How do I fix what ChatGPT and Google get wrong about my company?

Last updated October 14, 2026. Written by the Third Coast AI team.

Fix a wrong AI answer about your company by finding the page the engine cited for the wrong fact and correcting it there, because the engines with retrieval on are repeating what they read; then make your own site and every profile state the correct fact in the same words, use the feedback routes OpenAI and Google publish as a secondary measure, and re-run the fact-check prompt weekly until the correction shows. Corrections that reach a cited page usually appear in retrieval-on answers within days to a few weeks of the page being re-crawled; answers that come from training data with retrieval off change only when the model is updated, and the only lever there is to make the correct fact the most-repeated version on the web.

This page covers how to find the errors, where they come from, what to change and in what order, the published feedback routes and what they can and cannot do, and how long to expect each route to take. It is the method behind the fact-check log in our audit, and it can be run without us.

How do you find out what the engines get wrong?

Write a set of fact-check prompts about your company and run them in each engine with retrieval on, at least twice each. The set should cover the facts a buyer might act on: where you are, what you sell and do not sell, who you serve, your prices if public, who leads the company, how long you have existed, your certifications, your service area, and anything that changed in the last two years such as a move, a merger, a rebrand or a discontinued line. Phrase some as direct questions ("where is [company] based") and some as the buyer would ("is [company] a good fit for a [thing] project in [place]"), because the second kind surfaces errors in the engine's description rather than in a single fact. For every error, record the engine, the prompt, the wrong claim, and the source the engine cited for it, since that citation is where the fix has to start. The audit checklist includes this step among the others.

Where do the wrong answers come from?

The first row is the most common and the most fixable. Ahrefs' study of 75,000 brands found third-party mentions correlate with AI citations at 0.664 against 0.218 for backlinks, which is also why third-party errors travel so far: the engines trust what others say about you more than what you say about yourself, so a wrong address in a directory can outrank a correct one on your contact page. The mentions versus backlinks page covers the mechanism.

What should you change, and in what order?

1. Correct the cited source. Claim the listing if it is a directory or profile, submit the publisher's correction route if it is press, and ask for an update if it is a partner or association page. Keep a record of every request and its date.
2. State the correct fact on your own site, plainly. Put it in the first paragraph of the relevant page in a complete sentence a machine can quote ("Third Coast AI is based in Holland, Michigan"), in the raw HTML rather than rendered by JavaScript, and in Organization schema.
3. Reconcile every profile. Google Business Profile, Bing Places, LinkedIn, industry directories, the chamber, Crunchbase and whichever others your category uses should carry the same legal name, description, address and founding date. The entity guide walks through this.
4. Create or correct the Wikidata item and claim the knowledge panel. Google's knowledge panel update page says you can suggest edits from the panel and that feedback from verified users is reviewed within a few days, which is the fastest correction route Google publishes.
5. Use the engines' feedback routes as a secondary measure, described below.
6. Re-run the fact-check prompt weekly in each engine and log when the correction appears. Keep the prompt in the monthly set afterward so regressions show.

Which feedback routes do OpenAI and Google publish?

OpenAI's help article on reporting a violation describes the in-product route: thumbs down under the ChatGPT message, then Select an issue, then the category that fits, and it links a content reporting form at openai.com/form/report-content. The same article says reported domains and other content may be reviewed by OpenAI's Model Quality team, which may apply filters or other mitigations to help prevent ChatGPT from relying on unreliable sources in future responses, which is the closest thing to a published correction mechanism and is aimed at unreliable sources rather than individual facts. OpenAI's ChatGPT search help page adds the practical advice that if a result looks wrong you can ask ChatGPT to search again using a specific source, date or location, which fixes the answer for that user and not for the next one. For the retrieval side, OpenAI's crawler documentation says OAI-SearchBot must be allowed for a site to appear in ChatGPT search answers and that a robots.txt change takes about 24 hours to register, so if your corrected page is not being cited, check access before anything else.

Google's AI Overviews help page describes thumbs up and thumbs down at the bottom of each overview, with a Report a problem option that asks you to choose a category and add details, and the same route from the three-dot menu at the top of the overview; it applies to AI Mode as well. Google's documentation on AI features says there are no additional requirements to appear in AI Overviews or AI Mode beyond the normal Search guidelines, which also means there is no separate correction channel: the fix is the page and the record, and the feedback form is a signal rather than a ticket. For facts in the knowledge panel, Google's knowledge panel update page is the route with a stated review window of a few days for verified users. We use all of these and expect little from the feedback forms on their own; they are worth the two minutes, and the source correction is what moves the answer.

How long do corrections take?

• Retrieval-on answers (ChatGPT with search, Gemini with grounding, AI Overviews, AI Mode, Perplexity, Copilot) usually change within days to a few weeks after the corrected source is re-crawled. The timing depends on how often the engine's index refreshes the cited page, and a corrected page on your own site that you request indexing for in Search Console and Bing Webmaster Tools is usually picked up fastest.
• Knowledge panel edits are reviewed within a few days for verified users, sometimes longer, according to Google's knowledge panel update page.
• Third-party corrections take as long as the publisher takes, which ranges from same-day for a claimed directory to never for an abandoned site, and when a source will not correct, the work shifts to making other sources outnumber it.
• Training-data answers with retrieval off change at the next model update, which has no published schedule, and only if the corrected fact has become the dominant version online by then.
• Robots.txt and CDN changes take about 24 hours to register with OpenAI's systems according to OpenAI's crawler documentation, and Google and Bing pick them up on their normal crawl schedules.

Report the correction at the prompt level: the wrong claim, the source, the fix, the date, and the first run in which the correct answer appeared in each engine. That log is the most convincing section of an early monthly report, because it shows the engines responding to specific work, and it is the part of the first 90 days that produces visible results soonest.

Frequently asked questions