ChatGPT Search: What It Says About You and How to Correct It

ChatGPT search is the feature that lets ChatGPT look things up on the web while it answers, then show links to the pages it used. When someone asks ChatGPT about you or your business, the answer may come from those live web results, from what the model learned in training, or from a mix of both. You can’t edit what ChatGPT says, but you can report wrong answers, use OpenAI’s privacy request process for inaccurate personal data, and fix the sources it draws on.

This guide is about ChatGPT specifically. For the general approach across Google, Perplexity and other assistants, start with our guide to managing your reputation in AI search.

How ChatGPT search works

ChatGPT can answer in two broad ways. Sometimes it answers from its training data alone, which reflects the web as it was when the model was trained. Other times it searches the web, either because the user asked it to or because it decided the question needs current information, and then it writes an answer with citations.

The difference matters for your reputation:

Answer type	What you’ll see	Typical problem	How it gets fixed
From training data	No source links, or general statements	Old job titles, closed locations, stories since corrected	Slowly, as newer models are trained on updated sources
From web search	Inline citations and a list of sources	A wrong or outdated page being quoted, or a misread source	Fixing or outranking the cited page, then waiting for it to be recrawled

OpenAI has said ChatGPT search uses third-party search providers as well as its own technology, and it runs a crawler called OAI-SearchBot to find pages for search answers. A separate crawler, GPTBot, is used for collecting training data. Website owners can control each one in their robots.txt file.

What ChatGPT says about people and businesses

Ask ChatGPT “Who is [name]?” or “Is [business] legit?” and you’ll typically get a short profile: what the person or business does, where, notable history, and what reviews or coverage say. For businesses, it often summarizes review themes and any widely reported complaints. For people, it leans on professional profiles, company pages, interviews and news.

The weak spots are predictable. People with common names get blended with others. Small businesses with thin coverage get vague or padded descriptions. And old training data can keep a past role, address or controversy alive long after the web has moved on. When ChatGPT states something no source supports, that’s an AI hallucination, which needs a slightly different response.

How to check what ChatGPT says about you

Your own ChatGPT account can personalize answers. If memory or custom instructions are on, ChatGPT may use what it knows about you, which skews the test. Our guide to ChatGPT memory explains how that works.

1. Use a clean session. Test in a temporary chat, signed out, or in an account without memory or custom instructions, where available.
2. Ask with and without search. Ask your questions normally, then ask ChatGPT to search the web for the same thing. Compare the answers.
3. Use real customer questions. “Who is [full name]?”, “What does [business] do?”, “Is [business] a scam?”, “What do customers complain about?”, “[business] vs [competitor]” and “Best [service] in [city].”
4. Record everything. Save screenshots, the date, the exact prompt and every cited link. Note whether each claim came with a source.
5. Trace each claim. Open the cited pages and find the sentence behind each statement. Where there’s no citation, search the claim yourself to find which pages might have fed it.

Repeat the test a few times. ChatGPT’s answers vary, and a claim that appears once in five tries needs less urgency than one that appears every time.

How to correct wrong information in ChatGPT

1. Report the answer

Use the thumbs-down or report option on the specific answer and explain what’s wrong in a sentence, with the correct fact. This is feedback to OpenAI about the model’s quality. It doesn’t change the answer for other users right away, and OpenAI doesn’t promise a response.

2. Use OpenAI’s privacy request process for personal data

If ChatGPT produces inaccurate personal information about you, OpenAI accepts requests through its privacy portal to correct or remove personal data, and it considers them case by case. OpenAI has said that where it can’t technically correct an inaccuracy, it may be able to remove the personal data instead. Be specific: the prompt you used, what it said, why it’s wrong and what’s correct. Our guide to ChatGPT privacy covers the other privacy settings and requests available to you.

3. Fix the pages it cites

For answers based on web search, the cited pages are the lever. Correct your own website, LinkedIn and business listings first. Then ask third-party publishers and directories for corrections, with evidence. Once a page changes, ChatGPT’s search can pick up the new version when the page is recrawled.

4. Make sure ChatGPT can reach your side of the story

If your site blocks OAI-SearchBot in robots.txt, ChatGPT search may not be able to cite your pages at all, leaving other people’s pages to describe you. Check your robots.txt, or ask your web developer to. Whether you also allow GPTBot for training is a separate decision about how your content is used, not about search visibility.

Because ChatGPT search draws on third-party search providers, making sure your site is indexed properly by major search engines also helps. Our guide to Bing Webmaster Tools covers submitting your site and checking for crawl problems.

5. Strengthen the sources for training data

Older, uncited errors come from what the model learned. You can’t edit that, but future models learn from the web as it is then. Clear, consistent information on your site, professional profiles, directories and independent coverage improves the odds that later versions get it right.

A worked example

This is an illustrative scenario, not a real client. “Tessa Monroe” moved from a retail chain to a nonprofit two years ago. When a board member asks ChatGPT about her, the answer says she is still head of marketing at the retailer and “was involved in the chain’s 2019 data breach response.”

Tessa traces it. The old job comes from an unsearched answer, most likely training data from her former employer’s leadership page. The data breach line comes from a news article that quoted a different executive, which ChatGPT seems to have connected to her.

1. She reports both answers with thumbs-down and a one-line correction each.
2. She submits a request through OpenAI’s privacy portal explaining the false association with the data breach, with a link to the article showing the quote came from someone else.
3. She updates LinkedIn, her speaker bio and the nonprofit’s team page, and asks the former employer to remove her from an archived leadership page.
4. She publishes a short professional bio page on her own site that lists her roles and dates.

When she tests again with web search on, ChatGPT gives her current role and cites the nonprofit page. Without search, the old role still appears some of the time. That split is typical: search answers catch up faster than training data.

What you can’t do

You can’t pay OpenAI to change what ChatGPT says, and nobody can promise a particular answer. You also can’t remove accurate public information just because it’s unflattering. If ChatGPT accurately summarizes a real news story about you, the realistic work is adding context and building stronger current sources, the same approach we take in personal reputation management.

Common mistakes

• Testing in your own personalized account. Memory and past chats can make answers about you look better or worse than what others see.
• Correcting ChatGPT in the chat. Telling it “that’s wrong” changes nothing for other users.
• Blocking every AI crawler without thinking. Blocking search crawlers can remove your own pages from answers about you.
• Only fixing the cited page. Uncited errors come from older data and need consistent sources everywhere.
• Waiting for a reply. Feedback is useful, but you usually won’t hear back. Keep working on the sources.

Frequently asked questions

Is ChatGPT search the same as Google?

No. ChatGPT search writes an answer and cites pages, rather than showing a ranked list. It draws on third-party search providers and its own crawling, so the sources it picks can differ from Google’s page one, though strong pages about you tend to appear in both.

Can I delete what ChatGPT says about me?

Not directly. You can report the answer, and if it involves inaccurate personal data, ask OpenAI through its privacy portal to correct or remove it, which it reviews case by case. Fixing the underlying sources is the most dependable long-term step.

Why does ChatGPT say something different when it searches the web?

Without search, ChatGPT answers from training data that may be out of date. With search, it uses current pages. If the searched answer is right and the unsearched one is wrong, the web is already fixed and you’re waiting for newer models.

Does deleting my ChatGPT history change what it says about me?

No. Your chat history affects your own account, not what ChatGPT tells other people. Our guide on deleting ChatGPT history covers what that does control.