AI Claims in Product and Marketing Copy Under Section 5 of the FTC Act

By Hank FasthoffSeptember 11, 2026

A claim that an AI product identifies fraud, achieves a stated accuracy, replaces a professional, or verifies a user’s age makes a factual promise about the product. Your business needs evidence that supports the promise before customers see it. That applies to claims in advertising, sales presentations, app listings, and the product’s interface.

Section 5 of the FTC Act prohibits unfair or deceptive acts or practices in or affecting commerce. Under the FTC’s advertising substantiation policy, advertisers must possess and rely on a reasonable basis for objective claims before making them. The requirement covers both what an advertisement states and what it reasonably implies.

The Evidence Must Match the Claim

The required evidence depends on the claim, the product, the consequences of an inaccurate claim, and what experts consider reasonable, among other factors. If an advertiser represents that a particular kind of testing supports a claim, it must have that support. A numerical accuracy claim requires substantiation for the stated performance, but a percentage alone does not automatically promise a particular study design or independent certification. FTC advertising substantiation policy.

For an AI product, you should examine whether the supporting evidence concerns the version, settings, tasks, and users described in the advertising. A provider’s published evaluation might support a limited statement about that model’s performance on the evaluated task. It does not automatically support the same percentage for your application with different inputs or features.

Testing is central to many performance claims, but the law does not prescribe one universal testing procedure for every objective statement. A statement identifying the model provider may be supported by contracts and configuration records. Claims about diagnostic accuracy, demographic performance, or replacing a professional generally require a different evidentiary basis.

You should also assess the overall message. A headline promising autonomous operation may mislead customers if the product requires substantial human review, even when a qualification appears in a separate document. The FTC’s guidance on digital advertising disclosures explains that necessary qualifications must be noticeable and understandable, and should appear close to the claims they qualify.

Accuracy Claims Require Attention to What Was Measured

In In re Workado, LLC, the FTC challenged advertising that an AI content detector could distinguish human writing from AI writing with approximately 98% accuracy. The complaint alleged that Workado relied on testing involving academic content while promoting the detector for marketing and other ordinary text, including output from models outside the training conditions.

The complaint identified two different results for nonacademic text. The model developers’ best overall classification accuracy was 74.5%, while the model correctly identified AI generated text as AI generated only 53.2% of the time. Those figures measure different aspects of performance, and describing 53.2% as the overall accuracy obscures that distinction.

The FTC’s August 2025 final order required Workado to possess competent and reliable evidence for effectiveness claims and retain the supporting evidence. For other businesses, the practical lesson is to identify what an accuracy figure measures, which inputs were evaluated, and whether the advertised use matches those conditions.

In In re IntelliVision Technologies Corp., the FTC alleged unsupported claims that facial recognition software achieved exceptional accuracy, operated without racial or gender bias, and could not be fooled by photographs or video. The FTC also challenged a claim that the system was trained on millions of faces when the underlying images represented approximately 100,000 distinct individuals and their generated variations.

The January 2025 final order required competent and reliable testing for specified performance claims, including demographic performance and resistance to attempts to fool the system. A claim about performance across groups needs evidence addressing those groups. An overall average does not establish identical performance for each group.

Safety and Labor Savings Claims

In FTC v. Evolv Technologies Holdings, Inc., the FTC alleged that a security screening company overstated its ability to detect weapons, ignore harmless items, outperform conventional metal detectors, and reduce staffing costs by 70%. These were separate performance claims requiring support for the promised results.

The complaint alleged that a scanner failed to detect a knife used in a school stabbing and that increasing the system’s sensitivity resulted in a 50% false alarm rate. A court entered the settlement order in December 2024, prohibiting misleading or unsupported claims and requiring cancellation rights for specified school customers. The settlement resolved the allegations without a trial determining their merits.

Unsafe deployment also presents a separate legal question. In FTC v. Rite Aid Corporation, the FTC alleged that facial recognition deployment without reasonable safeguards caused consumers to be falsely identified and subjected to searches, accusations, or removal from stores. That case involved unfair deployment practices and alleged violations of an earlier security order. Accurate advertising alone does not resolve those operational duties.

Claims That AI Replaces a Professional

In In re DoNotPay, Inc., the FTC alleged that a service advertised as an AI lawyer lacked support for claims that it could substitute for a human attorney. According to the complaint, the company did not test whether its legal output performed at the level of a lawyer and did not retain attorneys to evaluate its legal features.

The January 2025 final order required $193,000 in monetary relief, notice to specified former subscribers, and support for claims comparing the service with professional expertise. You should distinguish claims about assisting with a defined task from claims about replacing a professional’s judgment. A disclaimer cannot reliably cure an advertisement whose overall message promises professional capabilities the product lacks.

Claims about replacing staff may also include earnings or business opportunity representations. In FTC v. Air AI Technologies, Inc., the FTC challenged representations about business growth, earnings, refunds, and service performance under the FTC Act and additional rules governing telemarketing and business opportunities.

In March 2026, the FTC announced a proposed settlement that would ban the defendants from marketing business opportunities. The proposed order provided for an $18 million judgment, largely suspended based on inability to pay, with a $50,000 payment.

State Consumer Protection Laws Apply Too

Texas law prohibits representing that goods or services possess characteristics or benefits they lack, or a quality or standard they do not meet. Its provision addressing nondisclosure also requires knowledge of the omitted information and intent to induce a transaction the consumer otherwise would not enter. An omission does not automatically satisfy those elements. Texas Business and Commerce Code § 17.46(b)(5), (7), and (24).

In State of Texas v. Pieces Technologies Inc., the Attorney General challenged accuracy representations about AI used to assist hospital personnel. The disputed metrics included a severe hallucination rate advertised as less than one per 100,000. A hallucination is generated output that contains incorrect or misleading information. Pieces denied wrongdoing and disputed the state’s allegations.

The agreement required explanations of advertised metrics and their calculation methods, with an alternative involving independent assessment, as well as disclosures to customers about intended uses and limitations. An advertised error rate is meaningful only when the reader understands what the metric measures. You should identify which errors count, the unit being measured, and the conditions under which the result was obtained.

Claims Inside the Product

Interface text can communicate the same factual promises as an advertisement. A signup screen stating that a payment step verifies adulthood represents that the process establishes the user’s age. If the process only authorizes payment, the business needs to revise the statement or implement a process that supports it. Whether the product also violates children’s privacy law requires a separate analysis of the applicable law and facts.

Privacy statements require similar precision. If a business says customer content is never used for training, it should have support covering relevant provider practices, account settings, and contractual commitments. A provider’s permission to train does not establish that training occurred, but it may prevent the business from supporting an absolute promise about future use.

You should include onboarding screens, help articles, settings, and chatbot responses about the service in the claims review. Promises about human access, deletion, identity checks, and security require support for what the product does. A marketing review limited to the homepage will miss those representations.

Disclosing That a User Is Interacting With AI

State disclosure requirements depend on the interaction and the business. In Utah, a supplier using generative AI in a consumer transaction must disclose its use when an individual makes an unambiguous inquiry about whether the interaction involves AI. Individuals providing services in a regulated occupation must make prominent disclosures for qualifying interactions involving sensitive information or advice that could inform significant personal decisions. Those disclosures must occur at the start of verbal interactions and before written interactions. Utah Code §§ 13-77-101 and 13-77-103.

Utah’s safe harbor addresses enforcement of that disclosure requirement. It applies when the system makes the specified conspicuous disclosures at the outset and throughout the interaction. It does not excuse deceptive product claims or other consumer protection violations. Utah Code §§ 13-77-102 and 13-77-104.

California’s bot statute prohibits specified online use of an automated account with intent to mislead someone about its artificial identity for the purpose of knowingly deceiving that person to encourage a commercial transaction or influence a vote. A compliant disclosure avoids liability under that section. The intent and purpose requirements limit its scope. California Business and Professions Code §§ 17940–17941.

California also requires disclosure when a reasonable person would mistake a covered companion chatbot for a human, with additional disclosures for known minors. Companion chatbots are systems capable of meeting social needs and sustaining relationships across interactions. The statutory definition excludes bots used only for specified purposes such as customer service, business operations, and technical assistance. California Business and Professions Code §§ 22601–22602.

Texas requires governmental agencies to disclose AI interactions and requires disclosures when AI is used in health care services or treatment. Health care disclosures are due no later than the first service or treatment, subject to an emergency exception. Section 552.051 does not impose a universal chatbot disclosure requirement on every private business. Texas Responsible Artificial Intelligence Governance Act.

Synthetic Reviews and AI Spokespeople

The FTC’s Consumer Reviews and Testimonials Rule prohibits businesses from creating or selling reviews or testimonials that materially misrepresent whether the speaker exists, used the product, or had the described experience. It also prohibits specified purchasing and dissemination practices when the business knew or should have known the representation was false. Generating the text with AI does not excuse the conduct. 16 C.F.R. § 465.2.

AI assistance does not automatically make a truthful review unlawful. Nor does the rule prohibit every synthetic spokesperson or virtual influencer. The FTC’s published guidance states that there is no blanket prohibition on AI avatars in marketing. Presenting a fabricated customer experience as a real testimonial is different from using an obviously fictional character to deliver an advertising message.

Material connections between an advertiser and an endorser require disclosure when consumers would not reasonably expect them and they could affect the endorsement’s credibility. An AI label does not substitute for disclosing a relevant paid relationship or excuse a false performance claim. FTC Endorsement Guides guidance.

The FTC’s treatment of review generation also changed after its 2024 enforcement sweep. In December 2025, the Commission set aside its order in In re Rytr LLC, concluding that the complaint’s facts did not support finding a Section 5 violation. That order should not be cited as an existing prohibition on providing an AI writing tool. Businesses that fabricate or disseminate false testimonials must separately comply with the review rule.

The review rule permits civil penalties for knowing violations. That authority is distinct from the restitution and disgorgement authority the Supreme Court addressed in AMG Capital Management, LLC v. FTC. In that 2021 decision, the Court held that Section 13(b) did not authorize those forms of monetary relief. The decision did not eliminate monetary remedies available under other provisions of the FTC Act.

Disclosures in Generated Images, Audio, and Video

California’s AI Transparency Act imposes specified duties on providers whose publicly accessible generative AI systems exceed one million monthly users or visitors. Its core provider requirements became operative on August 2, 2026. The law also contains separate obligations for certain hosting platforms and large online platforms beginning January 1, 2027. California AI Transparency Act amendments.

Covered providers must offer detection tools, embed required machine detectable disclosures in generated images, audio, and video, and offer an option for a disclosure people can perceive. Licensing provisions require preservation of the system’s disclosure capability. If a provider discovers that a licensee modified the system to disable the required capability, it must revoke the license within 96 hours. California Business and Professions Code §§ 22757.2–22757.3.

Those duties depend on statutory coverage and the business’s role. A startup’s use of generated artwork does not, by itself, make it a covered provider or impose a universal visible label on every asset. You should assess the applicable statute, the provider’s contract, and whether the presentation would mislead consumers about a person, endorsement, or product characteristic.

Maintaining Support for Product Claims

You should maintain an inventory of objective claims and the evidence supporting each one. The inventory should identify the claim, where it appears, the relevant product version, supporting evaluations or records, and any limitations. Your technical team and appropriate subject specialists should conduct and explain testing, while legal review assesses what the proposed statement communicates and whether the described support addresses it.

Changes to the model, provider, settings, or intended use should trigger a review of affected claims and whether additional testing is needed. You should also review contradictory results and customer reports that call existing representations into question. When the evidence no longer supports an objective claim, the business must stop making that claim unless it obtains adequate support.