What do I do if ChatGPT says something wrong about my business?
Published · By Mario Russo
Fix the source, not the answer. As of November 2026, we found no published process at OpenAI for correcting facts about a business, so there's no form that rewrites what ChatGPT says. What works is finding where the wrong fact came from — usually your own site, a stale directory, or an old listing — correcting it there, sending feedback on the answer itself, and re-checking over the following weeks.
Why does AI get my business wrong?
Usually because something on the web really does say the wrong thing. An old directory still lists your previous hours. A profile carries a former address. Two businesses share a similar name, and the assistant blends them together.
Sometimes nothing on the web is wrong and the model is simply mistaken. OpenAI's own help documentation warns that ChatGPT can produce incorrect output and suggests checking the sources behind an answer. That's why the first job is working out which case you're in.
How do I find where the wrong information came from?
Save the exact prompt and answer, then ask the same question a few more times, because a one-off error is different from a pattern. If the assistant shows sources, open each one. Perplexity and ChatGPT's search mode often link them, and those links are your shortlist.
If no source is shown, search for the wrong fact yourself: the old phone number, the old address, the outdated hours. Then check your own site, your Google Business Profile, Bing Places, Yelp, Facebook, and the directories in your trade.
What should I fix first?
Start with what you control. Correct your website in plain visible text, then your own profiles, then any listing you can claim. Use one exact version of your name, address, phone, hours, and services everywhere.
If the confusion comes from a similar business name, make the difference explicit on your site: your full legal or trading name, your city, and your owner's name. Then give feedback on the answer itself using the thumbs-down control if the assistant offers one. Treat that as a helpful extra, not the fix.
What if the answer is harmful, not just outdated?
Save screenshots with dates first. If an answer falsely claims something damaging, such as a legal problem or a fraud accusation, use the platform's formal reporting route. Google's legal reporting tool, for example, lists AI Overviews as a product you can report. For anything serious, talk to a lawyer, because we can't advise you on what your legal options are.
How long until the wrong answer goes away?
There's no guaranteed date. Assistants that read the live web can change within days of re-reading the corrected page, while listings can take weeks and anything stored in a model's training data can take much longer. Repeat your saved prompts weekly and record the results. Our free AI-Readiness Check starts that record for you: we capture what assistants say today and show which sources need fixing.
Sources
Keep reading
Find out what AI says about your business.
Run my free AI-readiness check