What can be done when ChatGPT gives wrong information about a brand?

When ChatGPT gives wrong information about a brand, the most effective response is to find the source the error comes from and correct it there, then report it to OpenAI, then ask the same questions again to measure whether it changed. GeoMono is a platform that measures how visible brands are in AI answers; this guide walks through the steps specific to ChatGPT, in order.

The short answer: you cannot change ChatGPT's answer directly. What you can change is the pages ChatGPT reads. Our general guide for all engines is When AI gives wrong information about your brand; this one focuses on ChatGPT.

Step 1: Record the error

Before you fix an error, record it exactly. A screenshot is not enough; write down:

• The question: the exact wording that produced the error.
• Date and version: when you asked, and which version of ChatGPT you used.
• The full answer: the whole answer, not just the wrong sentence.
• Cited sources: every source link the answer showed, if any.
• The correct fact: what is actually true, and which of your pages says so.

Ask the same question several times. Does the error appear every time, or only in some answers? An error in one answer and an error repeated in every answer carry different priority.

Note what kind of error it is: outdated information such as an old address or a closed branch, confusion with another company, a product or campaign that never existed, a wrong price or a wrong service area. Each comes from a different kind of source and needs a different fix.

Which errors come first?

Not every error matters equally. Start with the ones that would do the most harm if a buyer or stakeholder acted on them: a wrong price, a wrong service area, a product that does not exist, a sentence that describes you with another company's problems. Then move to errors that repeat in every answer. An error seen once, with no source cited, is a finding to record and watch, but not the first one to work on.

Step 2: Find where the error comes from

According to OpenAI's help page, ChatGPT can search the web automatically when it judges that a question would benefit from current information. So an error can arrive by two routes.

If the answer cites sources, the error most likely comes from one of those pages. Open each link and check whether the wrong fact is written there. Usually you will find one of these:

• An old page on your own site that was never updated (an old price list, a closed branch, a past campaign).
• An outdated entry on a directory, marketplace or news site.
• An unverified claim on a complaint site or forum.
• A page written about a different company with the same name.

If the answer cites nothing, you cannot see from outside what it rests on. The model may have produced the error from what it learned earlier. What you can do is make sure the correct information is stated on your own site plainly, currently and where it is easy to find.

Step 3: Fix the source

The lasting fix is made on the page ChatGPT reads.

Your own site. Update the page carrying the wrong information, and check the same fact is not written differently elsewhere on the site: the footer, the contact page, old blog posts, PDF brochures. Give the correct fact as text in server-rendered HTML.

ChatGPT's access to your site. According to OpenAI's crawler documentation, OAI-SearchBot is used to show sites in ChatGPT's search features, and sites that block it are not shown in ChatGPT's search answers. If you have corrected your site but your robots.txt blocks OAI-SearchBot, ChatGPT cannot read the corrected page and may keep taking the error from other sources.

Someone else's site. Write to the page owner with a source that shows the correct fact. Answer an old claim on a complaint site where it is read, in a calm tone.

Namesake confusion. State your full legal name, city, what you do and whom you serve plainly on your site. For ChatGPT to tell you apart from a company with the same name, the distinguishing facts have to be readable.

Next step

Is AI telling the truth about your brand?

The accuracy audit compares what the engines say about your brand with the facts on your site. With nothing to compare against, the result is shown as not measurable, not as clean.

See the accuracy audit

Step 4: Report it to OpenAI

According to OpenAI's content reporting page, you can report a message by tapping the thumbs-down icon under it and choosing the problem; there is a "Safety or Legal concern" option for legal matters. The same page says reported domains may be reviewed by OpenAI's model quality team and that filters may be applied to stop ChatGPT relying on unreliable sources. There is a separate form for trademark infringement.

If the wrong information concerns a person rather than the company (your founder or an executive, for example), OpenAI's personal data removal page explains that requests can be made through its Privacy Portal under certain privacy laws such as the GDPR. Requests are assessed individually, and information removed from ChatGPT is not thereby removed from other sites or search engines.

None of these channels promises that a report will change a particular answer. Make the report, but make the real fix at the source.

Step 5: Measure again with the same questions

After the fix, ask the questions from your first record again, in the same way, and record the results with their dates. If you change the questions, any difference between the two records may come from the questions rather than the fix.

Do not expect the result at once. A page takes time to be read again, and OpenAI does not publish how long. Ask the same questions at regular intervals.

One caution: an answer without the error does not necessarily mean the error is gone. Engines do not give the same answer to the same question every time. Count in how many answers the error appears, before the fix and after.

"No errors found" can mean three different things

When a check says "nothing wrong was found about you", it can be any of three situations:

1. The check never ran. The engine did not answer, or the answer was cut short.
2. There was nothing to compare against. The engine said something, but the correct fact is not on your site, or your site could not be read.
3. There genuinely is no error. The answer was read, its claims were compared with the correct facts, and no conflict was found.

Only the third is good news. For a brand whose site blocks crawlers, the check often ends in the second case; the brand most at risk of being described wrongly and the brand whose check comes back empty can be the same brand.

GeoMono's Accuracy Check keeps those three states apart. It runs inside the crawl-based analysis: it compares the verifiable claims in the answers with the brand's own pages, marks a conflict low, medium or high by how much harm it would do to a buyer who acted on it, and adds it to the analysis roadmap as a step. A check that did not run is never written as a clean result. Registered measurement reports do not include this section today, and the report says so plainly.

Frequently asked questions

Can I get ChatGPT to correct the wrong information directly?

No. Typing the correct fact into a conversation may change the answer in that conversation, but it does not change the answer other users get. The lasting fix is made in the source ChatGPT reads.

Does reporting with the thumbs-down button help?

OpenAI says reported domains may be reviewed by its model quality team, but it does not promise that a particular answer will change. Make the report; make the real fix at the source.

ChatGPT confuses us with another company. What should we do?

State your full legal name, city, founding details and what you do plainly on your site. If the answer cites sources, check whether the page behind the confusion is about you or about your namesake.

How long until the fix shows?

OpenAI does not publish a timeline. Ask the same questions at regular intervals and count in how many answers the error appears.