The appellants, IMS Health Inc., Verispan, LLC, Source Healthcare Analytics, Inc., and Pharmaceutical Research and Manufacturers of America (“PhRMA”) (collectively, “the appellants”) challenge a Vermont statute banning the sale, transmission, or use of prescriber-identifiable data (“PI data”) for marketing or promoting a prescription drug unless the prescriber consents. In 2007, Vermont enacted the statute at issue, namely Vt. Acts No. 80, § 17 (2007), codified at Vt. Stat. Ann. tit. 18, § 4631 (2007), as amended by Vt. Acts No. 89 (2008) (changing effective date of § 17 from January 1, 2008 to July 1, 2009) (Act 80, “section 17”). The appellants appeal from a judgment of the United States District Court for the District of Vermont (J. Garvan Murtha, Judge) finding section 17 to be a constitutional restriction on commercial speech pursuant to Central Hudson Gas & Electric Corp. v. Public Service Commission of New York, 447 U.S. 557, 561-66, 100 S.Ct. 2343, 65 L.Ed.2d 341 (1980), and finding that section 17 does not violate the Commerce Clause, art. I, § 8, cl. 3, of the United States Constitution.1 IMS Health Inc. v. Sorrell, 631 F.Supp.2d 434 (D.Vt.2009).
On appeal, the appellants argue (1) that section 17 restricts non-commercial speech and cannot withstand strict scrutiny, (2) that even if section 17 restricts only commercial speech, it cannot withstand intermediate scrutiny under Central Hudson, and (3) that section 17 violates the dormant Commerce Clause by prohibiting commerce wholly outside of Vermont. The appellees, Vermont Attorney General William H. Sorrell, Vermont Governor Jim Douglas, and Secretary of the Agency of Human Services of the State of Vermont Robert Hofmann, contend (1) that section 17 does not implicate the appellants’ First Amendment rights, (2) that even if section 17 is a restriction on the appellants’ commercial speech, section 17 survives intermediate scrutiny because it is a narrowly tailored statute that directly advances Vermont’s substantial interest in protecting medical privacy, in controlling health care costs, and in promoting public health, and (3) that the appellants lack standing to challenge section 17 under the dormant Commerce Clause and that, in any event, section 17 does not violate the dormant Commerce Clause because it regulates intrastate commerce.
*267We conclude that because section 17 is a commercial speech restriction that does not directly advance the substantial state interests asserted by Vermont, and is not narrowly tailored to serve those interests, the statute cannot survive intermediate scrutiny under Central Hudson. Therefore, we reverse and remand the judgment of the district court.
BACKGROUND
The Vermont legislature passed Act 80 in 2007, intending to protect public health, to protect prescriber privacy, and to reduce health care costs. Section 17 prohibits the sale, license, or exchange for value of PI data for marketing or promoting a prescription drug, and prohibits pharmaceutical manufacturers and marketers from using PI data for marketing or promoting a prescription drug, unless the prescriber consents. See Vt. Stat. Ann. tit. 18, § 4631(a) & (d). As amended, section 17 was effective on July 1, 2009. See Vt. Acts No. 89 (2008).
I.
When filling prescriptions, pharmacies in Vermont collect information including the prescriber’s name and address, the name, dosage, and quantity of the drug, the date and place the prescription is filled, and the patient’s age and gender. Pharmacies sell this PI data to the data mining appellants IMS Health Inc., Verispan, LLC, and Source Healthcare Analytics, Inc.2 These data mining companies, all located outside of Vermont, aggregate the data to reveal individual physician prescribing patterns and sell it outside of Vermont, primarily to pharmaceutical manufacturers. The PI data sold by the data-mining appellants is stripped of patient information, to protect patient privacy. Appellant Pharmaceutical Research and Manufacturers of America (“PhRMA”) is a non-profit association representing pharmaceutical researchers and manufacturers, the primary customers of the data mining appellants.
Pharmaceutical manufacturers market their products through various means, including advertising and detailing. “Detailing” refers to visits by pharmaceutical representatives, called detailers, to individual physicians to provide information on specific prescription drugs, including the use, side effects, and risks of drug interactions. Pharmaceutical manufacturers use PI data to identify audiences for their marketing efforts, to focus marketing messages for individual prescribers, to direct scientific and safety messages to physicians most in need of that information, to track disease progression, to aid law enforcement, to implement risk mitigation programs, and to conduct clinical trials and post-marketing surveillance required by the United States Food and Drug Administration (“FDA”).
While section 17 in part aims to decrease detailing, prescribers may want to receive the information detailers provide, and, in any event, prescribers are free to decline meetings with detailers.
As the district court noted, pharmaceutical industry spending on detailing has increased exponentially along with the rise of data mining. Detailing is only cost-effective for brand-name drugs. When a *268patent expires, competitors can introduce bioequivalent generic drugs. Bioequivalent generic drugs are not necessarily identical to the brand name version, but are required to demonstrate an absorption rate between 80 and 125 percent of the brand-name drug. Variations in absorption rates among branded or generic drugs may cause different reactions, such as side effects. The district court also noted that while a brand-name drug is not necessarily better than its generic version, the brand-name drug is typically more expensive.
Pharmaceutical manufacturers are not the only entities that purchase PI data from the data mining appellants, although pharmaceutical manufacturers and marketers are the only customers banned from using PI data in their marketing efforts by section 17. The state of Vermont itself uses PI data for law enforcement and other state programs. Researchers use PI data to identify overuse of a pharmaceutical in specific populations, to develop new drugs, and to facilitate identification of potential patients to participate in clinical trials. The FDA, the Center for Disease Control, and the federal Drug Enforcement Agency use PI data to monitor usage of controlled substances and to identify prescribers who need time-sensitive safety information. Insurance companies and pharmacy benefit managers use the data to process claims and manage formulary compliance. Moreover, insurance companies and state governments like Vermont’s use PI data to encourage the use of cheaper, generic medications — the very medications section 17 seeks to promote. While insurance companies and governments collect their own PI data, their databases are not as thorough as those maintained by the data mining appellants. To preserve the value of their data, data mining companies typically restrict re-publication of the data they provide their customers. The appellants argue that the sales covered by section 17 are essential to the ability of the data mining appellants to provide PI data for these other, permitted, uses.
II.
a.
The Vermont law was adopted in the wake of a similar statute that had been enacted in New Hampshire, and shortly before another similar statute adopted in Maine.
In 2006 the New Hampshire state legislature passed a statute prohibiting the transmission or use of patient-identifiable and PI data for most commercial purposes. See IMS Health Inc. v. Ayotte, 490 F.Supp.2d 163, 170-71 (D.N.H.2007), rev’d, 550 F.3d 42 (1st Cir.2008). In relevant part, the statute reads:
Records relative to prescription information containing patient-identifiable and prescriber-identifiable data shall not be licensed, transferred, used, or sold ... for any commercial purpose, except for the limited purposes of pharmacy reimbursement; formulary compliance; care management; utilization review by a health care provider, the patient’s insurance provider or the agent of either; health care research; or as otherwise provided by law. Commercial purpose includes, but is not limited to, advertising, marketing, promotion, or any activity that could be used to influence sales or market share of a pharmaceutical product, influence or evaluate the prescribing behavior of an individual health care professional, or evaluate the effectiveness of a professional pharmaceutical detailing sales force.
N.H.Rev.Stat. Ann. § 318:47-f. The stated intent of the statute, passed without any formal legislative findings, was to protect patient and physician privacy and to *269reduce health care costs. See Ayotte, 490 F.Supp.2d at 171, 177. The United States District Court for the District of New Hampshire found the statute unconstitutional because it restricted commercial speech without directly promoting substantial state interests, and despite the existence of alternative approaches to achieve these interests, in violation of the test for restrictions on commercial speech set out in Central Hudson. See Ayotte, 490 F.Supp.2d at 183.
Maine also enacted a law in 2007 regulating the use of PI data. The legislative findings indicate that the statute was passed to improve public health, to reduce costs, and to protect patient and prescriber privacy. See 22 Me.Rev.Stat. Ann. tit. 22, § 1711-E(1-A, 1-B), invalidated by IMS Health Corp. v. Rowe, 532 F.Supp.2d 153 (D.Me.2007), rev’d, IMS Health Inc. v. Mills, 616 F.3d 7 (1st Cir.2010). The Maine statute prohibits the use of PI data for marketing purposes when the prescriber opts out of its use. In relevant part, it reads:
[A] carrier, pharmacy or prescription drug information intermediary may not license, use, sell, transfer or exchange for value, for any marketing purposes, prescription drug information that identifies a prescriber who has filed for confidentiality protection....
22 Me.Rev.Stat. Ann. tit. 22, § 1711-E(2-A). The United States District Court for the District of Maine found the statute unconstitutional because it did not survive intermediate scrutiny despite the opt-out provision. See Rowe, 532 F.Supp.2d at 182.
While an appeal of the Maine district court decision was pending, the Court of Appeals for the First Circuit reversed the judgment of the New Hampshire district court and upheld the constitutionality of the New Hampshire statute. See Ayotte, 550 F.3d at 64. The majority found that the New Hampshire statute regulated only the conduct of data miners, and therefore did not violate their First Amendment rights. Id. at 50-54. Even if the statute did regulate commercial speech, the majority concluded that it would find that the statute survived intermediate scrutiny. Id. at 54-60. Concurring in the result, Judge Lipez concluded that the statute regulates commercial speech, but that it survived intermediate scrutiny review. Id. at 64-65, 79-102 (Lipez, J., concurring and dissenting).
The Court of Appeals for the First Circuit recently followed its decision in Ayotte. It reversed the District Court’s preliminary injunction in Rowe, and found the Maine statute regulating the use of PI data to be constitutional. Mills, 616 F.3d 7.
b.
In 2007, Vermont passed Act 80, section 17, legislation aimed at restricting the use of PI data in pharmaceutical marketing. The state legislature explained that:
It is the intent of the general assembly to advance the state’s interest in protecting the public health of Vermonters, protecting the privacy of prescribers and prescribing information, and to ensure costs are contained in the private health care sector, as well as for state purchasers of prescription drugs, through the promotion of less costly drugs and ensuring prescribers receive unbiased information.
Vt. Stat. Ann. tit. 18, § 4631(a). The statute adopts an opt-in approach, allowing prescribers to opt in to allow the use of their PI data for marketing purposes. See id. at § 4631(c)(1). Otherwise, the sale or transfer of PI data for marketing purposes, or the use of PI data for marketing *270purposes, is prohibited. The statute provides:
A health insurer, a self-insured employer, an electronic transmission intermediary, a pharmacy, or other similar entity shall not sell, license, or exchange for value regulated records containing prescriber-identifiable information, nor permit the use of regulated records containing prescriber-identifiable information for marketing or promoting a prescription drug, unless the prescriber consents as provided in subsection (c) of this section. Pharmaceutical manufacturers and pharmaceutical marketers shall not use prescriber-identifiable information for marketing or promoting a prescription drug unless the prescriber consents as provided in subsection (c) of this section.
Id. at § 4631(d). Marketing is defined by the statute to include
advertising, promotion, or any activity that is intended to be used or is used to influence sales or the market share of a presci'iption drug, influence or evaluate the prescribing behavior of an individual health care professional to promote a prescription drug, market prescription drugs to patients, or to evaluate the effectiveness of a professional pharmaceutical detailing sales force.
Id. at § 4631(b)(5).
The statute expressly permits the sale, transfer, or use of PI data for multiple other purposes, including the limited purposes of pharmacy reimbursement; prescription drug formulary compliance; patient care management; utilization review by a health care professional, the patient’s health insurer, or the agent of either; health care research; dispensing prescription medications; the transmission of prescription data from prescriber to pharmacy; care management; educational communications provided to a patient, including treatment options, recall or safety notices, or clinical trials; and for certain law enforcement purposes as otherwise authorized by law. See id. at § 4631(e)(1)-(7).
The Vermont state legislature issued thirty-one legislative findings in support of the statute. See Vt. Acts No. 80, § 1 (2007). The findings expressly state the legislature’s intent to interfere with the marketplace of ideas to promote the interests of the state. For example, the findings note that the legislature views the goals of pharmaceutical marketing as “often in conflict with the goals of the state.” Id. at § 1(3). The legislature expressed its concern that the “marketplace for ideas on medicine safety and effectiveness is frequently one-sided,” leading doctors to prescribe “drugs based on incomplete and biased information.” Id. at § 1(4). The legislature therefore found that “[pjublic health is ill served by the massive imbalance in information presented to doctors and other preseribers.” Id. at § 1(6). Section 17 is the state’s attempt to correct what it sees as an unbalanced marketplace of ideas that undermines the state’s interests in promoting public health, protecting prescriber privacy, and reducing health care costs.
III.
The data mining plaintiffs filed suit on August 29, 2007 against the Vermont Attorney General, seeking to enjoin enforcement of the statute prior to its taking effect. In November 2007 the action was consolidated with a suit by PhRMA against the appellees seeking declaratory and injunctive relief. An amended complaint was filed on May 14, 2008. , After a bench trial, the district court denied the plaintiffs’ motions for declaratory and injunctive relief and for summary judgment, and denied as moot the defendants’ mo*271tions for summary judgment. See Sorrell, 631 F.Supp.2d at 464.
The district court found that section 17’s restriction of commercial speech survived intermediate scrutiny under Central Hudson. See Sorrell, 631 F.Supp.2d at 455. The district court likewise found that section 17 did not violate the dormant Commerce Clause of the United States Constitution.3 See id. at 456-59.
The appellants appealed from the judgment of the district court, arguing that section 17 is either a restriction on speech requiring strict scrutiny, or a restriction on commercial speech that does not survive intermediate scrutiny. The appellants also argue that the statute restricts commercial activities outside of Vermont, in violation of the dormant Commerce Clause. The appellees respond that the statute restricts conduct rather than speech, that even if the statute does restrict commercial speech it survives intermediate scrutiny, and that it does not violate the dormant Commerce Clause. Because we find that section 17 is an improper restriction on commercial speech under the test set forth in Central Hudson, we find the statute unconstitutional and reverse and remand.
DISCUSSION
Because this case turns on constitutional issues, our review is de novo. See Boy Scouts of Am. v. Dale, 530 U.S. 640, 648-49, 120 S.Ct. 2446, 147 L.Ed.2d 554 (2000); Melzer v. Bd. of Educ. of the City Sch. Dist. of the City of New York, 336 F.3d 185, 198 (2d Cir.2003).
The appellants’ principal argument is that section 17 violates their rights under the First and Fourteenth Amendments. See U.S. Const. amend. I (“Congress shall make no law ... abridging the freedom of speech.... ”). The First Amendment has been applied against state action by the Fourteenth Amendment. See Gitlow v. New York, 268 U.S. 652, 666, 45 S.Ct. 625, 69 L.Ed. 1138 (1925) (incorporating First Amendment freedom of speech against the states under U.S. Const. amend. XIV). Because the appellees contend that section 17 merely regulates conduct that is not subject to First Amendment protections, it is necessary to determine whether the statute restricts protected speech before determining whether that restriction is permissible under the First Amendment.
I.
The district court found that section 17 is a restriction on speech, and does not merely regulate the appellants’ conduct. See Sorrell, 631 F.Supp.2d at 445-47. The appellees argue that the statute is simply a restriction on a commercial practice. They argue that the data miners are buying and selling a commodity, which can be regulated. They concede that the activities of the pharmaceutical companies who seek to use that information to market prescription drugs is a closer question under the First Amendment, but they contend that the statute is nevertheless a restriction on the commercial conduct of the pharmaceutical companies.
We agree with the district court. The First Amendment protects “[e]ven dry information, devoid of advocacy, political *272relevance, or artistic expression.” Universal City Studios, Inc. v. Corley, 273 F.3d 429, 446 (2d Cir.2001). See also Va. State Bd. of Pharmacy v. Va. Citizens Consumer Council, Inc., 425 U.S. 748, 761-70, 96 S.Ct. 1817, 48 L.Ed.2d 346 (1976) (drug price information in drug advertisements is speech); Universal City Studios, 273 F.3d at 446-49 (computer program is speech). Furthermore, it is plain that speech in a form that is sold for profit is entitled to First Amendment protection. See Va. State Bd., 425 U.S. at 761, 96 S.Ct. 1817.
The Court of Appeals for the First Circuit found that a similar New Hampshire statute was not a restriction on speech, but primarily a restriction on conduct, although it considered the statute only as it affected the activities of data miners rather than pharmaceutical manufacturers. See Ayotte, 550 F.3d at 50-54. The court therefore considered the statute to be “a species of economic regulation,” subject only to rational basis review, which the plaintiffs conceded the law satisfied. See id. at 54.
In Ayotte, the court treated the New Hampshire statute among the narrow categories of regulations restricting speech that are not entitled to First Amendment protection, in the tradition of Chaplinsky v. New Hampshire, 315 U.S. 568, 571-72, 62 S.Ct. 766, 86 L.Ed. 1031 (1942), which found lewd, obscene, profane, libelous, and fighting words to be categories of speech wholly outside the protections of the First Amendment. The Court of Appeals interpreted the New Hampshire statute as principally a regulation of conduct because it “restricts] the ability of data miners to aggregate, compile, and transfer information destined for narrowly defined commercial ends” in a transaction where the “information itself has become a commodity.” Ayotte, 550 F.3d at 52-53. The Court of Appeals thought it would “stretch! ] the fabric of the First Amendment beyond any rational measure” to treat a regulation of information differently from a regulation of “beef jerky” when the information is a product. Id. at 53. The majority of the Court of Appeals concluded that it was consistent with the First Amendment for the “legislature ... to level the playing field not by eliminating speech but, rather, by eliminating the detailers’ ability to use a particular information asset — -prescribing histories — in a particular way.” Id. at 54. However, as the Supreme Court recently affirmed, courts do not have “freewheeling authority to declare new categories of speech outside the scope of the First Amendment.” United States v. Stevens, — U.S.-, 130 S.Ct. 1577, 1586, 176 L.Ed.2d 435 (2010). The obscure distinction between speech and “information asset[s]” is an insufficient basis for giving the government leeway to “level the playing field” subject only to rational basis review.
Here, the legislature explicitly aimed to correct the “massive imbalance in information presented to doctors and other prescribes.” Vt. Acts No. 80 § 1(6). The statute specifically decries that “[t]he marketplace for ideas on medicine safety and effectiveness is frequently one-sided.... ” Id. at § 1(4). The statute is therefore clearly aimed at influencing the supply of information, a core First Amendment concern. Instead of mere rational basis review, the First Amendment teaches that courts should assume that truthful commercial information “is not in itself harmful,” Va. State Bd., 425 U.S. at 770, 96 S.Ct. 1817, and conclude that when a statute aims to restrict the availability of such information for some purposes, that restriction must be judged under the First Amendment.
*273The appellees also argue that the statute only regulates conduct and not speech because the appellants have no First Amendment right to access non-public health records without consent. However, the appellants have not claimed a First Amendment right to obtain information. They challenge the restriction on their ability to purchase and use information otherwise available to them but for the state’s restriction. The statute prevents willing sellers and willing buyers from completing a sale of information to be used for purposes that the state disapproves. Indeed, section 17 does not prohibit the collection of PI data so long as it is not used for purposes that the state has prohibited.
The appellees rely on the Supreme Court’s decision in Los Angeles Police Department v. United Reporting Publishing Corp., 528 U.S. 32, 120 S.Ct. 483, 145 L.Ed.2d 451 (1999). However, that case illustrates why the appellees’ argument is misplaced. In United Reporting, the Supreme Court held that restrictions on access to certain police department information were not facially unconstitutional under the First Amendment. Id. at 34-37, 120 S.Ct. 483. The Supreme Court noted that, “what we have before us is nothing more than a governmental denial of access to information in its possession.” Id. at 40, 120 S.Ct. 483. The Court also noted that “[t]his is not a case in which the government is prohibiting a speaker from conveying information that the speaker already possesses.” Id. In this case, the information is not in the government’s possession. Rather, the state seeks to limit the acquisition and use of information in the hands of pharmacies, data miners, and pharmaceutical companies. This is a case about the extent of the permissible governmental regulation of information in the hands of private actors. It is not a case about a claim by private parties to a First Amendment right to access information in government files.
Because we agree with the district court that the statute restricts protected speech, it is necessary to determine whether section 17 violates the appellants’ First Amendment rights.
II.
The appellants argue that section 17 restricts noncommercial speech, even though PI data is sold for a profit. They argue that the statute should be subject to strict scrutiny. See Bd. of Trs. of State Univ. of N.Y. v. Fox, 492 U.S. 469, 482, 109 S.Ct. 3028, 106 L.Ed.2d 388 (1989) (“Some of our most valued forms of fully protected speech are uttered for a profit.”) The appellees contend, and the district court agreed, that section 17 restricts only commercial speech, and therefore is subject to intermediate scrutiny under the test set out in Central Hudson. See Sorrell, 631 F.Supp.2d at 447-48. The district court noted that PI data has both commercial and noncommercial uses. See Sorrell, 631 F.Supp.2d at 447. The data can be used in research regarding the use of prescription medications, to identify harmful consequences of particular medications, and to warn doctors who have prescribed a particular medication of safety concerns that arise after FDA approval. The data can also be used for the purely commercial purposes of marketing branded prescription drugs.
Section 17 restricts the speech of both the pharmaceutical manufacturers represented by PhRMA, who are prohibited from using Vermont PI data for marketing purposes, and the data mining appellants, who are prohibited from selling or transferring Vermont PI data if the data is to be used for marketing purposes. See Vt. *274Stat. Ann. tit. 18, § 4631(d). We address each in turn.
a.
Section 17 prohibits pharmaceutical manufacturers from using PI data regarding prescriptions written and dispensed in Vermont in their marketing efforts. See id. The statute therefore affects manufacturers’ ability to promote brand-name drugs to doctors through detailing, for example, by making it harder to identify those physicians for whom the message will be most relevant and to tailor the detailing messages based on individual physicians’ prescribing histories.
“The ‘core notion’ of commercial speech is that “which does no more than propose a commercial transaction.’ ” Anderson v. Treadwell, 294 F.3d 453, 460 (2d Cir.2002), quoting Bolger v. Youngs Drug Prods. Corp., 463 U.S. 60, 66, 103 S.Ct. 2875, 77 L.Ed.2d 469 (1983). It cannot be seriously disputed that the primary purpose of detailing is to propose a commercial transaction — the sale of prescription drugs to patients. The manufacturers argue, however, that the detailing message includes fully protected speech, specifically “information regarding medical conditions the prescribers treat and [a manufacturer’s] innovative treatments for those conditions” and that strict scrutiny should apply here because Section 17 restricts commercial speech that is “inextricably intertwined with otherwise fully protected speech.” Riley v. Nat’l Fed’n of the Blind of N.C., Inc., 487 U.S. 781, 796, 108 S.Ct. 2667, 101 L.Ed.2d 669 (1988). However, the mere presence of non-commercial information in an otherwise commercial presentation does not transform the communication into fully protected speech. See, e.g., Bolger, 463 U.S. at 68, 103 S.Ct. 2875 (“We have made clear that advertising which ‘links a product to a current public debate’ is not thereby entitled to the constitutional protection afforded noncommercial speech.”); Bad Frog Brewery, Inc. v. N.Y. State Liquor Auth., 134 F.3d 87, 97 (2d Cir.1998) (holding product label to be commercial speech despite social commentary purportedly communicated by the labeling).
Therefore, although some of the information communicated by detailers might be fully protected in another context, we will analyze section 17 as a restriction on commercial speech with respect to the pharmaceutical manufacturers. See Bolger, 463 U.S. at 68, 103 S.Ct. 2875 (“A company has the full panoply of protections available to its direct comments on public issues, so there is no reason for providing similar constitutional protection which such statements are made in the context of commercial transactions.”).
b.
Section 17 also prohibits data miners from selling or transmitting PI data regarding prescriptions written and dispensed in Vermont if that PI data will later be used for marketing purposes. See Vt. Stat. Ann. tit. 18, § 4631(d). Data miners do not themselves use PI data in their own marketing efforts. Rather, data miners are in the business of aggregating and selling the data to pharmaceutical manufacturers, among other entities, so that pharmaceutical manufacturers can use the data in their marketing strategies. The data miners’ regulated speech is therefore one step further removed from the marketing goals of the pharmaceutical manufacturers, although it remains a necessary step in the pharmaceutical manufacturers’ marketing efforts.
The sale of information is protected by the First Amendment, and is not necessarily commercial speech. See, e.g., Univer*275sal City Studios, 273 F.3d at 446-58 (finding computer program is speech, and not scrutinizing it under the commercial speech doctrine). However, unlike the data miners’ sale of PI data here, the computer program in Universal City Studios was not a step in a chain intended to influence marketing efforts.
Because this Court finds that section 17’s restriction on data miners cannot survive even the lower intermediate scrutiny that applies to regulations of commercial speech, we assume without deciding that the statute restricts the data mining appellants’ commercial speech.
III.
Under Central Hudson, the government may regulate commercial speech when (1) “the communication is neither misleading nor related to unlawful activity;” (2) the government “assert[s] a substantial interest to be achieved” by the regulation; (3) the restriction “must directly advance the state interest;” and finally (4) “if the governmental interest could be served as well by a more limited restriction on commercial speech, the excessive restrictions cannot survive.” Central Hudson, 447 U.S. at 564, 100 S.Ct. 2343. There is no allegation that the commercial speech regulated by section 17 is either misleading or related to an unlawful activity. Therefore, for the statute to survive intermediate scrutiny, the government must assert a substantial state interest that is directly advanced by the statute, and the regulation must not be more extensive than necessary to achieve the government’s interest.
a.
The second prong of Central Hudson requires that the state “assert a substantial interest to be achieved by restrictions on commercial speech.” Id. Vermont alleges that section 17 advances three substantial state interests: (1) “the state’s interest in protecting the public health,” (2) “protecting the privacy of prescribers and prescribing information,” an interest the state sometimes also refers to as an interest in protecting “medical privacy,” and (3) the state’s interest in containing health care costs in both the private and public sectors. See Vt. Stat. Ann. tit. 18, § 4631(a).
The district court found that Vermont’s cost containment and public health interests were substantial government interests to justify the statute. Sorrell, 631 F.Supp.2d at 449-50. The court found that it was unnecessary to consider whether protecting prescriber privacy was also a substantial government interest. Id. at 450. The appellants do not seriously dispute that the state has a substantial interest in protecting public health and containing health care costs, although the appellants do argue that section 17 does not directly advance these substantial state interests.
The parties dispute whether protecting the privacy of prescribers and prescribing information is a substantial state interest. Section 17 itself refers to “protecting the privacy of prescribers and prescribing information,” but the statute plainly does not protect physician privacy. Vt. Stat. Ann. tit. 18, § 4631(a). Physician privacy might be protected if the statute prohibited the collection and aggregation of PI data for any purpose, or if the use of such data were permitted in only rare and compelling circumstances. The statute at issue here, however, does not forbid the collection of PI data in the first instance. Furthermore, the statute does not ban any use of the data other than for marketing purposes, including widespread publication to the general public. There is nothing in the statute that would prevent the use of *276such data for journalistic reports about physicians.
Vermont contemplates that the data will still be collected and used, albeit for purposes other than marketing. For example, the state acknowledges that the statute permits the use of PI data for “health care research, treatment, and safety-related uses.” The statute only imposes restrictions on the sale or use of such data for marketing or promoting a prescription drug. Vermont does not explain how the continued collection of PI data, and its use for non-marketing purposes, is compatible with an alleged interest in protecting physician privacy. Indeed, the concern that patient information can be gleaned from PI data is not reduced in any way by section 17, and the statute does not prohibit wide public dissemination of PI data.
The appellees argue that the state’s interest in privacy is “that pharmaceutical marketers should not be exerting undue influence and intruding on the doctor-patient relationship” by marketing prescription drugs using PI data. According to this argument, the state has an interest in preventing pharmaceutical manufacturers from using PI data to persuade doctors to prescribe brand-name medications “because patient care can be compromised [and] because patient trust in the health care system is undermined.” Therefore, what the appellees refer to as “medical privacy” is actually two distinct interests. The first is an interest in the integrity of the prescribing process itself, and the second is an interest in preserving patients’ trust in their doctors by preventing patients from believing that their physicians are inappropriately influenced by PI data-driven marketing.
However, the state’s asserted interest in medical privacy is too speculative to qualify as a substantial state interest under Central Hudson. Intermediate scrutiny requires that the state “demonstrate that the harms it recites are real.” Rubin v. Coors Brewing Co., 514 U.S. 476, 487, 115 S.Ct. 1585, 131 L.Ed.2d 532 (1995). On the record in this case, Vermont has not shown any effect on the integrity of the prescribing process or the trust patients have in their doctors from the use of PI data in marketing. Vermont’s own expert was unaware of any instance in which a detailing interaction caused a doctor to prescribe an inappropriate medication. To the extent that the record might suggest PI data has damaged the relationship between doctors and patients, the evidence is either speculative or merely indicates that some doctors do not approve of detailing or the use of PI data in detailing. For example, Vermont’s expert witness Dr. David Grande opined that the use of PI data “will make patients only feel more anxious about whether or not in fact their interests are being put first,” but he had not conducted any studies of patient perception of PI data to support that conclusion.
Therefore, we agree with the district court that Vermont does have a substantial interest in both lowering health care costs and protecting public health. However, the state’s asserted interest in “medical privacy” is too speculative to satisfy the second prong of Central Hudson.
b.
The third prong of Central Hudson requires that the regulation “directly advance the state interest involved.” Cent. Hudson, 447 U.S. at 564, 100 S.Ct. 2343; see also Edenfield v. Fane, 507 U.S. 761, 767, 113 S.Ct. 1792, 123 L.Ed.2d 543 (1993) (describing third prong of Central Hudson as “whether the challenged regulation advances these interests in a direct and material way”). “It is well established that ‘the party seeking to uphold a restriction *277on commercial speech carries the burden of justifying it.’” Edenfield, 507 U.S. at 770, 113 S.Ct. 1792 (alteration omitted) (quoting Bolger v. Youngs Drug Prods. Corp., 463 U.S. 60, 71 n. 20, 103 S.Ct. 2875, 77 L.Ed.2d 469 (1983)). This prong is “critical” and requires invalidating a regulation that restricts commercial speech “ ‘if it provides only ineffective or remote support’ ” for the government’s interest. Greater New Orleans Broad. Ass’n v. United States, 527 U.S. 173, 188, 119 S.Ct. 1923, 144 L.Ed.2d 161 (1999) (quoting Cent. Hudson, 447 U.S. at 564, 100 S.Ct. 2343).
The Vermont statute cannot be said to advance the state’s interests in public health and reducing costs in a direct and material way. Section 17 can advance the state interests in protecting public health and reducing health costs only by the following route: the statute prevents PI data from being transferred from data miners to pharmaceutical manufacturers for marketing purposes, who in turn are prevented from using the data in their marketing efforts. Failure to use PI data in marketing results in less effective marketing for brand-name prescription drugs, some of which — although not all — are more expensive yet provide no therapeutic advantage over generic alternatives. Less effective marketing will result in doctors writing fewer prescriptions for brand-name prescription drugs, thereby reducing health care costs and protecting public health by minimizing prescriptions for more expensive or less tested medications. The state’s own explanation of how section 17 advances its interests cannot be said to be direct. The statute does not directly restrict the prescribing practices of doctors, and it does not even directly restrict the marketing practices of detailers. Rather, it restricts the information available to detailers so that their marketing practices will be less effective and less likely to influence the prescribing practices of physicians.
The appellees have failed to cite to any case from the Supreme Court or this Court that has upheld a regulation on speech when the government interest in the regulation is to bring about indirectly some social good or alter some conduct by restricting the information available to those whose conduct the government seeks to influence. Cf. Cent. Hudson, 447 U.S. at 566 n. 9, 100 S.Ct. 2343 (“We review with special care regulations that entirely suppress commercial speech in order to pursue a nonspeech-related policy.”). Regulations of conduct are permitted, but only if the government interest is “unrelated to the suppression of free expression.” United States v. O’Brien, 391 U.S. 367, 377, 88 S.Ct. 1673, 20 L.Ed.2d 672 (1968). However, the legislative findings are explicit that Vermont here aims to do exactly that which has been so highly disfavored— namely, put the state’s thumb on the scales of the marketplace of ideas in order to influence conduct. The legislature found that the “marketplace for ideas on medicine safety and effectiveness is frequently one-sided in that brand-name companies invest in expensive pharmaceutical marketing campaigns to doctors. The one-sided nature of the marketing leads to doctors prescribing drugs based on incomplete and biased information.” Vt. Acts No. 80, § 1(4). In other words, the statute seeks to alter the marketplace of ideas by taking out some truthful information that the state thinks could be used too effectively.
The state’s approach to regulating the interaction between detailers and doctors is premised on limiting the information available to physicians as a means of impacting their conduct. This approach is antithetical to a long line of Supreme *278Court cases stressing that courts must be very skeptical of government efforts to prevent the dissemination of information in order to affect conduct. See 44 Liquormart, Inc. v. Rhode Island, 517 U.S. 484, 503, 116 S.Ct. 1495, 134 L.Ed.2d 711 (1996) (“The First Amendment directs us to be especially skeptical of regulations that seek to keep people in the dark for what the government perceives to be their own good.”); Va. State Bd., 425 U.S. at 770, 96 S.Ct. 1817 (alternative to ban on pharmacist advertising “is to assume that this information is not in itself harmful, that people will perceive their own best interests if only they are well enough informed, and that the best means to that end is to open the channels of communication rather than to close them.”). Even if section 17 is successful in altering the conduct of physicians in their prescribing practices, the Supreme Court reminds us that “[i]t is precisely this kind of choice, between the dangers of suppressing information, and the dangers of its misuse if it is freely available, that the First Amendment makes for us.” Va. State Bd., 425 U.S. at 770, 96 S.Ct. 1817; see also Thompson v. W. States Med. Ctr., 535 U.S. 357, 373, 122 S.Ct. 1497, 152 L.Ed.2d 563 (2002) (“If the First Amendment means anything, it means that regulating speech must be a last — not first — resort.”).
The appellees place extensive reliance on Anderson v. Treadwell, 294 F.3d 453 (2d Cir.2002). In Anderson, this Court upheld a New York statute banning in-person real estate solicitations of homeowners in certain zones designated by the Secretary of State if the homeowner indicated that the homeowner did not wish to receive such solicitations. Id. at 456-58. The statute was designed to prevent “blockbusting” — the practice of obtaining real estate listings by emphasizing that a neighborhood is undergoing a religious, racial, or ethnic change. Id. at 457. However, this Court upheld the statute on the basis of the government interest in protecting the privacy of homeowners from harassing real estate solicitations, an interest that is not present here. See id. at 461. The statute in Anderson directly regulated the potentially harassing sales calls. It directly targeted the harassing visits that were viewed as problematic. The statute in Anderson did not ban any entity from transmitting marketing data that would be useful to real estate agents in deciding which homeowners to target. It did not seek to affect the conduct of homeowners by limiting the information available to them. In contrast, section 17 does not ban detailing, even when that detailing is seen as harassment by an individual physician. It does not even restrict such detailing. The opt-in provision of section 17 does not make the statute comparable to the statute in Anderson. The opt-in provision in the Vermont statute relates solely to a physician’s agreement that the physician’s PI data can be used. Physicians in Vermont can always choose to decline to be visited by detailers, even without section 17. The opt-in provision in the statute in Anderson was a consent to be solicited by real estate licensees, not a consent to have information used.4
*279Because section 17 is an attempt to influence the prescribing conduct of doctors by restricting the speech of others — namely data miners and pharmaceutical manufacturers — it does not directly advance the state’s interests in protecting public health and reducing health care costs. Instead, the statute restricts protected speech when uttered for purposes the government does not approve of in order to reduce the effectiveness of marketing campaigns and, ultimately, alter the behavior of prescribers, who are not regulated by the statute. This route is too indirect to survive intermediate scrutiny.
c.
Section 17 also fails under the final prong of Central Hudson, which requires invalidating the restriction “if the governmental interest could be served as well by a more limited restriction on commercial speech.” 447 U.S. at 564, 100 S.Ct. 2343.
The Government is not required to employ the least restrictive means conceivable, but it must demonstrate narrow tailoring of the challenged regulation to the asserted interest — “a fit that is not necessarily perfect, but reasonable; that represents not necessarily the single best disposition but one whose scope is in proportion to the interest served.”
Greater New Orleans Broad., 527 U.S. at 188, 119 S.Ct. 1923 (quoting Fox, 492 U.S. at 480, 109 S.Ct. 3028). The burden is on the government to show that it “carefully calculated” costs and benefits of burdening speech. Id. While the fit need not be perfect, “if the Government could achieve its interests in a manner that does not restrict speech, or that restricts less speech, the Government must do so.” Thompson, 535 U.S. at 371, 122 S.Ct. 1497.
The regulation at issue here applies to all brand name prescription drugs, irrespective, for example, of whether there is a generic alternative or whether an individual drug is effective or ineffective. This is a poor fit with the state’s goal to regulate new and allegedly insufficiently tested brand-name drugs in cases where there áre cheaper generic alternatives available. The statute targets the use of PI data to market all brand name prescription drugs, not merely new brand-name drugs or those brand-name medications for which there are generic alternatives.
The appellees argue that the Court should defer to the legislative determination that the statute is a reasonable fit so long as that determination is itself reasonable. The appellees rely on this Court’s recent decision in Clear Channel Outdoor, Inc. v. City of New York, 594 F.3d 94, 104 (2d Cir.2010), for the proposition that this Court should defer to a government’s reasonable determination regarding how to regulate commercial speech. However, reliance on Clear Channel is misplaced because that decision specifically addresses a regulation of commercial billboards, a distinctive method of speech that poses unique problems such as the potential to distract drivers and is therefore particularly amenable to government regulation. See id. at 108. This Court stressed the *280particular government interests involved in “ ‘the law of billboards.’ ” Id. (quoting Metromedia, Inc. v. City of San Diego, 453 U.S. 490, 501, 101 S.Ct. 2882, 69 L.Ed.2d 800 (1981)).
In any event, we need not decide what level of deference is appropriate here. The statute prohibits the transmission or use of PI data for marketing purposes for all prescription drugs regardless of any problem with the drug or whether there is a generic alternative. The statute bans speech beyond what the state’s evidence purportedly addresses. It seeks to discourage detailing about new brand-name prescription drugs which may not be efficacious or which may not be more effective than generic alternatives. However, it does that by precluding the use of PI data for the marketing of any brand-name prescription, no matter how efficacious and no matter how beneficial those drugs may be compared to generic alternatives. Even if the Court defers to the legislature’s determinations, those determinations cannot support banning speech in circumstances that the state’s evidence does not address. The fact that section 17 sweeps beyond Vermont’s interests in public health and health care costs undermines the state’s argument that the statute is a reasonable fit with its interests.
Moreover, Vermont does have more direct, less speech-restrictive means available. The state could wait to assess what the impact of its newly funded counter-speech program will be, including academic detailing and sample generic vouchers. The state could mandate the use of generic drugs as a first course of treatment, absent a physician’s determination otherwise, for all those patients receiving Medicare Part D funds. All of these means could be targeted at new brand-name drugs particularly when there are alternatives available, unlike section 17’s approach that applies to every prescription drug regardless of whether it is a less tested version of an existing medication or a breakthrough drug with no reasonable alternative. All of these alternative means would directly promote the state’s interests, although they would do so without impacting First Amendment rights.
The district court found that section 17 satisfied the narrow tailoring requirement of Central Hudson because the statute allows prescribers to determine how their PI data would be used, just as the statute at issue in Anderson allowed homeowners to determine whether they would receive solicitations from real estate agents. See Sorrell, 631 F.Supp.2d at 455 (citing Anderson, 294 F.3d at 462). We reject the comparison of section 17 with the statute at issue in Anderson, for the reasons explained above. Moreover, the district court did not consider whether there are any reasonable alternatives that would be less speech-restrictive than section 17. While we agree with the district court that Central Hudson does not require the state to use the least restrictive means available to it to achieve its goals, this Court has examined the available alternatives in other cases to determine whether there was a reasonable fit between the regulation and the state’s asserted interests. See N.Y. State Ass’n of Realtors, Inc. v. Shaffer, 27 F.3d 834, 844 (2d Cir.1994) (invalidating regulation banning real estate brokers from soliciting residential property owners in certain designated areas when defendant failed to provide empirical evidence regarding whether less speech-restrictive approaches would sufficiently promote the asserted government interests).
The state argues that section 17 is narrow because it does not ban detailing and is therefore narrower than speech restrictions that have been struck down. See Ayotte, 550 F.3d at 53; id. at 97 (Lipez, J., *281concurring). The district court agreed with this reasoning. See Sorrell, 631 F.Supp.2d at 455. The statute may be narrow in the sense that it does not prohibit detailing and does not proscribe any particular claim or message. However, the statute does ban a set of messages that Vermont itself contends are particularly effective, namely, messages informed by PI data, and curbs the ability of pharmaceutical manufacturers to market brand-name drugs.
Vermont argues that, unlike other regulations that have been struck down, the statute at issue here does not ban an entire category of speech because doctors can permit their own PI data to be transmitted and used for marketing purposes. Cf. Alexander v. Cahill, 598 F.3d 79, 96 (2d Cir.2010) (finding statute banning potentially, but not actually, misleading use of nicknames in attorney advertising an unconstitutional regulation of commercial speech). However, the mere fact that the statute does permit doctors to choose to make their PI data available for marketing purposes, even if a substantial number of doctors would do so, “does not render the disputed provisions any less categorical.” See id. The statute bans the transmission or use of PI data for marketing purposes, unless the prescriber consents, without regard to whether the data pertains to a prescription drug that is efficacious and whether or not it has a generic alternative. It is the fact that the statute does not distinguish between brand-name drugs, no matter how unique and efficacious, that renders the statute a categorical ban.
The appellees failed to explain how section 17 is no more extensive than necessary to serve its asserted interests in health care costs and public health, or why the proposed alternatives would be inadequate. The state did present limited testimony at trial relating to these alternatives. For example, Dr. Aaron Kesselheim testified that the pharmaceutical industry’s total annual detailing budget was approximately $8 billion and that it was not realistic for Vermont to spend this amount on academic detailing. Dr. Kesselheim also testified that “[formularies, step therapy, and prior authorization] have been in place ... for a few years [but] ... we still see ... overuse of products that potentially place patients at risk.” However, the testimony fell far short of demonstrating that the alternatives would be inadequate. Therefore, section 17 cannot survive Central Hudson scrutiny because Vermont did “not offer[ ] any reason why these possibilities, alone or in combination, would be insufficient to [achieve the government’s interests].” Thompson, 535 U.S. at 373, 122 S.Ct. 1497.
Vermont does argue in its brief that the statute is narrowly tailored because it “focuses on the specific problem identified by the Legislature: the use of [PI data] to fuel marketing campaigns.” However, this argument is not responsive to the inquiry under Central Hudson. Vermont has not asserted a substantial state interest in curbing the use of PI data in marketing campaigns. To satisfy the final prong of Central Hudson, Vermont must show that section 17 is narrowly tailored to serve the substantial state interests that it contends justify the speech restriction — containing health care costs and protecting public health.
Because the statute restricts speech even with regard to prescriptions of breakthrough brand-name medications for which there are no generic alternatives, and because the state could pursue alternative routes that are directly targeted at encouraging the use of generic drugs the state wishes to promote, the state has not demonstrated that its interests in protecting public health and containing health care *282costs could not be as well served by a more limited restriction on speech. Therefore, section 17 cannot survive intermediate scrutiny and is an unconstitutional regulation of commercial speech under the test set forth in Central Hudson.5
CONCLUSION
For the reasons explained above, we reverse and remand the judgment of the district court.
. The district court also upheld sections 20 and 21 of Act 80, and the appellees do not challenge those holdings on appeal.
. The appellants describe themselves as "publishers,” a term that plainly furthers their First Amendment argument. The district court referred to the appellants as “data miners,” a term that has been used in other cases. It is undisputed that the appellants collect and pass on information. Their rights depend on what they do rather than what they are called. This opinion will follow the description used by the district court, namely "data miners.”
. The district court also upheld sections 20 and 21 of the Act, creating a program funded by a fee on pharmaceutical manufacturers to educate health care professionals concerning therapeutic and cost-effective utilization of prescription medications, and creating a consumer fraud cause of action for advertisements in Vermont that violate federal law. See Vt. Stat. Ann. tit. 33, § 2004 & tit. 9, § 2466a; Sorrell, 631 F.Supp.2d at 462, 464. The appellants do not dispute these holdings on appeal, and we do not address them here.
. Anderson is consistent with those cases that have approved procedures for unwilling listeners to decline to receive speech as less restrictive regulations than those preventing speech unless a listener has affirmatively chosen to receive such messages. See, e.g., Martin v. City of Struthers, Ohio, 319 U.S. 141, 147-49, 63 S.Ct. 862, 87 L.Ed. 1313 (1943) (invalidating ban on door-to-door solicitation while noting that regulation banning solicitation when homeowner has indicated a desire not to be disturbed is appropriate); see also Mainstream Mktg. Servs., Inc. v. F.T.C., 358 F.3d 1228, 1242-43, 1246 (10th Cir.2004) (upholding “do not call'' list as constitutional restriction on commercial speech in part be*279cause consumers actively joining “do not call” registry before commercial telephone calls are barred is less restrictive of speech than requiring consumers to consent to receiving such calls before they could be made).
The Court of Appeals for the First Circuit recently noted that Maine's statute was similar to "do not mail lists” because prescribers are entitled to have their information protected from disclosure only if they choose to seek confidentiality protections. Mills, 616 F.3d at 21-22. The Vermont statute at issue in this case, however, uses the broader approach of prohibiting the designated uses of PI data unless a prescriber affirmatively chooses to have that prescriber information made available.
. The appellants also argue that section 17 violates the dormant Commerce Clause because it restricts commerce outside Vermont. Because we find section 17 unconstitutional pursuant to the Central Hudson test, we need not reach this argument.