10 Ad Copy Claims AI Writes That You Legally Cannot Publish

TL;DR

Ten claims AI writes into ads, each matched to the rule it breaks - three false however you word them, seven lawful only with evidence in hand, a genuine former price, or the condition next to the claim.

Ask a model for ad copy and it will hand you claims you cannot run. Three of the ten below are false however you word them: the weight-loss claims on the FTC's own list, a testimonial nobody gave, and any claim whose necessary disclosure will not fit the ad. The other seven are restricted rather than banned — lawful only with substantiation in hand, a stated condition, or a disclosure placed next to the claim. An ad is the worst surface for all of them, because it is the one with no room.

Written against US Federal Trade Commission rules and guides, the UK CAP Code, and the Google Ads, LinkedIn and Meta ad rulebooks as retrieved on 2026-09-18. This is not legal advice; it is a list of the rules each claim runs into, with the wording quoted so you can read them yourself.

Key Takeaways

Seven weight-loss claims are treated as false outright, starting with a product that "causes weight loss of two pounds or more a week for a month or more without dieting or exercise" (FTC, Gut Check reference guide, retrieved 2026-09-18).

A model-written testimonial breaks a rule on its face. It is a violation to "write, create, or sell" a testimonial misrepresenting "that the reviewer or testimonialist exists" (16 CFR 465.2, retrieved 2026-09-18).

An earnings claim in an ad carries mandatory disclosures. It must state, "in immediate conjunction with the claim", the dates the earnings were achieved and "the number and percentage" of purchasers who achieved them (16 CFR 437.4(b), retrieved 2026-09-18).

If the disclosure will not fit, the FTC's answer is not to shrink it. Where a disclosure is needed and cannot be made clearly and conspicuously, "then that ad should not be disseminated" (FTC, .com Disclosures, March 2013, retrieved 2026-09-18).

Evidence comes before publication, not after a challenge. UK CAP Code 3.7: "Before distributing or submitting a marketing communication for publication, marketers must hold documentary evidence to prove claims" (ASA, CAP Code section 3, retrieved 2026-09-18).

Why ad copy is the worst place for an invented claim

Three things stack up. The claim is short, so the qualifier that would make it true gets cut for character count. The ad is the public surface a regulator reads first. And a model writing ad copy is optimising for a click, which is the pressure that produces a superlative nobody can prove.

Most of the rules below are not ad-specific. What is ad-specific is that the format removes your room to comply — a product page at least has space for the condition, and those errors are covered in 9 AI product description mistakes.

1. "Clinically proven to treat [condition]" — restricted

Health claims need a particular grade of evidence. The FTC's health guidance states that claims about "the health benefits or safety of foods, dietary supplements, drugs, and other health-related products require substantiation in the form of competent and reliable scientific evidence", generally "in the form of randomized, controlled human clinical testing". It also catches the claim you did not write out: "If elements of an ad imply that the product also provides a disease benefit, the advertiser must be able to substantiate the implied disease claim even if the ad contains no express reference to a disease" (FTC, Health Products Compliance Guidance, retrieved 2026-09-18, SOURCED).

Condition to publish: the trial exists, you hold it, and the wording matches what it measured.

2. The seven weight-loss claims on the FTC's list — prohibited

The FTC publishes the list for media buyers so they can refuse the ad. Do not run copy saying a product: causes two pounds or more of weight loss a week for a month or more without dieting or exercise; causes substantial weight loss no matter what or how much the consumer eats; causes permanent weight loss even after the consumer stops using it; blocks the absorption of fat or calories to enable substantial weight loss; safely enables more than three pounds per week for more than four weeks; causes substantial weight loss for all users; or causes substantial weight loss by wearing it on the body or rubbing it into the skin (FTC, Gut Check reference guide, retrieved 2026-09-18, SOURCED).

Softening does not help. The guide names the workarounds it already sees — "helps consumers lose substantial weight without diet or exercise", "up to three pounds a week for a month or more" — and calls them bogus too (same source, SOURCED).

Status: false as written. No disclosure fixes them.

3. "Earn $10,000 a month with our system" — restricted, and unworkable in an ad

If you are selling a business opportunity, an earnings claim in an ad is governed by the Business Opportunity Rule. Making one "in the general media" requires a reasonable basis, written substantiation in your possession, and a statement "in immediate conjunction with the claim" of "the beginning and ending dates when the represented earnings were achieved" and "the number and percentage of all persons who purchased the business opportunity prior to the ending date … who achieved at least the stated level of earnings" (16 CFR 437.4(b), retrieved 2026-09-18, SOURCED).

Read that against a 30-character headline. The rule does not prohibit the claim; the format does (ANALYSIS).

Condition to publish: the dates and the number-and-percentage sit beside the claim, in the ad.

4. "Guaranteed results, or your money back" — restricted

A refund promise is not evidence. The FTC's small-business guidance answers this directly: "Offering a money-back guarantee is not a substitute for substantiation. Advertisers still must have proof to support their claims." The same page disposes of the other thing a model reaches for — happy customers: "Statements from satisfied customers usually are not sufficient to support a health or safety claim or any other claim that requires objective evaluation" (FTC, Advertising FAQs, retrieved 2026-09-18, SOURCED).

Condition to publish: the underlying result is substantiated, and the refund terms are honoured as stated.

5. "#1 rated" and "better than the alternatives" — restricted

Comparative claims are allowed, and the evidence bar does not drop: "The Commission evaluates comparative advertising in the same manner as it evaluates all other advertising techniques" (16 CFR 14.15, retrieved 2026-09-18, SOURCED). In the UK the timing is explicit — CAP Code 3.7: "Before distributing or submitting a marketing communication for publication, marketers must hold documentary evidence to prove claims" (ASA, CAP Code section 3, retrieved 2026-09-18, SOURCED).

Condition to publish: you can name the ranking, its source and its date, and the ad says which.

6. A testimonial the model wrote — prohibited

This is the clearest line in the list. Under the FTC's consumer reviews rule it is an unfair or deceptive practice for a business to "write, create, or sell a consumer review, consumer testimonial, or celebrity testimonial" that materially misrepresents "that the reviewer or testimonialist exists", that they used or had experience with the product, or what that experience was (16 CFR 465.2, retrieved 2026-09-18, SOURCED). The endorsement guides say the same from the other direction: "Endorsements must reflect the honest opinions, findings, beliefs, or experience of the endorser" (16 CFR 255.1, retrieved 2026-09-18, SOURCED).

A generated quote has no endorser behind it, and asking for "a sample testimonial we will replace later" is how one reaches production (ANALYSIS). Quotes come from a bank of things real customers said.

7. "Was $199, now $99" when nothing sold at $199 — restricted

The former price has to have been real. Under the FTC's pricing guides a comparison is legitimate where "the former price is the actual, bona fide price at which the article was offered to the public on a regular basis for a reasonably substantial period of time", and improper where it "is not bona fide but fictitious" (16 CFR 233.1, retrieved 2026-09-18, SOURCED).

A model does not know your price history. It invents an anchor that makes the discount look better, and the invented number is the violation.

Condition to publish: the former price is one you genuinely charged, regularly, for a substantial period.

8. "Free trial" and "free gift" with the conditions left out — restricted

The word "free" has its own guide, unusually specific about placement. All terms "upon which receipt and retention of the 'Free' item are contingent should be set forth clearly and conspicuously at the outset of the offer", in "close conjunction with the offer" — and a footnote reached by an asterisk "is not regarded as making disclosure at the outset" (16 CFR 251.1(c), retrieved 2026-09-18, SOURCED). That rules out the default pattern: "Free trial*", condition parked on the landing page.

Condition to publish: the obligation — card required, auto-renewal, minimum purchase — sits next to the word "free".

9. "Eco-friendly" and "sustainable", unqualified — restricted

The Green Guides are direct: "because it is highly unlikely that marketers can substantiate all reasonable interpretations of these claims, marketers should not make unqualified general environmental benefit claims" (16 CFR 260.4(b), retrieved 2026-09-18, SOURCED).

In the EU the position hardens four days after this post publishes. Directive (EU) 2024/825 strengthens "consumer protection rules against greenwashing", and the Commission states: "From 27 September 2026, the Directive shall enter into application" (European Commission, sustainable consumption, retrieved 2026-09-18, SOURCED). What it changes for product copy is in 9 AI product description mistakes.

Condition to publish: name the specific attribute and its scope — "recycled aluminium body", not "sustainable".

10. Any claim whose disclosure will not fit — prohibited in that format

The FTC's digital-advertising guidance says to place a disclosure "as close as possible to the triggering claim" and, in a space-constrained ad, to "incorporate the disclosure into the ad whenever possible". Then comes the rule people skip: where a disclosure is needed and "it is not possible to make the disclosure clearly and conspicuously, then that ad should not be disseminated" — and "if a particular platform does not provide an opportunity to make clear and conspicuous disclosures, then that platform should not be used" for ads that require them (FTC, .com Disclosures, March 2013, retrieved 2026-09-18, SOURCED).

So "terms apply" in eight characters is not a fix, and neither is small print the reader meets only after the click. Change the claim or change the format (ANALYSIS).

The second gate: the platform rulebook

Platform policies are contract terms, not law, and they bite first — as a disapproval or an account review:

Rulebook	What it says	Why a model trips it
Google Ads, Misrepresentation	"Making inaccurate claims or claims that entice the user with an improbable result (even if this result is possible) as the likely outcome a user can expect is not allowed"	Copy states the best case as the expected case
LinkedIn Ads Policies	"Any claims in your ad must have factual support"; "unrealistic or misleading claims about health improvements" are prohibited	Same failure modes, enforced pre-click
Meta Advertising Standards	"Ads must not contain content that asserts or implies personal attributes", and no "content implying or attempting to generate negative self-perception in order to promote diet, weight loss or other health related products"	"Struggling with [condition]?" is a standard hook

Sources: Google Ads policies, Misrepresentation, LinkedIn Ads Policies and Meta Advertising Standards, all retrieved 2026-09-18, SOURCED.

What to do before the ad runs

Give the model a fact sheet and tell it to use nothing else — prices, plan names, ranked awards with dates, approved claim wording.
Keep a claim register: one row per claim, with the evidence and its date. CAP 3.7 wants it to exist before publication, and it is the fastest way to answer a platform appeal.
Run the disclosure test. If a claim needs a condition, write the condition into the ad. If it will not fit, the claim changes.
Read the rulebooks for the platforms you are actually buying.
Verify numbers at their source, not in the draft — the routine is in how to check if ChatGPT is right, and what a wrong claim costs once it is public is in eight real consequences.

How This Guide Was Sourced

Written and maintained by the LogicBalls editorial team (logicballs.com). Disclosure: LogicBalls builds AI writing tools, including tools that draft marketing copy. Every limitation here applies to ours as much as anyone's, which is why the fixes above are editorial controls rather than a product.

AI involvement. AI assisted the research and the draft. Every quotation was matched against source text fetched on 2026-09-18, and every rule was read at eCFR, ftc.gov, asa.org.uk or the platform's own policy page rather than in a summary.

Sources. FTC: Health Products Compliance Guidance, the Gut Check reference guide, Advertising FAQs, and .com Disclosures (March 2013). eCFR: 16 CFR 14.15, 233.1, 251.1, 255.1, 260.4, 437.4 and 465.2. ASA: CAP Code section 3. European Commission: sustainable consumption, for the application date of Directive (EU) 2024/825. Platform policy: Google Ads Misrepresentation, LinkedIn Ads Policies, Meta Advertising Standards.

What could not be verified. Meta's transparency site returns HTTP 400 to a plain automated request, so its two quotations were read through a browser-style fetch of the ad-standards page; the two policy sub-pages we tried returned 404, so both quotes are cited to the page that carries them. We did not read Directive (EU) 2024/825 at EUR-Lex, which blocks automated requests — the application date and the greenwashing description come from the Commission's own page. FDA labelling rules for specific product categories were out of scope.

What is not claimed. No count of how often models produce each claim, no ranking by frequency, and no enforcement action against AI-written ad copy specifically. Penalty amounts are not quoted: they are adjusted annually and we did not verify the current figure. Nothing here is legal advice.

No LogicBalls telemetry is used in this guide. Every figure above is external and linked.

Frequently Asked Questions

Is it illegal to use AI to write ad copy?

No. The rules address what the ad claims, not how it was drafted. The problem is that a model produces the unqualified version of a claim by default, and an ad is the format least able to carry the qualifier.

Which of these ten can never be published?

Three. The seven weight-loss claims on the FTC's list are treated as false however they are worded, a testimonial from a person who does not exist is a rule violation on its face, and a claim whose necessary disclosure cannot be made clearly and conspicuously "should not be disseminated" in that format. The other seven are publishable with evidence, a genuine former price, or the condition stated next to the claim.

The platform approved my ad. Does that mean it is compliant?

No. Platform review checks the platform's own contract terms. Those rulebooks are narrower and broader than the law in different places, and an approved ad can still be deceptive under the FTC Act or the CAP Code.

Can I put the small print on the landing page?

Sometimes, and not as a habit. The FTC says to incorporate the disclosure into a space-constrained ad whenever possible, and allows the linked page "under some circumstances". For a "free" offer the guide is stricter: the terms belong in close conjunction with the offer, and an asterisk to a footnote is not disclosure at the outset.

Conclusion

Ten claims: three false in any wording, seven lawful only with something the ad usually has no room for. The pattern is the same throughout — a model writes the strongest version of a claim and the rules ask for the qualified one. Hold the evidence before you publish, put the condition next to the claim, and when it will not fit, change the claim rather than the type size.