AI-Written Reviews and Review Responses: The Rules
Why this matters
A chat assistant can write a convincing five-star review in four seconds, in any voice, about a job that never happened. Since October 21, 2024 that is a federal violation with civil penalties attached, and the FTC said in plain terms that AI-generated reviews are inside it. The same tool drafting your reply to a real customer's one-star review is fine, and often a good idea. The line between the two is not "AI or no AI". It is whose experience the words claim to describe. A small shop needs to know exactly where that line sits, because the shop is usually one owner deciding alone, late at night, after a bad review.
The rule itself
The FTC's Trade Regulation Rule on the Use of Consumer Reviews and Testimonials, 16 CFR Part 465, took effect October 21, 2024. It is federal, it reaches businesses of every size, and a knowing violation exposes the business to civil penalties per violation, at a figure the FTC indexes each year. It does not replace state consumer-protection law, and several states' attorneys general pursue fake reviews under their own statutes as well.
Its core prohibitions, in plain terms:
- Fake reviews and testimonials (465.2). Writing, creating or selling a review that misrepresents that the reviewer exists, that they actually used your service, or what their experience was. Buying or posting one when you knew or should have known it was fake is covered too. A review generated by AI about a customer who does not exist, or an experience nobody had, is a fake review.
- Reviews bought for a sentiment (465.4). Offering anything of value conditioned, said or implied, on the review being positive (or on a competitor's being negative).
- Insider reviews without disclosure (465.5). Reviews by your officers, managers, employees or their relatives, posted as if from an ordinary customer, where the relationship is not clearly disclosed and you knew or should have known.
- Company-controlled review sites (465.6). A "reviews" website you run, presented as independent.
- Review suppression (465.7). Unfounded legal threats, intimidation or false public accusations used to get a review removed, and presenting the reviews you show as all or most of them when you have hidden the bad ones.
- Fake social indicators (465.8). Bought followers or views used to misrepresent your influence.
What the rule does not touch: your responses. A reply you post as the business, in the business's name, is your own speech. The general ban on deceptive practices in Section 5 of the FTC Act still applies to it, so a reply cannot misstate facts, but nothing in Part 465 stops you from drafting it with a tool.
Where the line sits
The test that sorts nearly every case: does the text claim to be a customer's experience? If yes, a customer must have had it and written it. If the text is the business speaking as the business, the tool is a typing aid.
| Task | Status | Why |
|---|---|---|
| Drafting your reply to a real review | Allowed | The business speaking as itself |
| Generating "sample" testimonials for your website | Banned | Testimonials from people who do not exist |
| Turning one real customer's emailed praise into a polished testimonial | Allowed only with their approval of the final wording | The polished text must still be their experience, and they must agree it is |
| Writing a review for a customer to post ("here's one you can just paste") | Avoid | The words are yours; if they misstate the experience, both of you are exposed, and most platforms forbid it outright |
| A discount for leaving any honest review | Not banned by Part 465 | The incentive is not tied to sentiment; but many platforms ban incentivised reviews entirely, and an incentivised review should disclose it |
| A discount for a five-star review | Banned | Conditioned on sentiment |
| Your service manager reviewing the shop | Banned without clear disclosure | Insider review |
| Threatening to sue over an honest bad review | Banned if unfounded | Suppression; see also the non-disparagement point below |
Review gating, asking only customers you expect to be happy, is the one practice the final rule did not ban by name. The FTC's revised Endorsement Guides (16 CFR Part 255, 2023) still treat selective solicitation that distorts what customers think as potentially deceptive, and the major review platforms prohibit it in their own terms, which can get your listing's reviews removed. See related: Automate Review Requests or Ask in Person: Decision Tree.
Non-disparagement clauses. The Consumer Review Fairness Act (15 U.S.C. 45b) makes void a form-contract clause that restricts a customer's honest review or penalizes them for one. If your service agreement template has one, take it out; it is unenforceable and it is evidence of the attitude the FTC looks for.
Why this matters more to a small shop than a big one
A national brand with ten thousand reviews can absorb a few planted ones without the pattern showing. A shop with 70 real reviews cannot: ten fake ones added to them are an eighth of the 80 total, they usually arrive in a burst, and they share the cadence of whatever tool wrote them. Platforms' detection looks for exactly that, and the usual consequence is not a fine but a filter: the platform removes the fakes, often removes real reviews posted in the same window, and may label the listing. The shop ends up with fewer reviews than it started with and a warning badge, which costs more calls than a few bad reviews ever did. The federal penalty is the tail risk; the platform filter is the likely one.
Worked example: one owner, one week, five decisions
A nine-person pest control shop picks up two one-star reviews in a week. The owner has a chat assistant open and five ideas.
1. "Write me three five-star reviews I can post from my sister's and my nephew's accounts." Banned twice over: fake (465.2, no such experiences) and insider (465.5, family). Not done.
2. "Write ten testimonials for the new website's homepage, realistic, varied names." Banned: fabricated testimonials. Instead the owner searched the shop's inbox and survey responses for real praise, found six, and emailed each customer asking permission to quote them by first name and town. Four agreed. The site launched with four real testimonials instead of ten invented ones.
3. "Offer a free re-treatment for anyone who leaves us five stars." Banned: conditioned on sentiment. The owner considered a small gift for any honest review, then dropped it after checking the main review platform's terms, which forbid incentives of any kind.
4. "Draft replies to the two one-star reviews." Allowed. This is where the tool earns its keep, and where the next section's check matters.
5. "Find out who left the anonymous one and tell them we'll take legal action if it stays up." The review said the tech "didn't treat the garage". The job record showed the garage was excluded at the customer's request on the work order. That makes the review mistaken, not defamatory, and threatening suit over it would be the suppression the rule names. The reply (decision 4) states the record politely instead. This is not legal advice; a review that genuinely makes false factual claims is a question for an attorney before anyone threatens anything.
Count: of five ideas, three were banned (1, 2 and 3), one was allowed (4), and one became the reply instead of a threat (5). The two allowed paths both cost more time than the banned ones, which is the honest reason the banned ones are tempting.
Checking an AI-drafted reply before it posts
The rule allows the reply; the tool's habits are what make it risky. Read every draft for four things before posting:
Invented events. Asked to reply to "the tech was rude and left a mess", the owner's first draft said "we're sorry our technician arrived late on Tuesday and didn't use shoe covers". The review mentioned neither lateness nor Tuesday nor shoe covers. The model added the details a reply like that usually contains. An apology for something that did not happen is a false statement and a gift to anyone collecting a pattern of complaints.
Customer details. Drafts reach for specifics to sound sincere: the address, the treatment, the product used, the customer's name. A public reply should confirm nothing about the customer that they did not post themselves. "We see your service in our records" is enough.
Promises. "We'll refund your visit in full" or "we guarantee this won't happen again" may be more than you meant to offer, and it is now in writing in public.
Admissions. "Our technician made a mistake" when you have not established that. Say what you will do ("I'd like to look into this with you directly") rather than conceding facts you have not checked.
The reply that posted on the garage review: "Thanks for the feedback. Our work order for this visit shows the garage was excluded from the treatment area at the time of booking. If that isn't what you wanted, please call me directly and we'll sort out the right plan." It is about one sentence of fact from the record, no customer details, no promise, and an open door.
How to know you are on the right side of it
Three questions, before anything posts: Did a real customer have this experience and write or approve these words? Is anyone being offered anything for a particular rating? Does the shop's reply state anything the job record does not support? A yes to the second, or a no to the first or third, stops it. Keep the job record you checked the reply against; if a reply is ever challenged, it is the only thing that shows the reply was true when posted.
References
- 16 CFR Part 465, FTC Trade Regulation Rule on the Use of Consumer Reviews and Testimonials, effective October 21, 2024
- 16 CFR Part 255, FTC Guides Concerning the Use of Endorsements and Testimonials in Advertising, revised 2023
- 15 U.S.C. 45b, Consumer Review Fairness Act
- See related: The Verify-Before-You-Send Rule for AI Output
- See related: Telling Customers When AI Is Involved