Ibanez, 512 U.S. at 146, 114 S.Ct. 2084. In this case, the proposed disclosure requirements would apply to both cigarette advertisements and cigarette packages. The Act bans any labeling or advertising representing that any tobacco product “presents a lower risk of tobacco-related disease or is less harmful than one or more other commercially marketed tobacco products,” “contains a reduced level of a substance or presents a reduced exposure to a substance,” or “does not contain or is free of a substance.” 21 U.S.C. § 387k. The Act also bans advertising or labeling using the descriptors “light,” “mild,” “low,” or similar descriptors. Id. In light of these restrictions, and in the absence of any congressional find*1215ings on the misleading nature of cigarette packaging itself, there is no justification under Zauderer for the graphic warnings.
The dissent’s argument that cigarette packages and other advertisements that fail to prominently display the negative health consequences of smoking are misleading, see Dissent at 1228-29, seems to blame the industry for playing by the government’s rules. The Companies have never argued that no disclosure requirements are warranted; they merely object to the form and content of the specific requirements proposed by the FDA. Indeed, it seems likely the FDA did not make any such claims because the industry has complied precisely with all of the government’s previous disclosure requirements, and continues to do so. Moreover, the Companies generally acknowledge the need for effective warnings and concede in their brief that they would be amenable to a number of new disclosure requirements, including putting the Act’s new text on the side of packages, the bottom front of packages and advertisements, or using less shocking graphics. Appellees’ Br. at 58.8
The amicus States suggest that the graphic warnings be evaluated in the context of the years of deception that preceded them.9 States’ Br. at 7. Citing Warner-Lambert Co. v. FTC, 562 F.2d 749 (D.C.Cir.1977) they claim this Court has found that even advertisements that do not appear deceptive in isolation can constitute “part of a continuing deception of the public” absent highly visible warnings. Id. at 769. But the States’ argument overlooks the broader context of that decision. Warner-Lambert involved a petition for review of an FTC order requiring the Warner-Lambert company to cease and desist from advertising that its product, Listerine mouthwash, prevents, cures, or alleviates the common cold. Id. at 752. As a remedial measure, the Commission required Warner-Lambert to include the following disclosure in every future advertisement for Listerine for a defined period: “Contrary to prior advertising, Listerine will not help prevent colds or sore throats or lessen their severity.” Id. at 753. In other words, the disclosure statement was required as part of a corrective order which the Commission found necessary to “dissipate the effects of respondent’s deceptive representations.” Id. at 769; see also Novartis Corp. v. FTC, 223 F.3d 783, 788-89 (D.C.Cir.2000) (upholding the Commission’s corrective order imposing disclosure requirements on drug manufacturer).
By contrast, FDA does not frame this rule as a remedial measure designed to counteract specific deceptive claims made by the Companies, nor did it offer a remedial justification for the graphic warnings during the rulemaking proceeding. While the Companies’ representations about *1216“light” or “low tar” cigarettes might have been misleading, see United States v. Philip Morris USA Inc., 566 F.3d 1095, 1124-26 (D.C.Cir.2009), the Act now prohibits such statements. See 21 U.S.C. § 387k. Unlike in Warner-Lambert, FDA has not shown that the graphic warnings were designed to correct any false or misleading claims made by cigarette manufacturers in the past.10 Nor did it show that absent disclosure, consumers would likely be deceived by the Companies’ packaging in the future. Rather, FDA framed the warnings as general disclosures about the negative health effects of smoking. The warnings thus represent an ongoing effort to discourage consumers from buying the Companies’ products, rather than, as in Warner-Lambert, a measure designed to combat specific deceptive claims.
Moreover, the graphic warnings do not constitute the type of “purely factual and uncontroversial” information, Zauderer, 471 U.S. at 651, 105 S.Ct. 2265, or “accurate statements,” Milavetz, 130 S.Ct. at 1340, to which the Zau&erer standard may be applied. The disclosures approved in Zauderer and Milavetz were clear statements that were both indisputably accurate and not subject to misinterpretation by consumers. See Zauderer, 471 U.S. at 633, 105 S.Ct. 2265 (describing the disciplinary rule that required “that any advertisement that mentions contingent-fee rates must disclos[e] whether percentages are computed before or after deduction of court costs and expenses”); Milavetz, 130 S.Ct. at 1330 (describing BAPCPA disclosure requirements, including, inter alia, a statement that “[w]e are a debt relief agency. We help people file for relief under the Bankruptcy Code.”).
The FDA’s images are a much different animal. FDA concedes that the images are not meant to be interpreted literally, but rather to symbolize the textual warning statements, which provide “additional context for what is shown.” Final Rule at 36,655. But many of the images chosen by FDA could be misinterpreted by consumers. For example, the image of a man smoking through a tracheotomy hole might be misinterpreted as suggesting that such a procedure is a common consequence of smoking — a more logical interpretation than FDA’s contention that it symbolizes “the addictive nature of cigarettes,” which requires significant extrapolation on the part of the consumers. Id. at 36,649. Moreover, the graphic warnings are not “purely” factual because — as FDA tacitly admits — they are primarily intended to evoke an emotional response, or, at most, shock the viewer into retaining the information in the text warning. See Appellant’s Br. at 33 (citing research showing that “pictures are easier to remember than words”); id. at 38 (citing FDA’s finding that a substantial body of scientific literature shows that emotional responses, such as worry and disgust, “reliably predict the likelihood that consumers will understand and appreciate the substance of the warnings”).
In fact, many of the images do not convey any warning information at all, much less make an “accurate statement” about cigarettes. For example, the images of a woman crying, a small child, and the man wearing a T-shirt emblazoned with the words “I QUIT” do not offer any information about the health effects of smoking. And the “1-800-QUIT-NOW” number, when presented without any explanation about the services provided on the hotline, hardly sounds like an unbiased source of information. These inflammatory images and the provocatively-named hotline ean*1217not rationally be viewed as pure attempts to convey information to consumers. They are unabashed attempts to evoke emotion (and perhaps embarrassment) and browbeat consumers into quitting. See Final Rule at 36,697 (“[R]isk information is most readily conveyed by warnings that elicit ... strong emotional and cognitive reactions .... ”). While none of these images are patently false, they certainly do not impart purely factual, accurate, or uncontroversial information to consumers. Consequently, the images fall outside the ambit of Zauderer.
b. Applicability of Central Hudson
Because this case does not fall within the narrow enclave carved out by Zauderer, we must next determine which level of scrutiny — strict or intermediate— is appropriate. The district court held that compelled speech that falls outside the Zauderer framework is subject to strict scrutiny. See Merits Op. at 274-75. See also Disc. Tobacco City & Lottery, Inc. v. United States, 674 F.3d 509, 554 (6th Cir.2012) (deciding between applying strict scrutiny or Zauderer to compelled commercial speech); Entm’t Software Ass’n v. Blagojevich, 469 F.3d 641, 652 (7th Cir.2006) (same). The government argues that we should view the graphic warnings as restrictions on commercial speech, which are analyzed under the less rigorous standard established by Central Hudson. Despite the contrary views of other circuits, our governing precedent makes clear that Central Hudson is the appropriate standard.
This Court recently evaluated the constitutionality of compelled commercial speech in United States v. Philip Morris, where it reviewed a district court order requiring the defendant tobacco manufacturers to publish corrective statements on their websites, in newspapers, and on major television networks. 566 F.3d at 1142-43. This Court began by noting that “[b]e-cause commercial speech receives a lower level of protection under the First Amendment, burdens imposed on it receive a lower level of scrutiny from the courts.” Id. After acknowledging that “the standard for assessing burdens on commercial speech has varied,” the Court concluded that “the Supreme Court’s bottom line is clear: the government must affirmatively demonstrate its means are narrowly tailored to achieve a substantial government goal.” Id. at 1143. See also Novartis Corp., 223 F.3d at 789 (evaluating a corrective remedy involving corrective statements under Central Hudson). Because this case also involves a compelled commercial disclosure, we follow the lead of Philip Morris and apply the intermediate standard set forth in Central Hudson.
III. Evaluating the Graphic Warnings Under Intermediate Scrutiny
Under Central Hudson, the government must first show that its asserted interest is “substantial.” 447 U.S. at 566, 100 S.Ct. 2343.11 If so, the Court must determine “whether the regulation directly advances the governmental interest asserted, and whether it is not more extensive than is necessary to serve that interest.” Id. The party seeking to uphold a restriction on commercial speech bears the burden of justifying it. Edenfield v. Fane, 507 U.S. 761, 770-71, 113 S.Ct. 1792, 123 L.Ed.2d 543 (1993). Because this case involves a challenge to final agency action, *1218the Administrative Procedure Act governs our review of the record. See 5 U.S.C. § 706(2)(B) (providing that the APA applies to allegations that agency action is “contrary to constitutional right, power, privilege, or immunity”). The APA requires us to “hold unlawful and set aside agency action, findings, and conclusions found to be ... unsupported by substantial evidence.” 5 U.S.C. § 706(2).
Unlike rational-basis review, the Central Hudson standard does not permit this Court to “supplant the precise interests put forward by [FDA] with other suppositions.” Edenfield, 507 U.S. at 768, 113 S.Ct. 1792. We thus begin by identifying FDA’s asserted interests.
A review of the statute and the administrative record makes clear that the graphic warnings are intended to encourage current smokers to quit and dissuade other consumers from ever buying cigarettes. One of the Act’s many stated purposes is “promot[ing] cessation to reduce disease risk and the social costs associated with tobacco-related diseases.” Act § 3.9. The only explicitly asserted interest in either the Proposed or Final Rule is an interest in reducing smoking rates. The Proposed Rule states in its preamble that the government has a “substantial interest in re-during the number of Americans, particularly children and adolescents, who use cigarettes and other tobacco products.” Proposed Rule at 69,525. And the preamble to the Final Rule reiterates the same interest. Final Rule at 36,629.12 Although counsel attempted to disclaim this interest at oral argument, the administrative record shows otherwise: the primary objective of the Rule was “both to discourage nonsmokers from initiating cigarette use and to encourage current smokers to consider quitting.” Id. at 36,630.
Assuming FDA’s interest in reducing smoking rates is substantial,13 we next evaluate whether FDA has offered substantial evidence showing that the graphic warning requirements “directly advance[] the governmental interest asserted,” Cent. Hudson, 447 U.S. at 566, 100 S.Ct. 2343, to a “material degree,” Fl. Bar v. Went For It, Inc., 515 U.S. 618, 626, 115 S.Ct. 2371, 132 L.Ed.2d 541 (1995). The government bears the burden of justifying its attempt to restrict commercial speech, Edenfield, 507 U.S. at 770, 113 S.Ct. 1792, and its burden is not light. A restriction that “provides only ineffective or remote support for the government’s purposes,” id. at 770, 113 S.Ct. 1792, is not *1219sufficient, and the government cannot satisfy its burden “by mere speculation or conjecture.” Rubin v. Coors Brewing Co., 514 U.S. 476, 487, 115 S.Ct. 1585, 131 L.Ed.2d 532 (1995). The requirement that a restriction directly advance the asserted interest is “critical,” because without it, the government “could [interfere with] commercial speech in the service of other objectives that could not themselves justify a burden on commercial expression.” Id.
FDA has not provided a shred of evidence — much less the “substantial evidence” required by the APA — showing that the graphic warnings will “directly advance” its interest in reducing the number of Americans who smoke. FDA makes much of the “international consensus” surrounding the effectiveness of large graphic warnings, but offers no evidence showing that such warnings have directly caused a material decrease in smoking rates in any of the countries that now require them. While studies of Canadian and Australian youth smokers showed that the warnings on cigarette packs caused a substantial number of survey participants to think — or think more — about quitting smoking, Proposed Rule at 69,532, and FDA might be correct that intentions are a “necessary precursor” to behavior change, Final Rule at 36,642, it is mere speculation to suggest that respondents who report increased thoughts about quitting smoking will actually follow through on their intentions. And at no point did these studies attempt to evaluate whether the increased thoughts about smoking cessation led participants to actually quit. Another Australian study reported increased quit attempts by survey participants after that country enacted large graphic warnings, but found “no association with short-term quit success.” Proposed Rule at 69,532. Some Canadian and Australian studies indicated that large graphic warnings might induce individual smokers to reduce consumption, or to help persons who have already quit smoking remain abstinent. See id. But again, the study did not purport to show that the implementation of large graphic warnings has actually led to a reduction in smoking rates.
FDA’s reliance on this questionable social science is unsurprising when we consider the raw data regarding smoking rates in countries that have enacted graphic warnings. FDA claims that Canadian national survey data suggest that graphic warnings may reduce smoking rates. But the strength of the evidence is underwhelming, making FDA’s claim somewhat misleading. In the year prior to the introduction of graphic warnings, the Canadian national survey showed that 24 percent of Canadians aged 15 or older smoked cigarettes. In 2001, the year the warnings were introduced, the national smoking rate dropped to 22 percent, and it further dropped to 21 percent in 2002. Id. at 69,532. But the raw numbers don’t tell the whole tale. FDA concedes it cannot directly attribute any decrease in the Canadian smoking rate to the graphic warnings because the Canadian government implemented other smoking control initiatives, including an increase in the cigarette tax and new restrictions on public smoking, during the same period. Id. Although FDA maintains the data “are suggestive” that large graphic warnings “may” reduce smoking consumption, id., it cannot satisfy its First Amendment burden with “mere speculation and conjecture.” Rubin, 514 U.S. at 487, 115 S.Ct. 1585.
FDA’s Regulatory Impact Analysis (“RIA”)14 essentially concedes the agency *1220lacks any evidence showing that the graphic warnings are likely to reduce smoking rates. One way in which the RIA analyzed the expected benefits of the Rule was by comparing the impact of similar warnings introduced in Canada in 2000. See Final Rule at 36,719-20. It (1) analyzed the change in smoking trends in Canada before and after 2000; (2) assumed any difference in the post-2000 change between Canada and the United States was solely attributable to the introduction of graphic warnings; and (3) assumed similar warnings would have an identical impact on U.S. smoking rates. See id. at 36,755. Describing its approach as “rudimentary,” FDA acknowledged that apart from differences in cigarette taxes, the RIA “d[id] not account for potential confounding variables,” id. at 36,720-21, such as the introduction of more stringent smoking bans and advertising restrictions in Canada during the relevant time period, or the fact that Canadian cigarette prices are generally higher than U.S. prices. Plaintiffs’ Comment Letter on Proposed Rule (Jan. 11, 2010) and Statement of Robert S. Maness.
Logic dictates that these procedural shortcuts would, if anything, lead to an overly optimistic prediction of the efficacy of the proposed graphic warnings. Not so. The RIA estimated the new warnings would reduce U.S. smoking rates by a mere 0.088%, Final Rule at 36,721, a number the FDA concedes is “in general not statistically distinguishable from zero.” Id. at 36,776. Indeed, because it had access to “very small data sets,” FDA could not even reject the statistical possibility that the Rule would have no impact on U.S. smoking rates. Id.
FDA has thus presented us with only two studies that directly evaluate the impact of graphic warnings on actual smoking rates, and neither set of data shows that the graphic warnings will “directly” advance its interest in reducing smoking rates “to a material degree.” Rubin, 514 U.S. at 487, 115 S.Ct. 1585. And one of the principal- researchers on whom FDA relies recently surveyed the relevant literature and conceded that “[t]here is no way to attribute ... declines [in smoking] to the new health warnings.” David Hammond, Health Warnings Messages on Tobacco Products: A Review, 20 Tobacco Control 327, 331 (2011), available at http:// tobaccocontrol.bmj.com/content/20/5/327. full.pdf In light of the number of foreign jurisdictions that have enacted large graphic warning labels, the dearth of data reflecting decreased smoking rates in these countries is somewhat surprising, and strongly implies that such warnings are not very effective at promoting cessation and discouraging initiation. While APA review of final agency action is deferential, it surely does not require us to accept a flawed interpretation of Canadian survey data or the agency’s own projected 0.088% decrease in the U.S. smoking rate as “substantial evidence” that its warnings will advance its stated interest.
FDA attempts to downplay the significance of the RIA by explaining that it “must be included in all federal rule-making to improve the internal management of the Federal Government,” and that it “was not intended to second-guess Congress’s judgment regarding the value of new health warnings.” Pet. Reply Br. at 15-16.15 FDA attempts to rehabilitate *1221its findings by noting the analysis made only the “unremarkable point” that it is “difficult [to] determine with statistical precision the relative causal impact of the relevant contributing factors,” particularly given the very small data sets to which FDA had access. Id. at 16. But FDA cannot get around the First Amendment by pleading incompetence or futility. Because FDA bears the burden of justifying its proposed restraint on speech, it cannot claim — rather perversely — that its own analysis was irrelevant because it lacked precision and was based on insufficient data. Central Hudson requires FDA to find and present data supporting its claims prior to imposing a burden on commercial speech.
Alternatively, FDA asserts an interest in “effectively communicating health information” regarding the negative effects of cigarettes. Appellant’s Br. at 28. But as FDA concedes, this purported “interest” describes only the means by which FDA is attempting to reduce smoking rates: “[t]he goal of effectively communicating the risks of cigarette smoking is, of course, related to the viewer’s decision to quit, or never to start, smoking.” Id. at 47. The government’s attempt to reformulate its interest as purely informational is unconvincing, as an interest in “effective” communication is too vague to stand on its own. Indeed, the government’s chosen buzzwords, which it reiterates through the rulemaking, prompt an obvious question: “effective” in what sense? Allowing FDA to define “effectiveness” however it sees fit would not only render Central Hudson’s “substantial interest” requirement a complete nullity, but it would also eviscerate the requirement that any restriction “directly advance” that interest. See 447 U.S. at 566, 100 S.Ct. 2343. In this case, both the statute and the Rule offer a barometer for assessing the effectiveness of the graphic warnings — the degree to which they encourage current smokers to quit and dissuade would-be smokers from taking up the habit. See Final Rule at 36,630, 36,707-08. As such, FDA’s interest in “effectively communicating” the health risks of smoking is merely a description of the means by which it plans to accomplish its goal of reducing smoking rates, and not an independent interest capable of sustaining the Rule.16
IY. Conclusion
In the Proposed Rule, FDA lamented that their previous efforts to combat the tobacco companies’ advertising campaigns have been like bringing a butter knife to a gun fight. According to the FTC, tobacco companies spent approximately $12.49 billion on advertising and promotion in 2006 alone, employing marketing and advertising experts to incorporate current trends and target their messages toward certain demographics. Proposed Rule at 69,531. The graphic warnings represent FDA’s attempt to level the playing field, not only by limiting the Companies’ ability to advertise, but also by forcing the Companies to bear the cost of disseminating an anti-smoking message. But as the Supreme *1222Court recently reminded us, “[t]hat the [government] finds expression too persuasive does not permit it to quiet the speech or to burden its messengers.” Sorrell v. IMS Health Inc., — U.S.-, 131 S.Ct. 2653, 2671, 180 L.Ed.2d 544 (2011). The First Amendment requires the government not only to state a substantial interest justifying a regulation on commercial speech, but also to show that its regulation directly advances that goal. FDA failed to present any data — much less the substantial evidence required under the APA— showing that enacting their proposed graphic warnings will accomplish the agency’s stated objective of reducing smoking rates. The Rule thus cannot pass muster under Central Hudson. The APA directs that we “shall ... set aside [the] agency action ... found to be contrary to constitutional right.” 5 U.S.C. § 706(2). We therefore vacate the graphic warning requirements and remand to the agency. In so doing, we also vacate the permanent injunction issued by the district court, in furtherance of our obligation to “set aside” the unlawful regulation. See, e.g., N. Air Cargo v. United States Postal Serv., 674 F.3d 852, 861 (D.C.Cir.2012) (“It was quite anomalous [for the district court] to issue an injunction. When a district court reverses agency action and determines that the agency acted unlawfully, ordinarily the appropriate course is to identify a legal error and then remand to the agency, because the role of the district court in such situations is to act as an appellate tribunal.”).
. The Companies also alleged the graphic warnings violated the Administrative Procedure Act ("APA”), specifically 5 U.S.C. §§ 553(b)(3) and 706(2)(A). See Compl. at 37. Because we hold the graphic warnings violate the First Amendment, we do not reach the Companies’ APA claims.
. FDA originally appealed the district court's grant of the Companies’ motion for a preliminary injunction, but that ruling was superseded by the court’s subsequent ruling on the merits.
. Countries/jurisdictions that have implemented pictorial warning requirements for tobacco packaging include Australia, Belgium, Brazil, Brunei, Canada, Chile, Colombia, Cook Islands, Djibouti, Egypt, Hong Kong, India, Iran, Jordan, Latvia, Malaysia, Mauritius, Mexico, Mongolia, New Zealand, Pakistan, Panama, Paraguay, Peru, Romania, Singapore, Switzerland, Taiwan, Thailand, Turkey, United Kingdom, Uruguay, and Venezuela. Countries/jurisdictions with pending requirements include France, Guernsey, Honduras, Malta, Norway, the Philippines, and Spain. It is worth noting that the constitutions of these countries do not necessarily protect individual liberties as stringently as does the United States Constitution. Proposed Rule at 69,525 n. 4.
. Tobacco manufacturers also criticized the "study's use of intentions to measure behavioral change and stated that FDA should have presented data showing actual effects on behavior.” Final Rule at 36,642. FDA disagreed that intentions were an inappropriate variable, explaining that while intentions do not perfectly predict future behavior, they are a “necessary precursor.” Id. FDA also cites the “scientific literature['s]” shocking conclusion “that one's intentions to quit smoking must be increased before one makes the actual quit attempt.” Id. In response to comments raising concerns about the lack of strong statistically significant results concerning intention, FDA explained that although its study made no attempt to show that increased intention to quit translated to actual (let alone successful) quit attempts, "the overall body of scientific literature” provides sufficient evidence that the warnings, “by increasing public understanding of and thoughts about the health risks of smoking, will be effective in encouraging smoking cessation and discouraging smoking initiation.” Id.
. The question here is whether the graphic warnings actually do constitute the type of disclosure requirements that are reviewable under Zauderer's relaxed standard” — what the dissent characterizes as "attempts only to prescribe what shall be orthodox in commercial advertising,” Dissent at 1226 (quoting Zauderer v. Office of Disciplinary Counsel, 471 U.S. 626, 651, 105 S.Ct. 2265, 85 L.Ed.2d 652 (1985)) — or whether they are more akin to *1212attempts to "prescribe what shall be orthodox in ... matters of opinion,” Zauderer, 471 U.S. at 651, 105 S.Ct. 2265, as the Companies contend. The dissent overlooks the element of compulsion, which at least creates an argument in favor of applying strict scrutiny.
. FDA, Tobacco Strategy Announcement (Nov. 10, 2010), available at http://www.fda.gov/ T obaccoProducts/NewsEvents/ucm23 2556. htm; see also Press Briefing by Press Secretary Jay Carney, Health and Human Services Secretary Kathleen Sebelius, and FDA Commissioner Margaret Hamburg (June 21, 2011), available at http://www.whitehouse. gov/the-press-office/2011/06/21/press-briefing-press-secretary-jay-carney-secretary-health- and-human-ser.
. Justice Souter noted that, although it was not cited by the government in Glickman, Zauderer represented “the closest pass at authority for his limited rationale of commercial speech protection” because it was "our only examination of a commercial-speech mandate before today.” 521 U.S. at 490, 117 S.Ct. 2130.
. The dissent also claims that the government has provided “more than sufficient evidence that cigarette packages and other advertisements remain likely to mislead consumers notwithstanding the existing warnings.” Dissent at 1228. In the Final Rule, the FDA found that consumers are uninformed about "the nature and extent of the health risks associated with smoking cigarettes,” Final Rule at 36,632, such as "the severity and magnitude” of those risks, their personal risks, the effects of secondhand smoke, and the highly addictive nature of cigarettes. See id. at 36,632-33. But none of the proposed warnings purport to address the information gaps identified by the government.
. To the extent that there is a concern about the Companies' past deception, the Act precludes them from "portray[ing] the use of tobacco as ... healthful to minors,” see Act § 2(17), 123 Stat. at 1778, by precluding the Companies from using "light” and other descriptors. See 21 U.S.C. § 387k. And Congress's objection to the Companies' portrayal of smoking as "socially acceptable” is likewise remedied by the constraints of the Act.
. Such matters are the subject of a pending- — and entirely separate — line of litigation against the Companies. See Philip Morris USA Inc., 566 F.3d 1095.
. Central Hudson also provides that commercial speech only receives First Amendment protection if it is a lawful activity and is not misleading or fraudulent. 447 U.S.' at 566, 100 S.Ct. 2343. Neither party seriously disputes that the cigarette packaging and advertisements regulated by the Act satisfy this threshold requirement.
. Moreover, the Institute of Medicine Report, on which FDA relies for some of its evidence supporting the Rule, states unequivocally that “the primary objective of tobacco regulation is not to promote informed choice but rather to discourage consumption of tobacco products ... as a means of reducing tobacco-related death and disease.” Institute of Medicine, Ending the Tobacco Problem: A Blueprint for the Nation 291 (2007), available at http://www.nap.edu/catalog.phpPrecord_ id= 11795. The Report goes on to state that “[e]ven though tobacco products are legally available to adults, the paramount public health aim is to reduce the number of people who use and become addicted to these products, through a focus on children and youths,” and recommends that the "warnings must be designed to promote this objective.” Id.
. Like the district court, we are skeptical that the government can assert a substantial interest in discouraging consumers from purchasing a lawful product, even one that has been conclusively linked to adverse health consequences. Nonetheless, the Supreme Court has at least implied that the government could have a substantial interest in reducing smoking rates because smoking poses "perhaps the single most significant threat to public health in the United States.” FDA v. Brown & Williamson Tobacco Corp., 529 U.S. 120, 161, 120 S.Ct. 1291, 146 L.Ed.2d 121 (2000).
. Such an analysis is required under Executive Order 12866, 58 Fed.Reg. 51,735 (Sept. 30, 1993), which directs agencies to assess all costs and benefits of available regulatory al*1220ternatives and, when regulation is necessary, to select the approach that maximizes net benefits.
. FDA also urges us to defer to Congress’s judgment regarding the efficacy of the graphic warnings. See Turner Broad. Sys., Inc. v. FCC. 520 U.S. 180. 196, 117 S.Ct. 1174, 137 *1221L.Ed.2d 369 (1997). But deference is only warranted where Congress "base[s] its conclusions upon substantial evidence,” id., and Congress’s predictive judgments are not “insulated from meaningful judicial review.” Turner, 512 U.S. at 666, 114 S.Ct. 2445. Deference is not appropriate here, because we find little evidence showing that the graphic warnings will advance the stated purpose of the statute — "promoting] cessation to reduce disease risk and the social costs associated with tobacco-related diseases.” Act. § 3.9.
. The dissent accuses us of "choosing to ignore” this interest, thereby ignoring our explanation that the government's stated interest in "effectively” communicating information is illusory absent some barometer for assessing that effectiveness.