How to correct a wrong fact an AI states about you

There is no edit button on an AI answer. You fix a wrong fact by finding where it came from, correcting it at that source, publishing the correct version clearly on your own site, and then waiting to be re-read. The first step is the one people skip, and it decides how long the rest takes.

What is actually happening underneath

A wrong fact reaches the answer by one of four routes, and each has a different fix.

An old page of yours. Old pricing, a discontinued product, a previous company name, a page you forgot about. This is the most common cause and the easiest to fix.

An out of date third party page. A directory profile from three years ago, an article written at launch, a listing with your old office address. You do not control the page, but you can usually ask.

Confusion with a similar name. If another company shares part of your name, or your name is also a common word, the assistant may be merging two entities. This shows up as facts that are true about somebody else.

The model filling a gap. When there is very little written about you, an answer may contain something plausible that was never stated anywhere. This is the hardest case, because there is no source to correct. The fix is to make the correct fact easy to find so there is no gap left.

How you tell them apart: ask the question again and look at whether the answer carries citations. If it cites a page, you have found your source. If it cites nothing, the fact came from memory and will be slower to shift.

The five step correction, in order

Doing these out of order is what makes corrections take a quarter instead of a month.

One, write down the wrong fact exactly as stated. Copy the sentence. Note the app, the date, the question you asked and whether there were citations. Without this you cannot tell later whether it improved.

Two, find the source. Open every citation. Search the web for the wrong value itself, in quotes, which often finds the page faster than reading the citation list does. Search your own site for it too, including old posts and any PDFs.

Three, correct your own pages first. This is fully in your control and it is where the wrong value usually lives. Correct in place and show the date of the change.

Four, ask the third parties. One short, factual message per page, with the correct value and a link to the page on your site that states it. No complaint, no legal tone. This works more often than people expect.

Five, make the correct fact easy to find. One page that states it plainly, with the date it became true, linked from somewhere visible. If the cause was a gap rather than a source, this step is the entire fix.

A worked example

On 6 August 2026 we found two price lists in public for the same product. clawlaw.in is our sister company. Its website and its app store listing carried different plan names and different prices.

ChatGPT noticed the contradiction and said so in its answer. That is worth sitting with. The wrong fact was not invented by the model and it was not hidden anywhere. Both versions had been published by the company, a reader could see both, and the assistant reported the disagreement to a buyer.

There was also nobody to write to, because both sources were ours. A case like this is the five steps above with step four removed: write down what the answer said, find every place the price is stated, correct each one in place with the date of the change, and make one page the plain statement of what the price is now.

The general rule it produces is the one at the top of this page. Where two of your own pages disagree, the assistant gets to choose, and it may choose to report the disagreement instead of picking. Consistency here is not tidiness. It is the difference between one fact and an argument your buyer reads.

What it means for a business

Deleting the old page is usually the wrong move. A deleted page leaves the wrong fact in circulation elsewhere with nothing to contradict it. Better to keep the URL and correct it in place, with a dated line saying what changed.

Consistency does the heavy lifting. If your pricing page, your directory profiles and your one line description all say the same thing in the same words, the correct version is what any reader finds. If they disagree with each other, you have given the assistant a choice, and it may not choose the current one.

Wrong facts also deserve priority over general visibility work. A wrong price or a wrong segment routes buyers away from you while sounding helpful, which is worse than not being mentioned at all.

What you cannot do

Five limits, stated plainly, because the false promises in this area are expensive.

You cannot edit an answer. There is no publisher console, no correction form that changes what the model says, and no account setting that pins your version.

You cannot rely on the thumbs down. Feedback controls exist and cost nothing to use. They are not a correction mechanism you can depend on, track or report.

You cannot make the model forget. If the wrong fact is in what the model holds, it can keep appearing in uncited answers until a future training run, and neither you nor your agency controls that date.

You cannot force a third party to change a page. You can ask well, with facts. Some will not reply.

You cannot be sure the correction is complete. Answers vary between runs, so one correct answer is not proof. Watch several readings before you call it fixed.

Legal routes exist for defamation and for personal data in some jurisdictions, and they are slow, expensive and out of scope for an ordinary wrong price. Treat them as a last resort for genuine harm, not as a correction workflow.

Common questions

Can I contact OpenAI or Google to fix it?

You can use the feedback control in the app, and for genuinely harmful or personal data issues there are formal channels. For an ordinary out of date fact, the reliable route is changing the sources and waiting. Plan on that, and treat any reply as a bonus.

How long does a correction take?

On the live path it can be weeks, because it depends on pages being re-crawled. On the remembered path there is no honest estimate, and anyone who gives you a date is guessing. Re-check monthly and report both paths separately.

The assistant is mixing us up with another company. What helps?

Distinguishing facts, stated consistently. Your full legal name, your city, your sector and what you do, in the same words everywhere, including on every profile. Where the other company is well known, add the clarifying detail to your own pages rather than mentioning them.

Should I delete the old page that carried the wrong fact?

Usually no. Correct it in place, keep the URL, and date the change. A deleted page removes your corrected version while copies of the old fact survive elsewhere.

It says something about us that was true but is embarrassing now. Can that be removed?

Not by request. What you can do is publish the current position clearly and get it into the sources, so the more recent and better documented version is the easier one to find.

Does asking the assistant where it got the fact help?

Not reliably. It may produce a plausible sounding source that was not actually used. Trust the citation list, which is evidence, rather than the explanation, which is generated text.

What to do first

• Write down the wrong fact exactly as the assistant stated it, with the app, the date and whether there were citations.
• Ask the question again on two assistants and save any URLs they cite.
• Search the wrong value in quotes on the open web, and search your own site for it, including old blog posts and PDFs.
• Fix every instance you control, in place, with the date of the change.
• Send one short factual message to each third party page that still carries it.
• Re-check monthly, and log the searched and unsearched answers separately.

Tell whoever asked you to fix it that the searched answer should move first and the remembered one lags. That sets the expectation correctly and stops a normal wait being treated as failure.