ChatGPT has wrong information about my company: how to fix it

A prospect forwards you a ChatGPT answer that gives your price from two years ago. Or Gemini says you don't offer something you've offered since launch. Or an AI Overview attributes a competitor's feature to you. The instinct is to find the place to report it. There mostly isn't one, but there is a reliable way to fix it.

Why there's no edit button

Assistants don't keep a database of facts about companies that you could correct. They write each answer fresh, from two places: what the model learned in training, and the pages it reads while answering. A wrong fact comes from one of those, so the fix is to change what they read. Our guide to how AI chooses which brands to recommend explains the two sources in more detail.

There are feedback buttons. On Google, you can give an AI Overview a thumbs-down and then "click Report a problem and choose the category that best describes your issue". ChatGPT and the other assistants have thumbs-down too. Use them, but don't rely on them: they're feedback on one answer, not a correction to the source.

Step 1: collect the wrong answers

Ask the questions that prompt the fact, in the ways buyers would: "how much does Prepco cost?", "does Prepco include practice exams?", "Prepco vs [rival]". Ask each several times, on each assistant. Note:

• The exact wrong statement.
• How often it appears (one answer in ten is a different problem from nine in ten).
• Whether the answer cites sources, and which.

Step 2: find where it comes from

Open every cited page and search it for the wrong fact. You'll usually find one of these:

Your own site is the most common culprit. Old blog posts announcing a price, a cached PDF brochure, a help article describing a retired plan: they're still crawlable and still read.

Step 3: correct it at the source

On your own site, update the old page with the current fact and the date it changed, or redirect it to the current page. State key facts in plain text near the top of the pages that should carry them: pricing, what's included, who it's for. Don't hide the new price in an image.

On third-party pages, write to the author or editor with the correction, a link to where the current fact is stated, and nothing else. Most publications fix errors gladly. Update your profiles on review sites and directories yourself.

When nothing is cited, the fact is coming from training data. Make the correct fact consistent across your site and the main sites that describe you, so future searches and future training snapshots find one story. This is the slowest case.

Step 4: make sure they can read the correction

A correction no crawler can see changes nothing. OpenAI says sites that opt out of its search crawler "will not be shown in ChatGPT search answers", so check robots.txt allows OAI-SearchBot (our guide to AI crawlers and robots.txt lists the others). Then ask for re-crawling where you can: Google's URL Inspection tool in Search Console, or an IndexNow ping where it's supported.

Step 5: check that it took

Ask the same questions again after two to three weeks, several times each, and count how often the wrong fact still appears. Answers that search the web usually update once the cited page is re-crawled. Answers from memory can lag for a long time; keep the correct fact consistent and keep checking.

Prevent the next one

• One page per fact. Keep pricing, plans and what's included on one current page, and link to it everywhere else.
• Date your facts, so readers and assistants can tell current from historical.
• Retire old pages deliberately: redirect them rather than leaving them live.
• Track fact questions weekly. A wrong fact caught in the week it appears is much cheaper than one a hundred prospects have read.

Fixing what's wrong comes first in the order of work in our guide to getting named in ChatGPT, because every time you're named, a wrong fact is repeated. Once it's fixed, writing answer-first pages keeps the right facts easy to find.