How Section 5 Reads an AI Claim
The FTC’s Section 5 framework gives the consumer-deception question a structure: identify what the provider represented, ask how a reasonable consumer would understand it, and determine whether the point could affect a decision about the service. Applied to AI, that structure moves beyond the slogan “the model can be wrong.” It helps readers distinguish an implied capability claim from a literal sentence, match evidence to the scope of the promise, and judge whether a disclaimer actually changes the impression created by the product.
The framework matters because AI claims often compress several promises into one word. “Accurate” may suggest reliable performance, an honest objective, or both. “AI lawyer” may imply professional knowledge, legal judgment, and the ability to substitute for a human lawyer. A provider may intend a narrower meaning, but consumer-protection analysis is concerned with the meaning that the overall presentation is likely to convey in context.
The three questions behind a deceptive-practice finding
The FTC’s proposed policy statement describes the established deception framework as having three linked elements. First, there must be a representation, omission, or practice likely to mislead. Second, the practice is assessed from the perspective of a consumer acting reasonably in the circumstances. Third, the misleading point must be material—likely to affect the consumer’s conduct or decision about a product or service. The statement also notes that the framework reaches implied misrepresentations and half-truths, not only statements that are plainly false word for word.
These elements are connected, but they are not interchangeable. A statement can be technically true and still contribute to a misleading overall impression. A claim can sound impressive without being material if it would not affect what a consumer does. A fact can be important in the abstract but not misleading if the provider presented it clearly and the reasonable consumer would understand the limitation. The analysis has to hold all three questions together.
Start with the representation, omission, or practice. In an AI product, the representation may be a direct claim about accuracy or efficacy. It may be a product name, a comparison with a professional, a demonstration, or repeated language describing the system as a reliable problem-solving tool. An omission may involve a material limitation or a competing objective that changes how the consumer would understand the claim. The relevant practice can also be the system’s behavior when it departs from the objective that the provider led users to expect.
The next question is the reasonable consumer’s interpretation. This does not mean asking whether the most skeptical reader would trust nothing, or whether the most credulous user would believe everything. It asks what a consumer acting reasonably in the circumstances would take away from the communication and the product setting. Context can include the wording of the claim, its prominence, the kind of task being offered, the audience, the surrounding explanations, and the consequences of relying on the result.
For AI, the context includes the ordinary purpose of an answer-producing service. A product marketed for research or advice may create an expectation that it aims at a useful and accurate answer, even if it disclaims perfection. A product described as an assistant may invite a different level of reliance from one described as a replacement for a licensed professional. The words do not operate in isolation. The product’s role, the provider’s demonstrations, and the user’s reason for seeking help can combine into an implied claim.
Materiality supplies the practical connection to consumer behavior. A misleading point is material when it is likely to affect conduct or a decision about the product. That can include whether a person pays for a service, chooses it over another option, uses it for a particular task, or relies on its output in deciding what to do. The supplied sources do not give a universal numerical threshold for materiality in AI cases. The point is contextual: the difference between the claim and reality matters because it can change behavior, not merely because an evaluator dislikes the wording.
This is why a narrow factual correction may be less important than the role a claim assigns to the system. If a tool is advertised as a way to organize questions before a professional consultation, a user may understand that it is assistive. If the same tool is presented as capable of performing the professional service itself, the claim can affect whether the user seeks independent help. The underlying software might be similar, but the consumer decision and foreseeable reliance are different. A legal conclusion would require the actual facts, but the framework explains why role and context matter.
A further complication is that a provider may not need to say something literally false for the overall presentation to mislead. Implied representations and half-truths matter because consumers form a net impression. A statement such as “our system uses advanced AI” may be literally true but say little about accuracy. Combined with demonstrations, professional imagery, or confident claims about outcomes, it may contribute to an inference that the system has been tested for a level of capability the provider has not established. That inference is not proven merely by a marketing style; it must be assessed against the whole communication and the evidence available.
The FTC statement presents the three elements as longstanding consumer-protection principles, while its AI-specific discussion is a proposed policy position. The difference in authority matters. The statement is direct evidence of what the Commission proposes to clarify, not a final rule that resolves every future dispute. The scholarly report on DoNotPay is a secondary analysis of a particular enforcement action and policy choices. It can illuminate the case and the policy debate, but it should not be treated as if it were an additional binding source of law.
The sequence can be expressed as a practical reading of the framework:
The diagram is not an official FTC checklist. It is a way to keep the questions in order. Evidence does not replace the misleadingness inquiry, and a disclosure does not erase materiality. The sequence prevents a provider from answering a consumer-expectation question with a benchmark that measures something else, or answering an evidence question with a warning that merely says the product is imperfect.
Evidence must match the promise
The central substantiation principle is simple: evidence must support the claim actually being made. A demonstration that an AI system can generate a document does not, by itself, support a claim that the document is legally valid. A benchmark on a narrow task does not automatically support a general statement that the system is accurate for research, advice, or professional work. Evidence has to match the task, comparison, conditions, and degree of confidence communicated to consumers.
This is especially important for comparative or professional-equivalence claims. Saying that a tool helps a user prepare a form is not the same as saying that it performs like a human lawyer. The latter claim raises questions about what performance is being compared, which legal tasks are included, what jurisdictions or facts are covered, how current the relevant material is, how errors are measured, and whether qualified professionals reviewed the system’s behavior. The supplied evidence does not establish a universal benchmark for every profession. It does establish why generic evidence cannot automatically substantiate a broad professional claim.
A useful way to test an accuracy or efficacy claim is to ask four related questions. What exact task was tested? Against what comparator? Under what conditions? And what limits were communicated to the consumer? These questions do not create a new legal test, and they cannot predict liability without the actual facts. They do expose a common mismatch: the evidence may support a narrow result while the marketing invites a broad conclusion.
The same reasoning applies to claims about the system’s objective. A provider may have evidence that a model performs well on a benchmark while failing to explain that the production system is subject to a material priority that changes how it handles certain questions. Conversely, a provider may clearly disclose a constraint but still lack evidence for a claim about accuracy or efficacy. Objective transparency and performance substantiation address different parts of the consumer’s expectation. One cannot be used as a substitute for the other.
This is also where omissions become important. Consumers do not need a complete inventory of every internal rule before using an AI service. The issue is whether the omitted fact would change the impression created by the claim. If a provider prominently presents a system as pursuing the user’s objective, but a material competing priority predictably changes the answer, the absence of that information may matter. The FTC’s proposed statement does not say that every safety, legal, or quality constraint must be disclosed in exhaustive detail. It focuses on whether the provider has left consumers with a misleading understanding of the system’s purpose.
Disclosures can change that understanding, but only if they are capable of reaching the consumer’s actual expectation. The FTC statement says an adequate disclosure would need to be clear and conspicuous enough to explain that the system prioritizes objectives different from what users request or otherwise expect. It also says that prominence and persistence matter. A disclosure buried in terms of service may not shift a belief created by prominent marketing. A one-time notice hidden in fine print may be inadequate when users encounter the contrary claim repeatedly or in a more visible setting.
The reason is the net impression. A consumer does not necessarily process a product page as a lawyer would parse a contract. They may notice a headline, demonstration, product name, or repeated promise before reading a qualification. If the main message says “dependable professional replacement” and the qualification says “may not always work,” the qualification may not correct the stronger implication. The FTC statement draws on the general principle that a less prominent disclosure may not remedy a more prominent misrepresentation.
That principle should not be turned into a rule that disclaimers are useless. A clear disclosure can narrow a claim when it appears where the consumer forms the expectation and directly explains the relevant limit. It may turn a claim of professional replacement into a claim of document assistance, or a claim of general accuracy into a narrower statement about a defined task. But the disclosure must do real interpretive work. “Results may vary” does not explain which results, under what conditions, or how the limitation changes the product’s role.
Nor does disclosure cure a lack of substantiation. If a provider claims that an AI service performs a professional function at human-equivalent quality, the provider still needs evidence relevant to that claim. A warning that the service can make mistakes may reduce the apparent certainty of the promise, but it does not show that the promised comparison was tested. The consumer must receive both an accurate account of the product’s limits and an honest account of what the evidence establishes.
The most defensible conclusion is therefore narrower than “AI marketing is dangerous” and stronger than “AI mistakes are unavoidable.” Section 5 analysis asks whether the provider’s representation or practice is likely to mislead a reasonable consumer and whether the point matters to the consumer’s decision. In AI settings, that inquiry requires attention to the implied role of the system, the objective it appears to pursue, and the evidence supporting the claimed capability. A documented case involving an AI service marketed as a professional replacement puts those questions under especially visible pressure, because the label itself can raise the stakes of reliance.