Is There an AI Exemption From Advertising Law?

What did the FTC actually say?

That existing law applies, without modification, to anything sold with AI attached. The announcement of Operation AI Comply puts it in a sentence: "there is no AI exemption from the laws on the books."

The accompanying statement from then-Chair Lina M. Khan is equally short: "Using AI tools to trick, mislead, or defraud people is illegal."

The framing of the sweep is worth quoting because it names the behaviour rather than the technology. The Commission describes firms that "have seized on the hype surrounding AI" in order to "lure consumers into bogus schemes".

That distinction runs through everything below. The enforcement is not about artificial intelligence; it is about claims that cannot be substantiated, made in a period when a particular word made extravagant claims easier to sell.

The timing is part of the message. The sweep arrived while the category was at its most enthusiastic, which is precisely when a regulator saying that nothing has changed is most useful and least welcome. Enforcement that waits for a hype cycle to end arrives after the harm.

What conduct was actually named?

Two distinct categories, and both have close analogues in the placement market.

The first is overstated capability. DoNotPay advertised "the world's first robot lawyer" and, in the Commission's account, "could not deliver on these promises"; it agreed to pay $193,000 and to notify customers about the service's limitations. Ecommerce Empire Builders was alleged to have falsely claimed it could help consumers build an "AI-powered Ecommerce Empire".

The second is manufactured evidence. Rytr was alleged to have provided a service generating an "unlimited number of detailed consumer reviews" that "almost certainly would be false" — a tool whose output was fabricated testimony.

Alongside these sat outright income schemes: Ascend Ecom, which the Commission says defrauded consumers of at least $25 million promising AI-driven "passive income", and FBA Machine, promising "guaranteed income" from "AI-powered" software at a cost to consumers of over $15.9 million.

What unites the five is not the technology but the structure of the promise. Each described an outcome the buyer could not verify in advance and the seller could not substantiate afterwards, with AI supplying the explanation for why the outcome was plausible. Remove the letters and the cases are ordinary false-advertising matters.

The settlements are instructive about remedy as well as liability. DoNotPay's obligation to notify customers about the product's limitations is a corrective disclosure rather than a prohibition, which is the Commission reaching for the same instrument it uses across this series: where a claim misled, the fix is telling the audience what is actually true.

Why does this reach the placement market?

Because "AI-powered" has become standard copy in exactly the services this series has spent sixty reports examining.

AI visibility monitoring, AI citation optimisation, answer-engine placement and guaranteed presence in AI answers are all sold routinely, and as documented throughout this series, the operators of those systems publish eligibility conditions and withhold selection criteria. A vendor promising a specific AI outcome is promising something they cannot verify.

The Commission's standard does not ask whether a claim is fashionable; it asks whether it can be substantiated. A claim that a service will place a company in AI answers is a performance claim, and the party making it bears the burden of supporting it.

The Rytr case has a sharper edge still. A tool producing consumer reviews that "almost certainly would be false" is manufactured evidence, and the same description would fit any service generating testimonials, engagement or citations that did not arise from genuine experience.

There is a particular trap in selling monitoring alongside outcomes. A dashboard reporting how often a brand appears in AI answers is a real product measuring a real thing; a promise to increase that number is a performance claim about a system whose selection criteria nobody publishes. The two are often sold together, and only one of them can be substantiated.

It also cuts against a defence that circulates in this category, which is that nobody could be expected to know how these systems behave. That is accurate and it is not exculpatory. Uncertainty about a mechanism is a reason to avoid asserting outcomes, not a licence to assert them freely because no one can check.

What does substantiation require?

Evidence held before the claim is made, which is the part most easily skipped when a category is moving quickly.

The Commission's long-standing position, applied here, is that objective performance claims need support at the time they are made. A vendor asserting that its work produces AI citations, ranking improvements or measurable visibility gains is making that kind of claim.

This is difficult in precisely the area where it is claimed most. As reported elsewhere in this series, no major operator publishes how sources are selected for citation, assistant outputs vary by user and session, and the one operator publishing citation data warns it indicates nothing about ranking or authority. Substantiating an AI visibility claim against that backdrop is close to impossible.

Which is the useful conclusion for a buyer: the difficulty of substantiating the claim is a reason to doubt the claim, not a reason to excuse the absence of evidence.

It is worth separating a forecast from a claim, because vendors often blur them. Saying that work may improve the chance of being cited, with the reasoning stated, is an argument a buyer can weigh. Saying that it will produce citations is an assertion of fact that requires evidence held in advance, and the second is what turns a sales pitch into a substantiation problem.

How does this fit the rest of the disclosure landscape?

It closes the loop on the theme running through this series.

The Commission's native advertising guidance asks whether readers can tell that content is advertising. Its endorsement guidance places responsibility on the brand rather than the platform. CAN-SPAM asks whether a recipient can tell who is writing and how to stop it. And Operation AI Comply says that none of these obligations pause because a product involves AI.

Read together, they describe one principle applied across formats: a claim must be true and supportable, and the person encountering it must be able to tell what they are looking at.

The technology changes what is easy to produce. It does not change what may be asserted, and the Commission has now said so in terms that require no interpretation.

For a company running placement, outreach and AI-branded services at once, that convergence is a simplification rather than a burden. One standard covers all of it: say what is true, hold the evidence before you say it, and make sure the person reading can tell who is speaking and why.

What this means for anyone buying placement

Ask what evidence supports any AI-related performance claim, and ask before purchase. Substantiation is required at the time the claim is made, so a vendor who cannot produce it now did not have it.

Treat guarantees about AI answers with particular scepticism. No operator publishes selection criteria, outputs are not reproducible, and a promise about them is unverifiable by design.

Refuse anything that manufactures evidence. Generated reviews, testimonials or engagement are the conduct the Commission acted on directly, and the exposure sits with the party whose product is being promoted.

Media Morpheus sells placement and search visibility work, and there is an obvious temptation in this market to describe that work in AI terms. What we can substantiate is narrow and real: an indexed, snippet-eligible page that answers a specific question, published on a named outlet. What no one can substantiate is a promise about what a model will say.

Methodology and limitations

This report quotes the Federal Trade Commission's published announcement of Operation AI Comply, September 2024, for its statements on AI and existing law, the conduct targeted, the named cases and the settlement figure reported for DoNotPay. Five limits apply. First, this is not legal advice; substantiation obligations are fact-specific. Second, allegations in enforcement actions are allegations, and except where a settlement is stated the companies named have not been adjudicated liable; this report reproduces the Commission's characterisations. Third, dollar figures are those stated in the Commission's announcement. Fourth, enforcement has continued since the initial sweep and this report describes the announcement rather than the full subsequent docket. Fifth, this addresses United States federal law only.

Conclusion

The FTC has removed the argument before anyone could make it: "there is no AI exemption from the laws on the books." Operation AI Comply targeted overstated capability claims, tools that manufactured consumer reviews, and income schemes wearing AI branding, with settlements attached. For the market this series has documented, the relevance is direct. AI visibility, AI citation and answer-engine placement are performance claims made about systems whose operators publish no selection criteria and whose outputs cannot be reproduced. The difficulty of substantiating such a claim is not a defence for making it unsupported — it is the reason the claim should not be made.

Frequently Asked Questions

Sources

1. official US Federal Trade Commission. FTC Announces Crackdown on Deceptive AI Claims and Schemes, September 2024.
2. official US Federal Trade Commission. Native Advertising: A Guide for Businesses, cited for the parallel disclosure principle.