# Report ChatGPT false information about a person

How to report ChatGPT false information about a person, and the same for Gemini, Copilot, Claude, Perplexity and Grok: which form, claim and route.

---


To report ChatGPT false information about a person, use the OpenAI Privacy Portal and choose the option to remove personal data from ChatGPT responses, with examples, links and proof of what is true. The thumbs-down button is a separate, lighter route. Every other major assistant splits the same way: a quick feedback button, a privacy or rights request about personal data, and a legal report for statements that may be unlawful.

This guide is a map of those three lanes, provider by provider, as each provider's own pages described them on the day we read them. It does not repeat how to diagnose a wrong answer or how to fix the web pages an assistant reads. The point here is narrower: which form to use, what it asks for, what it can and cannot change, and in what order to file when more than one assistant is wrong about you.

Providers rewrite these pages often, and some were not readable for this guide; the text says where. Nothing here is legal advice, and no provider page we read promises a result.

<div class="box"><b>The short version</b>
<ul><li>Every provider offers up to three lanes: feedback on one response, a privacy request about personal data, and a legal report. They are not interchangeable, and a feedback click is never a substitute for a request.</li><li>Lane choice depends on the claim. Wrong personal data points to the privacy lane. A statement you believe is defamatory or unlawful points to the legal lane where one exists. A one-off slip points to feedback.</li><li>A provider can act on its own assistant only. OpenAI's page says removal stops information appearing in ChatGPT responses and does not touch websites or search engines.</li><li>File with evidence that a stranger can check, one claim per request, and keep a log of what you sent, where and when.</li></ul></div>

## The three lanes every provider uses, and what each one can do

The first sentence to settle is what you want from the provider. Providers separate three requests that people lump together as "reporting".

The feedback lane is a button on the response. Its purpose is product improvement. Google's Gemini page, for example, says submitted feedback includes the conversation. A feedback click may help a specific response get reviewed, but none of the pages we read promise a reply, a correction or a timeline.

The privacy lane is a rights request about personal data. It rests on privacy law, mainly in the EU and UK but also in some other places, and on the provider's own policy. In this lane you are the person the data is about, you prove who you are, and the provider weighs your request against other interests. This is where OpenAI's removal option, Anthropic's correction request and xAI's portal sit.

The legal lane is a report that content breaks a law, such as defamation, or that a court has ordered action. Google's routing page and Gemini's "Report legal issue" belong here, and Microsoft points legal concerns to its Report a Concern page. This lane is aimed at content, not at your personal data, and it usually asks you to name the law.

Two things follow. You can use more than one lane for one error, and often should. And the lanes differ in work and reach, as the next table shows.

<div class="tablewrap"><table><thead><tr><th>Lane</th><th>Best for</th><th>Proof asked</th><th>What it can change</th></tr></thead><tbody>
<tr><td><strong>Feedback</strong></td><td>A single wrong response, low stakes</td><td>None, the response is attached</td><td>Possibly a review of that response; no promise</td></tr>
<tr><td><strong>Privacy request</strong></td><td>Wrong or unwanted personal data in answers</td><td>Identity, plus evidence for any disputed fact</td><td>What that assistant says about you; not the web</td></tr>
<tr><td><strong>Legal report</strong></td><td>A statement you say is defamatory or unlawful</td><td>The legal basis, often your details and the exact text</td><td>Availability of that content on that product, often in one country</td></tr>
</tbody></table></div>

## Which lane fits your claim

Match the claim to the lane before you open a form. A mismatched request tends to go nowhere: a privacy request that is really a complaint about opinion, or a legal report about something merely out of date.

A short list of claims and where each usually belongs:

- A wrong date, job title or employer: feedback first, then the privacy lane if it persists. These are inaccuracies in personal data, which is what correction and rectification requests are for.
- An invented fact, such as a qualification, award or role you never had: the privacy lane, with proof of what is true. If it also harms your standing, add the legal lane where the provider offers one.
- A statement that you committed a crime or professional misconduct, when you did not: the legal lane and the privacy lane together. Consider a lawyer before you file, because what you send becomes part of the record.
- Facts that belong to another person with a similar name: the privacy lane, because the error is that data about someone else is attached to you. Say plainly that it is a different person and provide what distinguishes you.
- An accurate but unflattering fact about a real event: no lane is a reliable fit. OpenAI's own page weighs freedom of expression and the public interest, and lists criminal allegations or misconduct and time passed among its factors. A request to hide an accurate public record tends to fail.
- A statement of opinion: usually neither a correction nor a legal matter.

<aside class="callout">
<span class="callout-label">Rule</span>
<h4>One request, one claim</h4>
<p>Put one false statement in each request, with its own evidence. Reviewers act on a specific sentence they can compare with a document. A request that bundles six complaints is easy to decline in part and hard to approve in whole.</p>
</aside>

## What each provider's pages say about reporting

Each section states what the page read said, and what it left out.

### OpenAI and ChatGPT: the Privacy Portal route

OpenAI documents the strongest personal route of the providers we read. Its help page on right to be forgotten and personal data removal from ChatGPT says to go to the Privacy Portal at privacy.openai.com, choose "Make a Privacy Request", and select "Remove personal data from ChatGPT responses". You can read the steps on [OpenAI's help page on personal data removal](https://help.openai.com/en/articles/20001057-right-to-be-forgotten-and-personal-data-removal-from-chatgpt).

The page asks for specific examples, including links to chats and any referenced URLs, the type of data (a name, email, phone number or URL), and an explanation of why the data is inaccurate, excessive, irrelevant or no longer appropriate. If you say something is incorrect, it asks for reliable supporting evidence, because OpenAI says it cannot independently investigate disputed facts. A government-issued ID or other acceptable proof is required.

OpenAI says it weighs each request against freedom of expression and the public interest, with factors such as a public or professional role, accuracy, the age of the information and the impact on you. Approval is not guaranteed.

The page also says you can complain to your local supervisory authority. We could not confirm a response deadline, so do not assume one.

What this route does not do is as important. OpenAI's page says removal stops information appearing in ChatGPT responses and does not remove it from external websites or search engines. If a live page is the source, the page still needs its own action.

### Google: AI Overviews, AI Mode and Gemini

Google runs three surfaces with different routes, and the pages we read treat them separately.

For AI Overviews, Google's help page says to click thumbs-down on the overview, choose "Report a problem", pick the category that fits and submit. The same page says AI responses may include mistakes. AI Mode has its own help page, which we do not summarize here, and the AI Overviews page offers no legal route, so that comes from Google's separate routing page.

That page, the [Google legal removal troubleshooter](https://support.google.com/legal/troubleshooter/1114905), asks whether the request relates to content generated by AI within a Google product, with AI Overviews or AI Mode as options. The legal reasons it listed include court order, defamation, copyright and trademark, among others. It also says you must file a separate report for each Google product where the content appears, and that legal removals typically apply only in the country or region where the content is deemed illegal.

Inside the Gemini app, Google's Gemini help page on feedback describes a Good or Bad response icon, a Send feedback option under Settings and help, and, below a response, More and then "Report legal issue". The same page says submitted feedback includes the conversation and may include uploaded files and personal content from connected apps, and that it describes no removal request process. Before you use it, think about what the conversation contains.

One wrong claim can appear in Search, AI Mode and Gemini, and each is a separate filing. The Google pages read describe no privacy-request route for AI answers.

### Microsoft Copilot

Microsoft's Copilot terms of use say Copilot can make mistakes and may not work as intended, and advise you to verify its information. They name two actions. For wrong or inappropriate responses, use the Report or Feedback features in Copilot. For legal concerns about what Copilot says, the terms direct you to the Report a Concern page, and the same page is listed for harmful content.

We could not read the Report a Concern page for this guide: its address returned the Microsoft homepage. We cannot say what it asks or who may use it, so read it before investing time. The terms page also describes no dedicated rights process for personal data in outputs, and Microsoft's general privacy statement was not read in this session. On the evidence we have, treat Microsoft as a two-lane provider: feedback and Report a Concern.

### Anthropic and Claude

Anthropic's [privacy policy](https://www.anthropic.com/privacy) is direct about outputs. It says Anthropic cannot guarantee the factual accuracy of outputs, that you can request correction of inaccurate personal data it holds about you, and that if an output contains inaccurate personal data about you, you can submit a correction request. Anthropic says it will make a reasonable effort to fix it, and that the technical complexity of its models means a correction may not always be possible.

The policy does not describe a dedicated request form. It says you or an authorized agent may contact Anthropic, and lists a privacy contact address for questions, requests and appeals. It says Anthropic may ask for information to confirm your identity, and that a third party must provide evidence of authority. On timing, it says Anthropic responds within the period required by the law that applies to you. As an example, under EU or UK GDPR it responds within one calendar month of a verifiable request, and for complex requests or several in a short period that period may be extended by up to two further months. If a request is denied, you can appeal by emailing the same address.

With no form, write the email as one would: your name, how you can be identified, the exact output and date, why it is wrong, and the document that shows the truth.

### Perplexity

Perplexity's privacy notice says that if a service does not let you update your account details, you can ask Perplexity to correct inaccuracies by contacting it. We could not re-read the notice's form, appeal and timeline details for this guide, so read them on the day you file.

One caution: the notice speaks of correcting inaccuracies in the context of personal data and accounts. It does not say that a wrong sentence in an answer can be corrected through the form, so ask for exactly that, and expect the real fix to lie in the sources the answer cites. Perplexity's help center could not be read for this guide, so we cannot say how its in-answer report option works.

### xAI and Grok

xAI's privacy policy says it cannot guarantee the factual accuracy of output from its models. If an output contains inaccurate personal information about you, you can submit a correction request, and the company will make reasonable efforts to correct it, though that may not always be feasible. [xAI's privacy policy](https://x.ai/legal/privacy-policy) points to a privacy portal for requests and asks that you include your full legal name, email address and city, state or province, and country of residence. Identity may need to be verified, an authorized agent must provide written permission, and a request that is not acted on within a reasonable time can be appealed through the same portal. The policy lists local contacts for the UK, the EU and Switzerland and a data protection officer address.

The portal asks for your full legal name and location. Decide what you are willing to hand over, and file only through the portal the policy names.

### Meta AI

Meta's privacy policy and generative AI pages could not be read for this guide: the site blocked the fetch. We make no statement about Meta AI's forms, objection rights or timelines. Open Meta's privacy center from inside its apps, read what it offers, and apply the same evidence rules.

## Compare the routes side by side

The table below puts the readable providers in one place. "Not described" means the pages we read said nothing on the point, not that no process exists.

<div class="tablewrap"><table><thead><tr><th>Provider</th><th>Personal-data route</th><th>Legal route</th><th>Gap on the pages read</th></tr></thead><tbody>
<tr><td><strong>OpenAI</strong></td><td>Privacy Portal, remove personal data from responses; ID required</td><td>Not described on the page read</td><td>No deadline confirmed</td></tr>
<tr><td><strong>Google Search</strong></td><td>Not described for AI answers</td><td>Legal troubleshooter, per product, often one country</td><td>AI Mode is on a separate page</td></tr>
<tr><td><strong>Gemini app</strong></td><td>Not described</td><td>Report legal issue under More</td><td>No removal process described</td></tr>
<tr><td><strong>Microsoft Copilot</strong></td><td>General privacy statement, not read</td><td>Report a Concern, page not readable</td><td>Scope of the form unknown</td></tr>
<tr><td><strong>Anthropic</strong></td><td>Correction request by email; reasonable effort</td><td>Not described</td><td>No form; correction may not be possible</td></tr>
<tr><td><strong>Perplexity</strong></td><td>Contact route; correction of inaccuracies on request</td><td>Not described</td><td>Timelines unconfirmed; answer-level scope unclear</td></tr>
<tr><td><strong>xAI</strong></td><td>Privacy portal; correction request</td><td>Not described</td><td>Asks for legal name and location</td></tr>
<tr><td><strong>Meta AI</strong></td><td>Not readable</td><td>Not readable</td><td>Page blocked for this guide</td></tr>
</tbody></table></div>

The table shows an asymmetry worth planning around. Personal-data routes are the most developed at OpenAI, Anthropic and xAI, and legal routes are the most developed at Google. If your claim is defamation and you are dealing with Google surfaces, the legal lane is the one with a documented path. If your claim is wrong personal data in a chat assistant, the privacy lane is.

## What to put in the request

The evidence rules are the same in every lane, and the lanes differ in how much the reviewer will do for you. Here is what a request itself should hold.

<figure class="fig">
<figcaption><b>Assembling one request</b></figcaption>
<ol class="fig-steps">
<li><b>Quote the sentence</b>Copy the exact false statement, the question that produced it, the assistant and the date. A screenshot alone is not enough where the form wants links or text.</li>
<li><b>State who you are</b>Give the name the assistant used, plus what distinguishes you from anyone with a similar name: a profession, a city, a registered number you are willing to share.</li>
<li><b>Say what is true</b>Write the correct fact in one sentence and attach a document that a stranger can check: a register entry, an official page, a dated letter.</li>
<li><b>Name the remedy</b>Ask for the specific outcome the lane offers: removal from responses, correction of the personal data, or a legal review of the content.</li>
<li><b>Prepare identity proof</b>Where ID is required, decide in advance which document you will send and send it only through the channel the policy names.</li>
<li><b>Log it</b>Record the date, the channel, the reference number if you get one and a copy of what you sent.</li>
</ol>
</figure>

Keep the tone factual: reviewers cannot verify anger. If you file for someone else, Anthropic and xAI each describe an authorized agent route that needs written permission or proof of authority, so get that first.

## The order to file in when several assistants are wrong

Start with the source. If the assistant cites a page, or the claim plainly comes from one, correct that page first. A provider can only change its own assistant, so a bad page keeps feeding every system that reads it. Then file the privacy or legal request at the assistant your audience is most likely to use, and at the others, each referring to the same evidence so the claim is consistent. Use feedback buttons alongside, not as the plan: a thumbs-down takes seconds, but it is not a record of a request.

## A worked scenario: one false claim, four assistants

This is an invented scenario with generic parties. A consultant discovers that ChatGPT, Gemini, Copilot and Claude all say she was removed from her professional register. In fact an old directory page lists a lapsed membership that she renewed, and the register shows her as current. The claim is false, damaging to her work and traceable to one source.

She first asks the directory to update the page and gets a reply that it will review it. While that is pending she files at OpenAI through the Privacy Portal with the register entry as proof, and at Anthropic by email with the same document. For Gemini she uses "Report legal issue" below the response, because the claim bears on her professional standing, and she reads the page's note that the conversation is included in what she submits. For Copilot she uses the in-app report, then opens the Report a Concern page, reads what it asks and files if it fits. She logs every filing.

The scenario shows the sequence, not a promise of outcomes. When the directory updates, she re-tests each assistant on the same question and files a short follow-up where the old claim persists.

## What a realistic timeline looks like

Provider pages mostly give no deadlines, so the windows below come from our experience on files and are written as ranges that vary. The only dated commitment we read is Anthropic's: under EU or UK law, one calendar month, extendable by up to two more, which describes its own handling and is not a rule for other providers.

<div class="tablewrap"><table><thead><tr><th>Stage</th><th>What you do</th><th>What usually happens</th></tr></thead><tbody>
<tr><td><strong>Day 1</strong></td><td>Save evidence, classify the claim, fix the source request</td><td>Nothing visible yet; this stage decides everything after</td></tr>
<tr><td><strong>Days 2 to 5</strong></td><td>File privacy and legal requests, click feedback</td><td>Acknowledgements, requests for identity proof, or silence</td></tr>
<tr><td><strong>Weeks 2 to 4</strong></td><td>Answer any request for more proof; re-test</td><td>Some responses change; others do not, or change on one phrasing only</td></tr>
<tr><td><strong>Weeks 4 to 8</strong></td><td>Appeal refusals; escalate where a regulator route exists</td><td>Decisions arrive, often partial; source fixes begin to show</td></tr>
<tr><td><strong>Month 3 onward</strong></td><td>Re-test on a schedule</td><td>Answers drift; new variants can appear</td></tr>
</tbody></table></div>

## Appeals, regulators and when to bring in a lawyer

A refused request is not the end of the road, but each step has its own cost. Some providers describe an appeal: Anthropic allows one by email when a request is denied, and xAI lets you appeal through its portal if a request is not acted on within a reasonable time. The appeal should add something: new evidence, a clearer identification, or a narrower request. Repeating the original does not help.

A complaint to a data protection authority is the next step where the law gives you that right. OpenAI's help page mentions a local supervisory authority. The EU text on rectification, [Article 16 of the GDPR as adopted](https://www.legislation.gov.uk/eur/2016/679/article/16/adopted), says you can obtain from the controller, without undue delay, the rectification of inaccurate personal data concerning you, and to have incomplete data completed. How that applies to a model's output is still being worked out.

Involve a lawyer when the false claim alleges a crime or professional failing, when it has caused a concrete loss, or when you intend to send a formal demand. Whether an AI-generated statement is actionable, and against whom, depends on your country. That is a question for counsel, not for this guide.

## Mistakes that waste a good request

Four patterns recur: filing before saving evidence, so you cannot prove what was said; one vague complaint instead of one claim per request; sending ID through a channel the policy does not name; and treating an auto-reply as a result, when it says only that your message arrived.

<aside class="callout">
<span class="callout-label">Warning</span>
<h4>Do not repeat the claim in public to fight it</h4>
<p>A public rebuttal that quotes the false statement gives it another page to be read from. State the correct fact instead, on a page you own, without repeating the accusation.</p>
</aside>

## What to do this week

File first at the assistant most likely to be read about you, in the lane that fits, with one quoted sentence and one checkable document. Then file at the others and log each filing.

<div class="box hot"><b>If this is happening to you now</b>
<p>If a false AI answer is affecting a client, a deal or your safety, we can help you sort the claim into the right lane, assemble the evidence and run the filings across providers. Start with a private audit and we will tell you which routes we think fit, and which will not work. See <a href="/audit">the private audit</a>.</p></div>

Reputation Partners handles this as part of monitoring what AI assistants say about a name, described on the [AI search reputation page](/ai-search-reputation).

## Questions people ask

### Does the thumbs-down button actually report a false statement?

It sends feedback on that response, and the providers describe it as a way to improve the product. None of the pages we read promise a review, a reply or a correction. Use it, but file the privacy or legal request as well, because that is the route with a documented handling process.

### Will OpenAI ask for ID when I ask to remove personal data?

Yes. OpenAI's help page says a government-issued ID or other acceptable proof is required. It also says it relies on the evidence you provide and cannot independently investigate disputed facts, so attach a document that shows what is true. Approval is not guaranteed, because each request is weighed against freedom of expression and the public interest.

### Can one request fix all AI assistants at once?

No. Each provider acts only on its own product, and OpenAI's page states that removal does not remove information from external websites or search engines. File separately with each assistant that repeats the claim, and fix the web page that feeds them.

### What if the provider says the information is accurate and public?

Then the request is probably weak on the merits, and OpenAI's page lists public interest and criminal allegations among its factors. Check whether the information is outdated or misleading rather than false, and whether the page is the real problem. Do not escalate a request that asks for an accurate public record to disappear.

### Can I report a false AI answer about someone else, such as a client or relative?

Anthropic and xAI each describe an authorized agent route that needs written permission or evidence of authorization. We could not confirm the same for every provider. Get the person's written consent first, and expect the provider to verify both of you.