AI Made Up Facts About My Business: How to Trace, Report and Replace the Fiction

The service you never offered, at a price you never charged

This is the error that makes owners angriest, and with reason. A wrong phone number is clerical. An invented fact is a claim about your business that you never made, delivered in the calm voice of an encyclopedia. Customers repeat it on the phone. Some arrive expecting a service you do not offer, at a price you would lose money on.

• Services: "offers 24-hour emergency repairs" when you close at six, or a whole service line you dropped or never had.
• Prices: a starting price, a discount or a free estimate policy that appears nowhere you have published.
• People and history: the wrong owner, a founding year off by a decade, a "family business" story that belongs to someone else.
• Reviews and ratings: a star average that matches no platform, or a paraphrased complaint no customer ever wrote.
• Credentials: an award, certification or license you do not hold, which can be worse than an insult in a licensed trade.

Start by confirming the problem is real and repeatable. Ask the same question again, ask it in a second engine, and run it through our free AI visibility check to see what several engines say side by side. A fabrication that appears once and never again matters less than one every engine repeats.

Why a confident machine fills in blanks

The vendors say it plainly. Google says Gemini can hallucinate and present inaccurate information as fact, and that AI Overviews may include mistakes. Anthropic says Claude can give incorrect or misleading responses and should not be treated as a single source of truth. These systems write the most likely next sentence. When the record about you is thin, the most likely sentence is borrowed from businesses like yours.

The distinction that matters most: is the claim invented or copied? If a cited page says it, the engine copied an error, and you fix that page. If no source says it, the engine invented it to fill silence, and you fix the silence. Wrong hours and phone numbers are nearly always copied; that problem has its own walk-through.

Document, trace, correct, report

1. 1

   Capture the evidence

   Screenshot the full answer and copy the question word for word. Note the engine, the date, whether web search was on, and save the share link if the engine offers one. If the claim ever becomes a legal matter, this record is what counts.

2. 2

   Open every citation

   Read each linked source and search it for the false claim. If a page says it, you have a copied error: contact that site's owner with the correction and proof. For a Google knowledge panel, Google says descriptions cannot be edited directly; you have to ask the original source to correct it.

3. 3

   If nothing is cited, find the silence

   Ask yourself what a customer would need to know to contradict the answer, and check whether your site says it plainly. Usually it does not: no prices, a vague services page, no about page, no ownership history.

4. 4

   Publish the facts, including the negatives

   One clear page or section per topic: services you offer and services you do not, how pricing works, who owns the business and since when, license numbers, service area. A sentence like "We do not offer emergency service after 6 p.m." gives an engine something to quote instead of something to guess. See content AI can quote.

5. 5

   Make the facts readable by machines

   Put the text in the page HTML rather than behind scripts or tabs, and keep any structured data matching what visitors see. Update the description in your Google Business Profile, Apple Business and Bing Places to the same wording.

6. 6

   Report the answer in each engine

   Use the feedback route in the table below for every engine that repeats the claim. Keep a note of when you reported it.

7. 7

   Retest on a schedule

   Ask the same question in the same engines every few weeks and log the result. When the fabrication stops appearing, keep testing monthly; invented facts can come back when a page changes.

Where to report a made-up answer, engine by engine

Every major engine has a way to flag a bad answer, and none promises to act on a single report. Treat reporting as one step, not the fix. The routes below come from each company's own help pages, apart from ChatGPT, where the route is simply the button under every answer.

When an invented fact turns into a legal problem

Most fabrications are a nuisance. A few are damaging: an engine claiming a business is under investigation, has lost a license or was sued. In 2025 a Minnesota solar company, Wolf River Electric, sued Google for defamation, alleging an AI Overview falsely said the state attorney general was suing it and that customers canceled contracts as a result, according to press reports. Google denies the claims, and the case does not settle what the law requires of AI companies.

An AI engine fills silence with the average of your industry, and the average is rarely flattering.

If a false statement is costing you work, save the evidence described above, report the answer through every route available, and speak to a lawyer about your options. DataCram is not a law firm and cannot advise on defamation; what we can do is help correct the public record the engines read, which is useful whatever the lawyers decide.

Don't fight fiction with more fiction

The temptation is to flood the web with counter-claims. Some of the ways people try that break rules that carry real penalties, and none of them work as well as a plain page of facts.

Reviews matter more here than most owners realize, because engines summarize them. Our guide to reviews and AI recommendations explains how, and if an engine is quoting a real negative review rather than an invented one, see AI quotes a bad review.

How long the fiction lasts, and when we can help

Copied errors fade once the source page is corrected and recrawled, on each engine's own schedule. Invented ones fade when the engine finds a clear fact to use instead, which can take weeks for engines that search the web and much longer for answers drawn from a model's memory. No engine publishes a timeline, and nobody can promise a date.

On sites built with Athena CMS, the Delphi agent asks ChatGPT, Google's AI answers and the others what they say about the business each week, flags wrong prices, hours or claims, and drafts the facts and FAQs the engines need. For any site, AI visibility monitoring uses AICiteBeam to track what each engine says and alert you when it changes.

Call when the same invention keeps returning after you have published the facts, when it appears across several engines, or when it touches licenses or legal claims. A free GEO audit traces each false claim to its cause and comes with a fixed monthly quote; our AI recommendation optimization service does the work, with GEO pricing explained up front. For related problems, see AI search problems or the AI search fixes hub.