One review, on repeat
AI Keeps Quoting a Bad Review About My Business: What Works, and What Is Illegal
Two hundred customers were happy and said so in four words. One was furious and wrote four paragraphs with dates, names and a dollar figure. Guess which one an AI answer finds worth quoting. The fix is not to bury that review. It is to give the machine better material to read.
October 9, 2026 · 8 min read
The short answer
AI engines quote a bad review when it is specific, visible on sources they read, such as Google Maps or Yelp, and not outweighed by newer detailed reviews. Find where the engine read it, reply publicly, fix and document the cause, report it only if it breaks platform policy, and ask every customer for honest reviews. Never buy, filter or suppress reviews: the FTC rule bans it.
On this page
When the summary of your business is somebody's worst day
Owners usually find this problem by accident: a prospect mentions it on a call, or a spouse asks Gemini about the business over dinner. The answer is polite, balanced in tone and devastating in content. Common versions:
- "Some customers report billing surprises," based on a single review from two years ago.
- An AI summary of your Google listing that leads with the one complaint about a missed appointment.
- Perplexity quoting a Yelp review nearly word for word, with a link beside it.
- ChatGPT recommending a competitor and adding that your company "has mixed reviews," while your rating is 4.7.
- A Reddit thread from a disgruntled former employee showing up as a source for questions about your service.
Note what the engine got right and wrong. If it accurately quotes a real review, the work below applies. If it invents a complaint or misstates one, also see AI made up facts about my business.
Which review pile each engine reads
Reviews do not live in one place, and the engines do not share a single feed. Knowing where each one looks tells you which platform to work on first. Several of these relationships come from press reports and third-party studies rather than the engines themselves, and they change.
| Review sources it uses | Where to work | |
|---|---|---|
| [Gemini](/generative-engine-optimization/gemini) | Google Maps data, including ratings, reviews and an AI summary of a place, per Google | Google Business Profile |
| [Google AI Overviews](/generative-engine-optimization/google-ai-overviews) and AI Mode | Local business information from Business Profiles, plus web pages, forums and videos Google indexes | Business Profile, then forums and articles |
| [Perplexity](/generative-engine-optimization/perplexity) | Yelp data, including reviews, licensed since March 2024 according to press reports | Yelp |
| [Siri and Apple Maps](/generative-engine-optimization/apple-intelligence-siri) | Mainly Yelp, Tripadvisor and Foursquare in the US, plus Apple's own ratings, as reported in July 2024 | Yelp and Apple's ratings |
| [ChatGPT](/generative-engine-optimization/chatgpt) | Web search results; a 2024 BrightLocal study found Yelp, Facebook and Google Maps absent from its local directory sources, so quoted complaints often come from forums, articles or business sites | Ask for its sources, then work that page |
Forums matter more than many owners realize. Peec AI, an AI visibility vendor, ranked Reddit the most-cited domain across five AI engines in a March 2026 analysis of 30 million citations, with Yelp also in its top ten. A single angry thread can travel further than a hundred star ratings. Our page on Reddit, forums and directories covers how to take part honestly.
Why one complaint outweighs a hundred compliments
It is the most specific thing anyone wrote
"Great job, thanks" contains nothing to summarize. "Quoted $400, billed $1,150, no call back for nine days" contains a story. Models summarize the text that has content in it.
Nothing newer replaced it
If your last detailed review is eight months old, the old complaint is still the freshest substance on the page. A steady flow of recent reviews changes the mix.
It went unanswered
A complaint with no reply reads as uncontested. A calm, factual reply sits right under it and gives any reader, human or machine, the other half.
It matches a pattern
If three reviews mention slow callbacks, the AI is not being unfair. It is summarizing accurately, and the fix is operational before it is digital.
That last card is worth sitting with. AI answers about reputation are usually a compression of what customers wrote. When the compression stings, read the reviews it came from before deciding the machine is the problem. More on the mechanism in reviews and AI recommendations.
Lines you cannot cross, however tempting
The FTC's rule on consumer reviews and testimonials took effect on October 21, 2024, and it allows civil penalties for knowing violations. Google's review policy is stricter still, and Yelp's is stricter than Google's. A shortcut that quietly removes a bad review can end with a fine, a removed listing or a warning label, and AI engines would then have something genuinely damaging to quote.
Allowed, and worth doing
- Asking every customer for an honest review, on Google, with the same message whatever their experience
- Replying publicly and calmly to every review, good or bad
- Reporting a review that genuinely breaks the platform's rules
- Fixing the problem the review describes, and saying so
- Publishing your own policies, guarantees and process on your site
Banned, by the FTC or the platforms
- Fake reviews, including AI-generated ones and reviews by people who do not exist
- Paying or rewarding reviews conditioned on a positive sentiment; Google bans review incentives of any kind
- Asking only happy customers (review gating), which Google prohibits
- Undisclosed reviews by staff, relatives or other insiders
- Groundless legal threats, intimidation or false accusations to get a negative review removed
- A review site you control presented as independent
Answering the review, then outgrowing it
- 1
Find the exact source
Ask each engine about your business and ask it to cite its sources. Our free AI visibility check shows several engines' answers side by side, though API answers can differ a little from the apps. Note which platform and which review keeps coming up.
- 2
Decide whether it breaks a rule
Google removes reviews that violate policy, such as spam, conflicts of interest or off-topic rants, and says not to report a review just because you disagree with it. If it qualifies, report it in the Reviews Management Tool; evaluation typically takes several days and there is a one-time appeal.
- 3
If it is extortion, do not engage
A sudden run of one-star reviews followed by a demand for money is a scam Google treats as a serious policy violation. Do not pay; collect screenshots and use Google's extortion report form.
- 4
Reply in public, once, with facts
Thank them, state what happened without arguing, say what changed, and offer a direct contact. Keep private details out. The reply is written for the next reader, including the AI.
- 5
Fix the cause and document it
If the review is about callbacks, pricing or cleanup, change the process and publish it: a pricing page with ranges, a written callback promise, a warranty page. Content that answers the complaint in plain sentences gives the engines a second side to quote. See content AI can quote.
- 6
Ask every customer, every time
Send the same review request after every completed job, not just the good ones. Within a few months, recent detailed reviews outnumber the old complaint, which is the only legitimate way to dilute it.
- 7
Flag the answer if it distorts
If the AI exaggerates, saying "many customers" about one review, use thumbs down or Report a problem in Google's AI features, Gemini, Claude and ChatGPT, and describe the error.
How long before the answer softens
Slower than you want, faster than you fear. A reply usually shows up on the platform quickly. A removal request takes Google several days to evaluate, longer with an appeal. Changing the overall picture takes months of steady new reviews, and AI summaries move only as the sources under them change. Track the same handful of questions monthly; the goal is the complaint shrinking from the headline to a footnote, then disappearing as newer material crowds it out. No one can promise a date, and anyone selling a guaranteed removal is selling something the platforms forbid.
You cannot argue an AI out of quoting a review. You can give it fifty newer ones to read.
Keeping it from happening again
Treat reviews as a weekly routine, not a crisis. Reply within days, keep requests going after every job, and review complaints for patterns each month. In Athena CMS, the Echo agent drafts replies to Google, Yelp and Facebook reviews and flags unhappy ones first, so nothing sits unanswered. Owners who want the whole job handled can fold review generation and replies into our AI and GEO search programs, priced on the GEO pricing page after a free GEO audit.
Get help when the review involves a legal dispute (talk to your attorney, not your marketer), when one complaint is costing visible business, or when several platforms disagree badly about you. Related problems sit in the AI search fixes section, including ChatGPT recommending a competitor instead of you.
Questions owners ask us about AI search
Only if it breaks the platform's rules. Google removes reviews that violate its policies, such as spam or conflicts of interest, and says not to report a review just because you disagree with it. Report qualifying reviews through the Reviews Management Tool. For genuine reviews, the legitimate path is a public reply, a fix, and more recent reviews.
Under the FTC rule, a business may offer incentives for reviews only if there is no requirement, stated or implied, that the review be positive. Google is stricter: its policy bans any incentive for reviews and bans asking only happy customers. Yelp tells businesses not to ask for reviews at all. Offering a discount puts your Google listing at risk.
ChatGPT draws on web search results rather than one review platform. A November 2024 BrightLocal study of local searches found Yelp, Facebook and Google Maps did not appear among its directory sources, so complaints it repeats may come from forums, articles or other sites. Ask ChatGPT to cite its sources for the answer and work on those pages.
Reply. An unanswered complaint reads as uncontested to customers and to AI engines summarizing the page. Keep it short and factual: acknowledge, state what changed, offer a direct contact. Never reveal private customer details or argue. The reply is for future readers, not the reviewer.
That is a question for an attorney, not a marketing firm. Be aware that the FTC rule bans groundless legal threats, intimidation and certain false public accusations used to prevent or remove negative reviews. A real defamation claim is a legal matter; a threat to scare a customer into deleting a review is prohibited.
There is no published number, and burying is the wrong goal. AI summaries reflect what customers wrote, so the aim is for recent, detailed reviews to represent your business as it is now. Ask every customer, reply to every review, and recheck the AI answers monthly to see whether the complaint still leads.
Keep reading
- Stars, read by machinesHow Reviews Affect AI Recommendations, Engine by EngineWhich review platforms ChatGPT, Gemini, Perplexity, Copilot and Siri draw on, what the FTC fake review rule bans, and how to build reviews that hold up.
- Third-party sourcesReddit, Forums and Directories as AI Sources: What the Engines Read About YouWhy Reddit threads, forums and directories feed AI answers, what the citation studies really found, and why faking a thread backfires.
- When the answer is fictionAI Made Up Facts About My Business: How to Trace, Report and Replace the FictionAn AI invented a service, a price, an owner or a review for your business. Save the evidence, find the cause, report it to each engine and fill the gap.
- Engine guide: PerplexityHow to Get Cited by Perplexity: Its Crawlers, Its Sources and the Pages It QuotesPerplexity shows every source it reads. Learn how PerplexityBot works, why Yelp matters for local answers, and how to become the page it cites.
- Schema, without the mythsStructured Data for AI Search: What Schema Does, What It Doesn't, and What to KeepDoes schema help you show up in AI search? What Google and Microsoft say, the end of FAQ rich results, and how to use Organization, LocalBusiness and Product.
- When ChatGPT picks themChatGPT Recommends My Competitor Instead of Me: Find What It Read and Change the AnswerChatGPT names the shop across town instead of you. See which sources it cited, why they favor the rival, and the steps that change the answer.
- Everything in AI Search FixesSee the section
See which review the engines repeat
Tell us which review keeps surfacing. We will check where each engine reads it and send a plan that stays inside FTC and platform rules.
Which of these sounds like you? (pick any)
Rather talk now? Call (210) 346-0848.