How to correct wrong information AI says about you

Correction and removal are different jobs

Removal asks a page to stop existing. Correction asks the answer to change — which means an assistant has to find something better than what it found last time.

If you only delete the stale page, the assistant falls back to the next-best source, which is often older and worse. Correction works by giving the crawlers a clearly better page, then getting the wrong one fixed.

Fix it where it was scraped

Trace the claim to the exact URL that carries it. One page is usually the origin and the rest are copies; fixing the origin is what eventually moves the copies.

Write the request so it can be acted on without a conversation: the URL, the exact sentence as published, what is wrong, the correct version, and one piece of evidence a stranger can verify. Ask for a correction on the page, not a note in a reply.

Keep the original wording. When you later want to prove an answer changed, you need the before text, and the moment the page is edited it is gone.

Publish something better than the wrong page

Assistants cite what they can read and resolve. A single canonical page that states your current role, organization, and links — plainly, in text, at a stable URL — is the cheapest correction available, because it gives every crawler a source that agrees with itself.

OpenProfiles publishes that as a machine-readable record: your corrections sit next to the source claim they answer, so a reader and an agent can both see what was said and what you dispute. Handle ownership stays labeled unconfirmed unless the address is verified — an unverifiable claim marked verified would be the same failure we are trying to fix.

Confirm it, then confirm it again

A sent email is not a correction. Re-run the same audit after the source updates and check whether the claim still appears and what it now cites.

Expect assistants to change one at a time. The one whose crawler visits most often moves first, and an old answer from a slower index is not evidence that the fix failed.

Can I just tell ChatGPT it is wrong?

Only for that conversation. Corrections in a chat do not propagate to other users or other sessions. The durable fix is the source page it cites.

The wrong claim is about someone with my name. What do I do?

Do not ask for a deletion — ask for disambiguation. A page that clearly distinguishes you, at a stable URL, gives the assistant the evidence it currently lacks.

How do I know the correction worked?

Re-run the identical audit and compare citations. The claim should either be gone or now cite the corrected page. Anything else means the source you fixed was not the one being read.

What ChatGPT actually knows about you

Assistants answer from two different places: frozen training data and a live web search run at question time. Here is how to tell which one is talking, and how to test it without fooling yourself.

How to remove your information from AI answers

You cannot edit a model. You edit the page it read, then the index that cached it. Here is the order the layers actually come off in, and how long each one takes.

A repeatable AI visibility check

A ten-minute audit you can run monthly on a person or a brand, and the four pieces of evidence worth writing down each time.