What Should I Check When AI Assistants Invent Reviews for My Local Client?
When AI assistants appear to invent reviews for a local client, first check whether the review is really invented. Many turn out to be something else: a real review of a business with a similar name or of another location, an old or removed review, or a paraphrased summary of several reviews presented as a single quote. Truly fabricated quotes, reviewer names and star ratings do happen, most often when a client has few reviews or is missing from major review platforms, leaving assistants to fill the gap. Search every platform, check the cited sources and look for name collisions before deciding. Then document the error, report it, fix the data gaps that invite guessing, and explain to the client what can and cannot be done.
Why does it matter?
People rely on AI summaries of reviews. BrightLocal's 2026 research found half of consumers trust AI to summarise reviews, and 23% rely on the summary alone without checking review profiles. An invented complaint, or an invented five-star quote the business cannot back up, reaches buyers who never look further.
How do you classify the error?
| Type | How to recognise it | What to do |
|---|---|---|
| Fabricated quote or reviewer | No match on any platform, in archives, or in the client's records | Document and report; fix data gaps |
| Misattributed review | The quote exists, but for another business with a similar name or another branch | Disambiguate the client's identity; report |
| Stale or removed review | The quote existed but was edited, removed or belongs to a closed period | Check removal status; report; publish current information |
| Paraphrased summary shown as a quote | The words blend several real reviews | Usually acceptable if accurate; report if the framing is misleading |
| Wrong rating or count | "4.1 stars from 30 reviews" when the real figure differs | Check every platform's rating; correct listings; report |
| Review on a platform where the client has no profile | "Reviewers on [platform] say..." with no profile there | Report; consider claiming or creating the profile |
What should you check, step by step?
- Capture the answer in full: the prompt, assistant, date, location and any cited links. How to preserve AI answers as evidence is covered in how to document a false statement generated by an AI assistant.
- Search for the quote in quotation marks on search engines, and on each review platform the client uses.
- Search for businesses with similar names in the same area and category, and check their reviews for the quote.
- Check other branches or former locations of the client.
- Check the client's records for the reviewer name or situation described, while respecting privacy obligations.
- Check web archives for removed or edited reviews.
- Compare ratings and counts across platforms against what the answer says.
- Open the cited sources. If none are shown, the answer probably came from model memory or pure generation; rerun it with search switched on or ask the assistant for its sources.
Run the same question several times and on other assistants. A fabricated review that appears once is a different problem from one that appears in most answers.
What data gaps make invention more likely?
- Few reviews overall, so there is little real material to summarise.
- Missing profiles on the platforms assistants rely on for the client's category.
- Inconsistent names, addresses or phone numbers, which make it easy to blend the client with other businesses. Name mix-ups are covered in why ChatGPT recommends the wrong business when several share one address.
- No reviews shown on the client's own website, or reviews shown only in a script-loaded widget.
- Generic category and service descriptions, which let assistants fill in details from similar businesses.
How do you fix it?
- Report the answer through the assistant's feedback option, stating that the review does not exist or belongs to another business, with evidence.
- Build real review volume on the platforms that matter for the category: ask every customer for an honest review in their own words, on platforms that allow requests. Yelp does not allow businesses to ask for reviews, so there the work is a complete, claimed profile.
- Complete and align profiles: name, address, phone, categories and website identical everywhere.
- Show real reviews on the website in text, with the platform and date, rather than only in a widget.
- Disambiguate from similarly named businesses with a clear location, full business name and distinctive details.
- Monitor the same questions monthly until the invented review stops appearing.
Where the invented review is damaging, such as a fabricated safety or fraud claim, treat it as a reputational incident. Escalate within the agency, involve the client, and consider legal advice.
What extra care do regulated clients need?
Healthcare, legal and financial clients need more caution when checking and responding. When searching records for the reviewer or situation described, stay within what the client's privacy obligations allow, and let the client do the checking where patient or client files are involved. Never confirm in public that someone was a patient or client. In the US, the Department of Health and Human Services has penalised dental practices for disclosing patient information in replies to online reviews. Keep public statements general, such as "We cannot find this review and have reported it", and keep the evidence file internal. If the invented review alleges misconduct, involve the client's compliance or legal adviser before anything is published.
How should you explain it to the client?
Keep the explanation short and factual:
- What happened: "ChatGPT showed a review quote that we cannot find on any platform."
- What type it is: fabricated, misattributed, stale or a summary.
- What cannot be done: no one can edit an AI answer directly.
- What will be done: reporting, data fixes and monitoring, with dates.
- What the client can do: keep asking every customer for an honest review, and tell the agency about any new reviews or disputes.
How to report AI sentiment and issues to clients more generally is covered in how to report AI citation sentiment to clients.
What might an illustrative case look like?
Consider an illustrative agency whose client, a family dental practice, was described by an assistant as having "complaints about surprise billing, according to several Yelp reviewers". The practice had no Yelp profile. The agency searched the quote and found it came from reviews of a dental group with a similar name in a neighbouring town. The client's name, phone and address were inconsistent across four directories, and its own site showed reviews only in a script widget. The agency aligned the listings, created and claimed the Yelp profile, added a text reviews section with dates to the website and reported the answer with links to both businesses. A month later, the billing complaint no longer appeared in answers about the practice.
What do people ask about invented reviews in AI answers?
Can the client sue the AI company over invented reviews?
Possibly, in some cases, but it is rarely the fastest route. Take legal advice for serious, damaging claims, and fix the sources meanwhile.
Should we post a response to the invented review?
There is nowhere to respond, since the review does not exist. Publish accurate information on the client's own site and profiles instead.
Does this happen more with some assistants?
It varies by assistant and question. Test the same questions across several assistants, and prioritise the ones the client's customers use.
How long until an invented review stops appearing?
There is no fixed timeline. Answers that search the web can change once the corrected pages are recrawled; answers from model memory can take longer. Keep monitoring monthly.
Published by The Kovmere Team · updated Sep 29, 2026 · More resources