AI marketing mistakes the FTC has already brought cases over

Short answer: The most useful list of AI marketing mistakes is the one the Federal Trade Commission has already written in its own cases. They point to five mistakes: buying an “AI-powered” income machine, trusting a refund guarantee, claiming your AI does a professional’s job, quoting an accuracy number nobody tested, and letting a tool write your reviews. In these cases, most of the money was lost by people buying AI programs, not by people fined for using AI.

This list is built from what the FTC alleged, case by case. The twist: the person who gets hurt is more often the buyer than the advertiser. We say what to do instead, and what changed at the FTC in 2025 and what did not.

This is general information, not legal advice.

Why read FTC cases instead of another tips list?

On September 25, 2024, the FTC announced five actions in one day under a sweep it called Operation AI Comply. Then-chair Lina M. Khan said the actions “make clear that there is no AI exemption from the laws on the books.”

In four of the cases here (Ascend Ecom, Ecommerce Empire Builders, FBA Machine and Air AI) the people who paid were buyers of an AI program. The FTC says many Air AI buyers were small business owners and entrepreneurs.

A complaint is an allegation, not a finding. For the basics, see what AI marketing is.

Mistake 1: buying the “AI-powered” income machine

Three of the 2024 cases follow one pattern.

Ascend Ecom. The FTC alleged a business opportunity scheme that claimed its AI-powered tools would help people earn passive income by opening online storefronts, and that it defrauded consumers of at least US$25 million. A federal court in the Central District of California temporarily halted it and put it under a receiver.

Ecommerce Empire Builders. The FTC alleged the company claimed to help people build an “AI-powered Ecommerce Empire”, with training programs at almost US$2,000 and stores costing as much as US$35,000, and that the promised profits did not materialize. A court in the Eastern District of Pennsylvania temporarily halted it and appointed a receiver.

FBA Machine, earlier called Passive Scaling. The FTC alleged it promised guaranteed income through online storefronts using “AI-powered” tools, rebranded after buyers sought refunds, and cost consumers more than US$15.9 million. A court in the District of New Jersey temporarily halted it and appointed a receiver.

What to do instead: treat “AI-powered” next to “passive income” or “guaranteed” as a reason to slow down, and ask for the earnings evidence in writing.

Mistake 2: trusting a refund guarantee on an AI program

The Air AI complaint was filed on August 25, 2025, in federal court in Arizona, by a 3-0 Commission vote. All of it is allegation.

The products were business coaching and support, the “Air AI Access Card”, and licenses to resell the services, with “conversational AI” advertised as a replacement for human customer service representatives. The FTC alleges buyers were told they would earn back tens of thousands of dollars within days or months, with a full refund if they did not earn a set amount, typically two or three times what they paid.

The FTC alleges the defendants “rarely honor their guarantee” and that buyers lost “as much as $250,000” and were often left in debt.

It names two rules. The Telemarketing Sales Rule covers misrepresenting performance, refund policies and earnings potential. The Business Opportunity Rule requires sellers to give buyers a disclosure document and an earnings claim statement, which the complaint says were not provided. That gives you a practical test: ask for those two documents.

Our view: reselling licenses were part of this offer, and a business that resells a white-label AI tool and repeats the vendor’s earnings claims is publishing those claims itself.

What to do instead: a guarantee is only as good as the seller’s record of paying it. Ask who has been refunded, and read the refund terms in the contract.

Mistake 3: saying your AI does a professional’s job

DoNotPay marketed itself as “the world’s first robot lawyer”. The FTC alleged the company did not test whether its chatbot’s output matched a human lawyer’s, and did not hire or retain attorneys. The proposed settlement required a US$193,000 payment, on a 5-0 Commission vote.

The alleged failure was not using AI. It was claiming equivalence to a professional without testing it.

For a small business, the same claim sounds like these illustrations, not quotes from any case: “our AI receptionist replaces your front desk”, “an AI bookkeeper”, “AI does your SEO”. If you publish a line like that, show it was checked against how that person works. If you cannot, say what the tool does do: it answers calls after hours, or drafts a first pass.

An AI-written claim is still your claim. Our AI marketing guide for US small business covers that rule.

Mistake 4: quoting an accuracy number nobody tested

Workado promoted its AI Content Detector as “98 percent” accurate at telling AI-written text from human-written text. The FTC said independent testing found 53 percent accuracy on general-purpose content, and alleged the model had only been trained or fine-tuned on academic content. The FTC’s Bureau of Consumer Protection director, Chris Mufarrige, said the product “did no better than a coin toss.”

The order bars effectiveness claims unless they are not misleading and the company holds competent and reliable evidence when the claim is made, and it must keep that evidence. The Commission approved it on August 28, 2025, 3-0.

For a small business, think “95% accurate lead scoring” or “answers 9 in 10 questions correctly” on your own site, or a vendor’s number repeated on your page. Those are illustrations, not quotes from any case.

Ask three things. Tested on what? Tested by whom? Does the test look like how your customers use the tool? If you cannot name the test, drop the number. “Answers common questions about hours and booking” is a claim you can check this afternoon.

Mistake 5: letting a tool write your reviews

A tool will write a review if you ask. The Consumer Reviews and Testimonials Rule is about the business that posts it. The rule went into effect on October 21, 2024.

It prohibits creating, selling, buying or disseminating fake or false reviews and testimonials, including reviews that misrepresent they are by someone who does not exist, “such as AI-generated fake reviews”. It also covers company-controlled “independent” review sites, suppressing negative reviews by threats, and fake followers or views. Courts can impose civil penalties for knowing violations.

Two details matter. Per the FTC’s Q&A on the rule, incentives for reviews are not prohibited when there is no express or implied requirement about sentiment. The problem is compensation conditioned on a particular sentiment. And reviews from insiders, such as officers, managers and others with a material connection, need a clear and conspicuous disclosure of that connection.

Now Rytr. In 2024 the FTC went after the tool maker, whose “Testimonial & Review” use generated detailed reviews, often with details that had nothing to do with what the user typed in. The proposed consent order passed 3-2. In December 2025 the FTC set that order aside.

The rule is separate from that order, and the December release does not mention it. Our reading: the reversal is about whether a tool maker is liable, not about whether a business may post invented reviews.

What to do instead: ask real customers, do not tell them what to say, and use AI only to help you reply.

Did the 2025 changes at the FTC make any of this safe?

No. The Rytr reversal is real. On December 22, 2025, the Commission voted 2-0 to reopen and set aside the order. It said the complaint “failed to satisfy the legal requirements of the FTC Act” and that the order “unduly burdens” AI innovation. The release title cites the administration’s AI Action Plan.

But the Workado final order passed 3-0 in August 2025, and the Air AI complaint came the same month, also 3-0. The December release itself says the FTC “will continue to hold accountable actors that use AI to violate the law”.

Our conclusion: what eased is pressure on tool makers. Deceptive claims and schemes are still being pursued.

A hypothetical: a Tulsa roofing company gets the pitch

This is a hypothetical: not a real client, a fictional seller, no measured results. A roofing company owner in Tulsa, Oklahoma, gets a video call pitching an “AI sales agent” program. The fee is an upfront US$18,000 (invented for this example). The seller says it pays for itself in 60 days, offers a money-back guarantee, and says the owner could resell it to other contractors.

Here is how the five mistakes play out.

Mistake 1. “Pays for itself in 60 days” is an earnings claim. The owner asks for the evidence in writing. A reasonable seller can send it the same day.

Mistake 2. The owner asks for the refund terms in the contract and the names of buyers who have actually been refunded. Because resale is part of the offer, the owner also asks for the disclosure document and the earnings claim statement, or why they are not required.

Mistake 3. If the agent “replaces a sales rep,” the owner asks how that was tested against a human rep’s work.

Mistake 4. If it “books 9 in 10 qualified appointments”, the owner asks tested on what, by whom, and whether it resembles storm-season roofing calls.

Mistake 5. If the sales page shows testimonials, the owner asks who wrote them.

A seller who cannot answer has given the owner the answer. The owner also skips resale, since repeating the seller’s claims makes them the owner’s.

Instead, the owner tries a small, month-to-month text-back tool for missed calls. Our notes on AI chatbots for lead capture cover that first step, and our AI automation service covers having it built.

The test to run before you sign or publish

Before you buy:

1. Can the seller show earnings evidence in writing?
2. If it is sold as a business opportunity, will they hand over the disclosure document and the earnings claim statement?
3. Are the refund terms in the contract, and who has actually been refunded?

Before you publish:

4. Can I prove this claim today?
5. Was this number tested on customers like mine?
6. Did a real customer write this review, with nobody telling them what to say?

For ad claims and endorsements, see digital advertising for small business. For adoption figures, see our AI marketing statistics.

Where this article stops

• Complaints are allegations, and we have not tracked how each court case ended.
• We give no penalty figures, because the FTC pages we read give none.
• State attorneys general and state laws are outside this article.
• Bot disclosure, AI voice calls under the TCPA, HIPAA and copyright are covered in our main AI marketing guide.
• None of this is legal advice.

Frequently asked questions

What did the FTC allege against Air AI?

In a complaint filed in August 2025 in federal court in Arizona, the FTC alleged that Air AI sold business coaching, an access card and reseller licenses with claims that buyers would earn back tens of thousands of dollars, backed by refund guarantees. The FTC alleges the defendants rarely honor the guarantee and that buyers lost as much as US$250,000. These are allegations.

Is it illegal to call a product or service “AI-powered”?

Our reading of the cases is that no rule we read bans the phrase itself. The cases turn on whether the claim is true and can be backed up. Workado was ordered to hold competent and reliable evidence for its accuracy claims, and the FTC alleged DoNotPay never tested its “robot lawyer” claim. If you can prove what you say, the label is not the problem.

Did the FTC drop its case against Rytr?

Not exactly. The FTC announced a proposed consent order against Rytr in September 2024, then on December 22, 2025 voted 2-0 to reopen and set aside the final order. It said the complaint failed to satisfy the legal requirements of the FTC Act. The same release said the FTC will continue to hold accountable actors that use AI to violate the law.

Can a US business offer customers something in return for a review?

The FTC’s Q&A on its reviews rule says incentives for reviews are not prohibited when there is no express or implied requirement about what the review says. The problem is compensation or incentives conditioned on a particular sentiment, such as a discount only for five stars. Reviews from owners, managers or other insiders need a clear and conspicuous disclosure of that connection.

What should I ask for before buying an AI business program?

Ask for the earnings evidence in writing. If it is sold as a business opportunity, ask for the disclosure document and the earnings claim statement, which the FTC’s Air AI complaint says the Business Opportunity Rule requires. Then read the refund terms in the contract and ask who has actually been refunded. A seller with real answers will give them without delay.

Have the FTC’s AI cases been proven in court?

The court cases began as complaints, which are allegations. In 2024 a court temporarily halted three of the storefront schemes and put them under a receiver. DoNotPay agreed to a proposed settlement, and Workado’s consent order was made final in August 2025. We have not tracked how each case finally ended, so treat every description in this article as what the FTC alleged, not as a finding.

---

Being pitched an AI program? Get a free AI marketing audit first.