Skip to content
ReputationKiln

A reference on online reputation  ·  Calm, sourced, free


An AI is saying something wrong about your business

By ReputationKiln Editorial · Published

More and more customers ask an AI before they ask you. Is this place any good, are they still open, what do they charge. And the AI is wrong a lot. One large study found that close to half of AI answers contained errors. It will tell someone you are closed when you are busy, list a service you stopped offering years ago, confuse you with a competitor who has worse reviews, or invent your prices. In a handful of cases it has gone further and stated something flatly defamatory: one small solar firm sued after an AI answer wrongly claimed it was being sued by the state, and put a lost contract worth a hundred and fifty thousand down to it.

You cannot reach in and edit the model. But you do not have to, because the AI is only reflecting what is written about you across the web. Change those sources, and you change what it says. The defence is the same one the whole of this site keeps arriving at: make yourself consistently, verifiably real, so the machine has no room to guess.

Find out what it is saying

Ask the main tools, ChatGPT, Gemini, Perplexity, Copilot, the questions your customers actually ask about you, and write down every error. You cannot fix what you have not checked, and the answers differ between tools.

Fix the sources, not just your own site

This is the part most people get wrong. Updating your own website helps, but the AI weighs the whole web, so the faster fix is correcting the information out there, on the third-party sources it leans on.

  • Give it a machine-readable source of truth. Add clear, structured business information to your site, your correct name, address, founding date, services, in the format search engines read, and keep dedicated, accurate pages for the basics.
  • Claim and correct your Bing listing. It directly influences what some AI tools say about you, and it is often the one nobody has touched.
  • Report the error through each tool's own feedback option.
  • Get the correct facts onto authoritative outside sources. Consistency across several trusted, recent places is what actually moves it. In documented cases, a correction propagated within a couple of months once a credible outside source carried the right information.
  • Expect it to take weeks, not minutes, and to be an ongoing habit rather than a one-off job.

The deeper point, and the limit

This is the whole thesis of the site in one task. In a world where customers ask a machine, the business that wins is the one that is consistent and well-evidenced everywhere, because that is the one the AI describes correctly. A thin or inconsistent presence gets hallucinated, and a faked one is doubly fragile. The limit, said plainly: the law here is new and unsettled, and no honest helper can "guarantee" an AI fix. Which is exactly why "AI is lying about you, pay us monthly to fix it" is the newest flavour of the panic tax. The core steps above are largely free and yours to do.

Sources

  1. AI search and chat tools frequently return confident but incorrect answers and rarely decline to answer when they cannot; in one study a leading tool gave incorrect responses 153 times yet admitted uncertainty only seven. — Columbia Journalism Review, Tow Center, AI search citation studies (2025). https://www.cjr.org/tow_center/we-compared-eight-ai-search-engines-theyre-all-bad-at-citing-news.php · checked 2026-06-04