Find out where the wrong detail comes from, then fix that source. If ChatGPT found it on a web page, correct or answer that page and publish the right facts plainly on your own site. If no page carries it, the error sits in the model's memory, and the slower route is more pages, on your site and others, that state the facts.

Report the answer too. But neither Google nor OpenAI says a report will change what the next buyer sees.

Why AI tools get firms wrong

The clearest test I have found comes from Ahrefs. Mateusz Makosiewicz built a fake luxury paperweight company called Xarumei and asked eight AI tools 56 questions built on false claims about it.

He then published an official FAQ on the brand's site that denied those claims. At the same time he planted three fake stories elsewhere: a blog post, a Reddit post and a Medium "investigation". The stories contradicted each other and the FAQ.

After the stories went up, Gemini and Perplexity repeated false details in about 37-39% of their answers. ChatGPT-4 and ChatGPT-5 stayed under 7% and cited the official FAQ in 84% of their answers. Five of the tools trusted the Medium piece over the FAQ. One of them was Google's AI Mode.

The most useful finding for a firm that exists is about detail. The FAQ said "We don't publish unit counts or revenue." The fake sources gave exact figures. The tools that went wrong used the figures.

The test has limits. It used one invented brand, and the models were versions from late 2025. Ahrefs graded the answers itself and sells brand-monitoring tools.

Find out where the error comes from

Start by asking each tool your buyers use what it knows about your firm. Ahrefs suggests this exact check: "What do you know about [your brand]?" Then ask the questions a buyer would ask, such as who you work with and what you charge.

Ask more than once. In a small test by Orbit Media, Claude kept only 30% of the pages it had cited when the same questions ran two days later. One run shows you one answer, not the pattern.

When an answer is wrong, open the sources it cites. If one of them carries the wrong detail, you have found the problem.

If none of them does, the claim came from the model's memory. A University of Toronto study found that 46% of ChatGPT's mentions of a brand that the question did not name had no cited page naming that brand. A tool can describe your firm without reading anything about it. We cover that gap in getting cited by AI search isn't the same as being recommended.

Fix the pages you control

Your own site is the one source you can change today. Ahrefs' advice from the experiment is an FAQ with direct lines, such as "We have never been acquired", plus dates and numbers wherever you can give them.

For a services firm, that means stating plainly when you were founded, where you are based, who leads the firm, which sectors you serve and what you don't do. In Ahrefs' test, a vague line left a gap that a more specific page elsewhere filled.

Check your Google Business Profile as well. Google's guide to its AI features names business profiles as one way to get a firm's services into its AI responses. Those features draw on Google's search index.

Fix or answer the pages you don't control

If a directory or an old article carries the wrong detail, ask the owner to correct it. Send them the right information and a source for it.

If the page won't change, publish a specific answer on your own site. In Ahrefs' test, ChatGPT resisted the fake stories by citing the official FAQ.

If an AI tool repeats something false and damaging about your firm, take legal advice before anything else.

Report it, but don't rely on it

In Google's AI Overviews, click the thumbs-down under the overview, choose "Report a problem", pick a category and add details. Google says the report includes your search and its results, and that feedback "helps us improve AI Overviews". It does not promise to change a specific answer.

In ChatGPT, you can rate a reply with a thumbs-down. I found no OpenAI process for correcting facts about a company. If the report lets you add a comment, give the right fact and a link to the page that proves it.

A few common follow-up questions

Can I ask OpenAI to correct what ChatGPT says about my firm?

I found no published process for that. A thumbs-down flags one reply. It is not an edit to what ChatGPT knows.

How long does a fix take to show up?

No AI provider publishes a timescale. A corrected page can only change answers once the tool reads it again, and an error held in the model's memory can outlast every page you fix.

Does Claude get firms wrong too?

In Ahrefs' test, Claude never repeated the false stories. It also never used the brand's own site, and kept saying the brand did not exist. The test used Claude Sonnet 4.5, and Anthropic has released newer models since.