AIRU

AI11 min read

ChatGPT is saying wrong things about your business: how to get it corrected

ChatGPT has no button that corrects a wrong answer about a company, and OpenAI has said it can only block output, not rewrite it. The fix that works is repairing the pages the assistant reads when it searches.

Seated crane holds a magnifying glass and a stack of papers
A wrong answer gets checked line by line before anyone files a request

You cannot make ChatGPT correct a wrong statement about your business on request. OpenAI told a complainant in April 2024 that it could only block data from appearing in some prompts, and could not correct itmarket. What you can change is the web pages the assistant reads when it searches, and that is where a repair usually starts.

Why does ChatGPT get facts about a company wrong?

ChatGPT builds an answer from two places: what the model absorbed in training and what it retrieves from the web when it runs a search. A wrong company fact usually comes from one of them, and the repair differs for each.

When the assistant searches, it can only cite pages it is allowed to read. OpenAI recommends allowing its crawler OAI-SearchBot in robots.txt, and sites that opt out are not shown in ChatGPT search answersmarket. A site that blocks the crawler by accident hands the answer to old directory listings and third-party pages. Pricing the page rewrites is simpler once you compare an SEO content agency with an in-house hire.

Wrong answers show up across every major assistant. In a study coordinated by the EBU and led by the BBC, 45% of answers had at least one significant issuemarket. Separately, 31% of the answers had serious sourcing problems such as missing or incorrect attributionmarket. To see how long such errors persist, a paid AI visibility tracker can log them weekly, though a manual check costs less.

The test covered news questions, so it does not measure company facts directly.

Gaps in the web make errors more likely. Google's head of search said in May 2024 that some wrong AI Overviews came from questions where very few high-quality sources existedmarket. A small business with a thin website and a stale listing sits in exactly that position, and our write-up of what makes ChatGPT cite one page covers which page features the assistant tends to quote.

Bald man with glasses looking worried at a document
Most owners first notice the error when a customer repeats it

What does OpenAI's own correction form do?

OpenAI runs a Privacy Portal request for personal data in ChatGPT answers, and it promises nothing in return. The company says submitting a request does not guarantee the information will be removed from outputs, and that it weighs privacy rights against the public interest in access to informationmarket.

The request asks for email verification, examples of the problematic output, the URLs or context where the information came from, and government IDmarket. The form is built around personal data about a named person. A wrong opening hour or a price list is a poor fit for it.

For a founder whose own name is attached to a false claim, the form is the only direct channel OpenAI offers. The complaint that shaped public understanding of its limits came from noyb, which filed against OpenAI with the Norwegian data protection authority in March 2025 after ChatGPT falsely said a man had murdered two of his childrenmarket. noyb asked the regulator to order the output deleted and the model fine-tuned.

Which correction route fits which problem?

Five routes exist, and each changes something different. The table matches each route to the kind of error it can reach.

RouteBest forWhat it can changeResponse time
OpenAI Privacy Portal requestFalse claims about a named personBlocks specific outputs, no guaranteeNo stated deadline
GDPR rectification requestPeople in the EU or UKObliges a response on inaccurate personal dataOne month, extendable by twomarket
Fixing your own pages and robots.txtWrong business factsThe sources a search-enabled answer draws onAbout 24 hours for crawler rulesmarket
Feedback on a Google AI OverviewWrong facts in Google answersSignals the answer was inaccurateNo stated deadline
LawsuitSerious reputational harmDamages, if the claim succeedsYears, no fixed date

The GDPR route rests on Article 16, which gives a person the right to have inaccurate personal data rectified without undue delaymarket. It applies to personal data, so it fits a sole trader or a named director, and it does not cover a company as such.

Most small-business errors sit in the third row. A wrong address, a discontinued service or an outdated price is a page problem, and pages are the one lever fully in your hands. The guide on how to rank in ChatGPT explains why the first paragraph of a page does most of the work in what gets quoted.

How do you fix the sources behind a wrong answer?

Fix the pages the assistant could have read, then wait for it to read them again. Start with a log: ask the same question in a fresh chat, save the full answer with the date, and note which sources it cites.

  1. Find the claim's origin. Open each cited page and check whether the error is printed there or whether the assistant invented it.
  2. Correct your own site first. Put the right fact in the first paragraph of the relevant page, in plain words, with the date it was last checked.
  3. Update third-party profiles, such as Google Business Profile, directories and review sites, so they say the same thing as your site.
  4. Check robots.txt and allow OAI-SearchBot if the site blocks it, since the crawler is what feeds ChatGPT search answersmarket.
  5. Ask the question again after a few weeks and record the change.

Training data does not update on request, so an error that lives in the model itself can persist for months. A search-enabled answer can change as soon as the retrieved pages do, and our look at how long until a new post gets cited gives a sense of the lag to plan for.

Do not pay for a promise of guaranteed correction. No vendor controls what OpenAI's model says, and the company itself declines to guarantee removalmarket.

Hands writing on tax documents with laptop, glasses, and currency on desk
A dated written record is what any later request or complaint rests on

Can you sue OpenAI over a false answer?

Suing is possible and has not worked so far in the best-known US case. On 27 May 2025 a Georgia state court granted summary judgment to OpenAI in the Walters defamation casemarket. The court gave three grounds: the output was not defamatory as a matter of law, there was no evidence of negligence, and there were no damagesmarket.

One detail matters for any business owner. Punitive damages were also unavailable because the plaintiff never asked OpenAI to correct or retract the output, which Georgia law requiresmarket. The user in that case had also seen on-screen warnings that ChatGPT can produce inaccurate information about people, places or facts.

A written correction request therefore costs little and may keep a legal option open later. It is not legal advice, and the rules differ by country, so a lawyer in your jurisdiction decides whether a claim exists.

How do you check the fix worked?

Repeat the original question under the same conditions and compare it to your saved log. Use a fresh chat, the same wording, and the same location settings. Run it on more than one day, since answers vary.

Check the cited sources as well as the text. If the assistant now cites your corrected page, the retrieval side is fixed. If the text is right but cites an old directory, that directory is the next page to update.

Keep the log going monthly. Our guide to what a ChatGPT recommendation costs helps separate paid placement from organic mentions when reading the results.

Frequently asked questions

Can I report wrong information about my business to OpenAI?

Yes, through the Privacy Portal, but the form concerns personal data and OpenAI says a request does not guarantee removalmarket. For company facts such as hours or prices, correcting the web pages the assistant reads gives a more reliable result.

Does a GDPR request work for a company?

Article 16 gives the right to rectification to individuals, so it fits a named person such as a sole trader or directormarket. A company as a legal entity has no personal data to rectify. The controller must respond within one month, extendable by two for complex requestsmarket.

Should I allow OAI-SearchBot?

If you want to appear in ChatGPT search answers, yes. OpenAI recommends allowing the crawler in robots.txt, and sites that opt out are not shown in those answersmarket. Blocking it does not affect the separate crawler used for model training.

How do I fix a wrong Google AI Overview?

Use the feedback control on the answer to mark it inaccurate, then correct the pages and profiles Google reads. Google has said some wrong overviews came from topics with few high-quality sources, so a clear page on your own site fills that gapmarket.

Is a wrong ChatGPT answer a legal problem for OpenAI?

Not necessarily. In the Walters case the court found no defamatory meaning, no negligence evidence and no damagesmarket. Each country and fact pattern differs, so legal advice is needed for a real claim.

Sources

  1. VRT communication on the EBU/BBC study "AI assistants misrepresent news content", 22.10.2025; Advanced Television, same date. Used for the 45% of answers with a significant issue and the study scope.
  2. AIB, "AI assistants misrepresent news content, says major study", October 2025; Sofia Globe, October 2025. Used for the 31% with serious sourcing problems.
  3. OpenAI Help Center, "Right to be forgotten and personal data removal from ChatGPT"; The Decoder, May 2024. Used for the no-guarantee wording and the form's introduction.
  4. OpenAI Help Center (same article); BGR, "How to delete your personal data from ChatGPT". Used for the verification, examples, URLs and ID requirements.
  5. Onmanorama (AFP), 20.03.2025; Engadget, March 2025. Used for the noyb complaint to the Norwegian regulator.
  6. 80.lv, April 2024; Mezha, 29.04.2024. Used for OpenAI saying it could block but not correct data.
  7. Loeb & Loeb, "Walters v. OpenAI LLC", May 2025; Covington, 16.06.2025. Used for the 27 May 2025 summary judgment and its three grounds.
  8. Loeb & Loeb (same article); Cleary Gottlieb, June 2025. Used for the punitive-damages point and the on-screen warnings.
  9. GDPR text, Article 16; IAPP, "DSARs: do you know how to calculate your legal response time under the GDPR". Used for rectification and the one-month response time.
  10. OpenAI developer docs, crawlers overview; Botcrawl, OAI-SearchBot. Used for the robots.txt recommendation and the delay for crawler rules.
  11. Maginative, 31.05.2024; BGR, June 2024. Used for Google's statement about information gaps behind wrong AI Overviews.

Check your site in a minute

Paste your address: we'll show your score out of 100, whether an AI assistant names you, and which searches your site has no page for. Free.