AI Testimonials and the FTC Fake Reviews Rule: Who Gets Charged

On 22 December 2025 the Federal Trade Commission published two announcements about fake reviews.

The first set aside a 2024 order against Rytr, an AI writing service the FTC had accused of letting subscribers "generate false and deceptive online reviews". That order had banned Rytr from offering "any AI-enabled service generating consumer or customer reviews or testimonials." The Commission now said the facts in its own complaint "fail to support allegations that Rytr violated Section 5 of the FTC Act", and that the order "unduly burdens innovation in the nascent AI industry".

The second announced warning letters to ten companies under the Consumer Reviews and Testimonials Rule, which since October 2024 has let the FTC seek civil penalties against knowing violators.

Read only the first, and it sounds as if the FTC has lost interest in AI-written reviews. The months since point the other way. In April 2026 the FTC charged the company behind the TruHeight supplement brand under the rule, and in May the Justice Department filed a civil penalty suit on the FTC's behalf against a home-repair business. Neither case, as the FTC announced it, is about AI-written reviews. Both are about businesses publishing reviews from people who didn't exist or never bought anything.

The rule doesn't ban using AI. It bans a review or testimonial that materially misrepresents whether the person exists, whether they used what you sell, or what happened to them — and when the FTC announced the rule, it named "AI-generated fake reviews" as an example.

We read the rule on the government's eCFR site, the FTC's announcements and business guidance, and the court filings named below, on 10 September 2026, and re-read the rule sections quoted here on 14 September. This isn't legal advice, and we aren't lawyers. If you've received a letter from the FTC, that's a job for a lawyer who handles FTC matters.

A one-person business is covered by definition

The rule is 16 CFR Part 465, published in the Federal Register in August 2024 and in force since 21 October 2024. Its definitions leave no room for "too small to count": "Business means an individual who sells products or services", and "Officers include owners". We found no small-business exemption.

The parts that matter if you sell a product or a service on your own:

• Writing or creating a fake review or testimonial (§ 465.2(a)). A business may not "write, create, or sell" a review or testimonial that materially misrepresents that the reviewer exists, that they used the product or service, or what their experience was. Unlike the paragraphs around it, this one has no "knew or should have known" wording for a violation, though civil penalties still require a knowing one.
• Buying fake reviews, or publishing testimonials you should have known were fake (§ 465.2(b)). This paragraph does carry the "knew or should have known" standard, for reviews you buy and testimonials you publish that someone else wrote.
• Paying for positive reviews (§ 465.4). Incentives "conditioned expressly or by implication on" a particular sentiment are banned. Asked whether a business can pay for 5-star reviews as long as the reviewers disclose the incentive, the FTC's rule Q&A answers: "No. That conduct would violate Section 465.4."
• Reviews by yourself and your family (§ 465.5). An owner writing a review of their own business without "a clear and conspicuous disclosure" of the relationship is a violation. Asking a spouse, parent, child, sibling or employee for a review is a violation too, if they then post one without disclosing the relationship and you didn't tell them to disclose it, encouraged them not to, or knew or should have known it went up and took no remedial steps. A general request sent to all your customers is excluded.
• Suppressing reviews (§ 465.7). No "unfounded or groundless legal threat", intimidation, or public accusation made knowing it's false or "with reckless disregard" for whether it is, "in an attempt to" stop a review being written or get it removed. And if your own site displays reviews, you can't hide the negative ones because of their rating while implying you show them all.
• Buying followers (§ 465.8). Buying fake "indicators of social media influence" is banned for anyone who "knew or should have known" they were fake, where they "materially misrepresent" your influence "for a commercial purpose".

The penalty is up to $53,088 per violation. That figure comes from the FTC's 2025 inflation adjustment, and as of September 2026 it hadn't risen: the Office of Management and Budget told agencies in April there would be no 2026 adjustment, because the shutdown left no October 2025 price data to calculate it from.

Two limits on that number. The FTC's Q&A says the rule "authorizes courts to impose civil penalties for knowing violations", so a court sets the amount, and only for knowing violations. And the rule "does not provide a private right of action": customers and competitors can't sue you under it. The FTC can, and states bring claims of their own. Illinois joined the May case as a co-plaintiff.

Where AI fits into that

The FTC made the connection itself. Its announcement of the rule described it as covering reviews and testimonials "by someone who does not exist, such as AI-generated fake reviews".

There are three ways a small business realistically uses AI near its testimonials. Here they are, from clearly over the line to allowed with conditions.

1. AI invents the customer or the quote. A made-up name, a generated face, words nobody said. That misrepresents that the person exists or had the experience, which is § 465.2(a). There's no grey area, and the paragraph covering writing it has no knowledge requirement for a violation (civil penalties still need a knowing one).

2. AI tidies up a real customer's words. This is the common case, and no FTC document we found addresses it; the rule Q&A has no question about editing. The closest tests are in two texts, and applying them to AI editing is our reading, not the FTC's.

• The rule covers a testimonial that misrepresents the person's "experience".
• The FTC's Endorsement Guides set out how the FTC reads deception in endorsements. They allow paraphrase, with a condition. You "need not present an endorser's message in the exact words of the endorser unless the advertisement represents that it is presenting the endorser's exact words, such as through the use of quotation marks". But the endorsement "may not be presented out of context or reworded so as to distort in any way the endorser's opinion or experience".

With AI, the failure is specific: models add detail. The FTC's case page for Rytr says the complaint alleged generated reviews containing "specific, often material details that had no relation to the user's input". That order is gone. The problem it described is still how an edited testimonial becomes a false one — a "saved me ten hours a week" the customer never said.

3. An AI avatar presents your product. The Q&A: "The rule has no blanket prohibition on the use of AI-generated avatars in marketing." It adds that an avatar's use "might be considered a 'testimonial'", which the rule prohibits "only if the underlying testimonials were fake or false", and that avatars "could also be deceptive in violation of the FTC Act." Our reading: an avatar explaining what a product does is a presenter, and an avatar saying it uses the product every day is a testimonial from someone who doesn't exist.

If you advertise in New York, there's a separate duty. Senate Bill S8420A, signed as Chapter 617 on 11 December 2025, requires advertisements to disclose a "synthetic performer". That's a generated asset meant to look like a human performer who isn't any identifiable real person. The duty falls on a business that "produces or creates" an ad for what it sells, where that business has "actual knowledge" (one of the two versions of the text on the Senate's page adds its agents and employees; we couldn't tell which version was enacted). Ads for films, TV and other expressive works are exempt, and one version also exempts audio ads. Penalties are $1,000 for a first violation and $5,000 for later ones, and it took effect 180 days after signing, in June 2026.

What happened to the AI case, and what it didn't decide

Rytr sold an AI writing assistant with a review-writing feature. According to the FTC's case page, the complaint said the reviews it produced "almost certainly would be false for the users who copied them and published them online", and the 2024 order banned the service outright.

In December 2025 the Commission reopened that order and set it aside, by a 2-0 vote, with Rytr's consent. Its announcement gives two reasons: the complaint "failed to satisfy the legal requirements of the FTC Act", and the order "unduly burdens artificial intelligence (AI) innovation".

What the announcement doesn't do matters as much:

• It doesn't mention the reviews rule. Rytr was charged under Section 5 of the FTC Act, in a complaint brought before the rule took effect.
• It concerned the company that made the tool, which the complaint had held responsible for what its subscribers published. It says nothing about the businesses that publish fake reviews.
• It doesn't say AI-written fake reviews are lawful.

We read the FTC's announcement and its case page for Rytr, but not the set-aside order itself. Law-firm summaries of the order's reasoning go further than the announcement does, so we've left them out.

Has anyone been charged under the rule? Yes, but not for AI-written reviews

"Nobody has been charged under the fake reviews rule" was arguably true until spring 2026. It isn't now. Here's what we found.

• Ten warning letters, December 2025. The FTC said the letters "are not formal determinations that the recipients have violated" the rule, and that violations "can result in the filing of a federal lawsuit…". The FTC's announcement didn't name the recipients.
• TruHeight, April and July 2026. An administrative complaint charged the company behind the brand under the FTC Act and the rule. The FTC said its website carried "several thousand five-star reviews" written by company employees. It also said the company offered consumers free product or discounts for 5-star reviews, and used fake social media profiles "run by automated bots" to post "software-generated comments". The announcement doesn't say which of that conduct was charged under the rule rather than the FTC Act, and we didn't read the complaint. The final order, approved in July, bars misrepresenting reviewers and buying reviews conditioned on sentiment, and requires a $750,000 payment. The proposed order had set a $4 million judgment, partially suspended.
• Premium Home Service, May 2026. The Justice Department, acting for the FTC, and Illinois sued the company behind it in the Northern District of Illinois: United States v. B.E.S.T. GDR, LLC, No. 1:26-cv-05415. The complaint says the business ran invented local contractors whose ratings were boosted by "fabricating profiles of people to pose as customers". Its fourth count charges fake reviews under § 465.2(a), (b) or (c) and asks for civil penalties. These are allegations; nothing has been decided.

We found no rule case about AI-generated reviews. We searched the FTC's site, read the case pages for these matters, and checked law-firm enforcement roundups published between May and July 2026. The FTC's case database wouldn't give us a filtered search, and we didn't page through every 2026 press release. So read that as "none we could find", not "none exists".

That record invites two readings, and both are wrong.

The first is that the rule only catches operations with thousands of fake reviews. That describes the cases brought so far, not the rule, and the penalty is set per violation.

The second is that AI is safe because no AI-review case has been brought. The FTC named AI-generated reviews when it announced the rule. And the one AI case it dropped was never brought under the rule.

The argument that could narrow the May case

On 17 July 2026 the Premium Home Service defendants asked the court to dismiss, among other things, the fake-review count. Their argument runs in three steps:

• the reviews the complaint quotes were posted in or around January 2024;
• the rule didn't take effect until 21 October 2024;
• and the rule "contains no express retroactivity provision".

The complaint alleges more than that. It says that even after an FTC investigative demand in February 2024, the defendants "continue to write, create, procure, or purchase fake positive consumer reviews". The motion answers that the complaint alleges this "Without any details".

The government filed its response on 7 August and the defendants replied on 21 August. The parties agreed to pause discovery until the judge rules, and as of 10 September 2026 there's no ruling.

We haven't read the government's response, so we can't say who has the better argument, and nobody should plan around the outcome. The questions are whether reviews posted before the rule existed, including ones still on display after it took effect, can support a violation, and whether the complaint pleads enough later conduct. The motion also attacks the case's other counts. If your site carries reviews or testimonials you can't trace to a real customer, the court's answer won't change what to do about them: take them down.

What to do if AI is anywhere near your reviews

These steps come from the rule, the FTC's Q&A and the Endorsement Guides. Where a step is only good sense rather than something those texts say, we say so.

• Use testimonials only from real customers, and keep the original. Save the email, form or message it came from, with its date. No rule text requires this, but it's how you'd show the person exists and said it.
• Let AI tidy, never add substance. This one is our reading of the texts above, not an FTC instruction. No numbers, results, timeframes or details the customer didn't give. Quotation marks tell readers these are the customer's exact words, so if the wording has changed at all, even for grammar, drop them and don't change what the customer meant (§ 255.1(b)). Asking the customer to approve the final wording isn't required anywhere we read, but it's the simplest way to know you haven't distorted it.
• Don't present an AI avatar or a generated face as a customer. The rule doesn't ban an AI presenter, but the FTC's Q&A warns avatars "could also be deceptive in violation of the FTC Act", and in New York an ad has to disclose a synthetic performer. Giving an avatar a customer's experience crosses the rule's line.
• Ask every customer for a review, not only the happy ones. The Q&A says the rule doesn't specifically ban asking only satisfied customers, "But this practice could violate the FTC Act."
• Incentives are allowed if they don't depend on what the review says. The rule bans conditioning them on a sentiment, "whether positive or negative". A discount for leaving a review of any kind is allowed. The Q&A warns that "failing to disclose incentives could be a violation of the FTC Act". A discount for a 5-star review isn't allowed, disclosed or not.
• Family reviews need the relationship stated. If you ask a relative, tell them to say who they are. Don't review your own business without saying it's yours.
• Don't threaten reviewers. A groundless legal threat or intimidation to get a review removed is banned. If your site shows reviews, don't filter out the low ratings.
• Don't buy followers, likes or views.

What we checked on our own site

A site that writes about disclosure rules should pass them, so we checked ours before publishing this.

Our affiliate labels were wrong in three ways. RunSolo earns commission from two programs, Make and Tidio.

1. We labelled links that don't pay us. All 49 product cards in our articles said "Affiliate link — we may earn a commission at no extra cost to you", although 31 of them link to tools that pay us nothing. All 25 entries on our tools page carried the same kind of label, and only 2 of those pay us.
2. The wording was the one the FTC questions. The FTC's Endorsement Guides FAQ says readers "might not understand that 'affiliate link' means that the person placing the link is getting paid", while "'Paid link' right next to an affiliate link should be an adequate disclosure".
3. One paid link had no label and our promise pointed to the wrong place. A Make link inside our tutorial text had no label at all. And our About page promised disclosure "in the article footer": the bottom of the page, which the same FAQ says readers may miss.

What we changed on 10 September 2026. A link that pays us now carries "Paid link — we earn a commission if you sign up through it, at no extra cost to you" directly beneath it, and links that don't pay us carry no label. That applies to product cards, comparison tables, verdict boxes and the tools page. The Make link in the tutorial now carries its own label in the sentence. The line at the end of each article pointing to our policy stays, but as a supplement, not as the disclosure.

Labelling every link was over-inclusive, and in small type with the wording the FAQ questions, it may not have told readers which links actually paid us.

On 14 September 2026 we also rewrote the line at the end of each article. It used to say "Some links may be affiliate links", which is vague in exactly the way the guidance is about. It now says either that links marked "Paid link" earn us a commission, or that no link in the article pays us.

Testimonials. RunSolo displays no testimonials, no ratings of itself and no reader counts. It did once publish something worse: an article on AI customer support built around an invented small-business owner, quoted four times, on a page recommending a tool we earn from. We deleted it in August 2026, and the rewritten article says so at the top.

The rest of the fine print

This is one of a set of articles on the terms and rules underneath the AI tools a one-person business uses:

• who owns what AI tools produce;
• whether you can put client data into ChatGPT;
• whose job consent is when a notetaker joins a call;
• what Canva's license allows for logos and client work;
• what payment processors' contracts let them do with your money;
• AI accessibility widgets after the FTC's accessiBe order.

If you work in a licensed trade, advertising rules add another layer, which we mapped in AI real estate advertising rules.

---

Last reviewed 14 September 2026. Not legal advice; we are not lawyers. Read at source on 10 September 2026, with the rule sections and New York bill text re-read on 14 September: 16 CFR Part 465 and Part 255 on eCFR (current to 10 September 2026); the FTC's announcements of the rule (August 2024), the warning letters and the Rytr set-aside (both 22 December 2025), and the TruHeight (April and July 2026) and Premium Home Service (May 2026) actions; the FTC's Rytr case page; the FTC's rule Q&A (November 2024) and its undated Endorsement Guides FAQ; OMB Memorandum M-26-11 (17 April 2026); New York Senate Bill S8420A; and, for Premium Home Service, the complaint, the defendants' motion to dismiss and the docket on CourtListener. Not read: the Rytr set-aside order itself, the government's response to the motion, and the TruHeight complaint. No link in this article pays us; see how we make money.