Key Takeaways
- Find the source before you react. Almost every wrong answer is repeating something that exists in writing somewhere. Ask the assistant where it got that, then go and look, because the fix lives at the source and nowhere else.
- Check it is actually about you. A confident description of a similarly named business is a different problem from a false claim, and it is fixed by making your own identity unmistakable rather than by chasing a correction.
- Work outward from what you control. Your site first, then the profiles you own, then the pages you can ask someone to change. Time spent arguing with the assistant itself is time not spent on the record it reads.
- Publish the correct version rather than only denying the wrong one. A clear, dated page that states the accurate facts gives an assistant something better to use. Denials without a positive statement leave the original claim as the only thing on the record.
- Expect weeks, not minutes. Nothing you change updates the answer instantly, and a fix that has not shown up after a week is not proof it failed. Re-check on a schedule instead of refreshing in hope.
An AI assistant just told a potential client something about your business that is wrong. Maybe it is out of date, maybe it belongs to somebody else with a similar name, maybe it is simply invented.
The instinct is to argue with the assistant. That does nothing, because the conversation you are having is not where the answer came from.
The answer came from a record. This post is about finding that record, working out which parts of it you can change, and doing those in the order that actually pays.
What do you do when AI says something wrong about you?
Correct the source, not the answer. Find the pages the claim came from, fix the ones you own, ask the owners of the ones you do not, and publish a clear statement of the accurate version somewhere permanent.
That order matters because of how these answers are produced. An assistant is assembling a reply from things it has read, so a correction only sticks if it lands in the material being read. Typing "that is wrong" into the chat window changes that one conversation and nothing else.
There is a second reason to start at the source. Doing so tells you which of the four kinds of wrong you are dealing with, and they need genuinely different responses.
Find where the claim came from first
Ask the assistant directly which pages it used, then go and read them. Tools that answer from live sources show their citations, which turns the whole investigation into a few clicks.
Perplexity's help center describes answers built from current sources with those sources shown, and Google's guidance on AI features in Search describes the same pattern of links accompanying generated answers. Where the links are visible, the source hunt is already done for you.
Where they are not visible, work backwards. Search the exact wrong phrase. Search your business name plus the wrong detail. Nine times out of ten the claim is sitting in an old profile, a stale directory entry, or a page about a business that shares part of your name. The map of where those live is in where ChatGPT gets information about local businesses.
Screenshot everything before you start changing things, with the date visible. You will want the record of what it said, and you cannot recreate it once the answer moves.
The bar lengths are illustrative, standing for how much of each source is yours to change rather than any measured figure. Work top to bottom: the bottom row is the one everybody tries to fix first and the only one that cannot be edited directly, because it refreshes from everything above it.
What you can actually change, in order
Your own site, then the profiles you own, then the pages you can ask somebody to update. Those three tiers cover the great majority of wrong claims, and they get progressively slower, so start at the top.
On your own site, state the correct fact plainly and give it a date. Not a denial. A positive statement of what is true, in a sentence that stands on its own, on a page that is going to stay put. That is what an assistant can use in place of the old claim.
On profiles you own, check every one, including the ones you forgot you created. Stale listings are the single most common origin of an out of date answer, and the full repair routine is in how to fix an out of date picture of your business in AI.
On pages somebody else owns, one polite, specific email works far more often than people expect. Give the URL, the exact sentence, the correction, and a link to the page on your site that supports it. Editors correct facts. They do not respond well to being told to take something down.
| What it looks like | Where it usually comes from | What actually moves it | Realistic pace |
|---|---|---|---|
| Out of date | An old listing, an old page of yours, an old article | Update the source, publish the current version with a date | Days to weeks |
| Mixed up with someone else | A similar business name, a shared city or specialty | Make your own identity unmistakable everywhere | Weeks |
| Invented detail | A gap in the record filled by inference | State the fact plainly where nothing was stated before | Weeks |
| Unflattering but true | A real review, a real complaint, a real report | Respond publicly and add accurate context; do not hide it | Ongoing |
| Potentially defamatory | A false claim of fact somewhere on the open web | Legal advice, dated evidence, and the source owner | Varies; start immediately |
How long a correction takes to show up
Days to weeks where the tool reads live pages, and considerably longer for anything baked into the model itself. Neither timescale rewards refreshing the chat window, which is the main thing to accept early.
The split is worth understanding. Some answers are assembled from pages fetched at the moment you ask, and those move as the pages move. Others draw on what the model absorbed during training, and OpenAI's own explanation of ChatGPT is clear that the model was trained on data up to a point in time. That part does not update because you edited a page this morning.
So make the changes once, write down the date you made them, and re-ask the same question on a schedule. Weekly for the first month is plenty. The baseline discipline is the same one described in how to check if AI recommends you, and without it you cannot tell a slow fix from a failed one.
When it is worse than merely wrong
When the claim is a false statement of fact that could cost you clients or your license, treat it as a legal matter alongside the visibility work. That is a different kind of problem and it deserves proper advice.
Practically, that means three things at once. Preserve dated evidence of exactly what was said. Speak to a lawyer in your jurisdiction about the source of the claim, since the source is usually where any remedy lies. And keep doing the ordinary correction work in parallel, because a legal process does not change what an assistant reads next week.
It is worth saying plainly that most cases are not this. Most are stale, confused, or filled in from silence, and most are fixed by publishing better material than the assistant currently has. That is unglamorous and it is usually enough.
If you would rather see the whole picture in one pass than discover errors by accident, the free check runs 25 real buyer questions from your niche through ChatGPT at $0 and emails a plain-English readout within the hour. It runs the moment you ask for it.
Questions we hear the most
What do I do if AI says something wrong about my business?
Find the source, then correct the source. Ask the assistant which pages it used, go and read them, fix the ones you own, and ask the owner of anything else you cannot edit. The answer follows the record, not the complaint.
Can I get ChatGPT to correct an answer about me?
Not by telling it in a chat. Anything you type in a conversation affects that conversation only. What changes future answers is changing the openly readable pages the model and its live sources draw on.
Why does AI say something about my business that was never true?
Usually it merged you with a similarly named business, or it filled a gap by inference where your own pages said nothing. Both look like invention and both are fixed by stating the facts plainly somewhere readable.
How long before a correction shows up in AI answers?
Days to weeks for tools that read live sources, and potentially much longer for anything the model absorbed during training. Make the change once, record the date, and re-check on a schedule rather than repeatedly.
Should I reply to a bad review if AI is quoting it?
Yes, calmly and once. A public reply that states the facts becomes part of what gets read alongside the review, which is a better outcome than an unanswered claim sitting on its own.
What if the wrong answer is genuinely damaging?
Treat it as a legal question as well as a visibility one, and get advice from a lawyer in your jurisdiction. Keep dated screenshots from the moment you notice it, because evidence of what was said and when is hard to reconstruct later.
How would the free check surface a wrong answer about me?
By quoting what came back. It asks ChatGPT 25 real buyer questions from your niche and reports the answers in plain English, so a false or outdated claim shows up as text you can act on rather than something you happened to notice.
Can I run the free check today?
Yes. It runs the moment you ask for it, and the report is in your inbox within the hour, 7am to 5pm Central, or by 10am the next morning. Asking your own buyer questions in a fresh logged out session will surface most of the same problems if you would rather check by hand.
Your buyers are already asking. Find out what AI tells them.
The free check asks ChatGPT 25 real buyer questions about your niche and sends you the report within the hour, at $0. If it shows you are already getting named everywhere, we will say so in plain words.
Get my free check →