How to report incorrect information to ChatGPT, Gemini and others

To report incorrect information to ChatGPT, click thumbs down under the answer and describe the error; for safety or legal problems, choose "Select an issue" and then "Safety or Legal concern."^[1] Google AI Overviews, Gemini, Perplexity, Copilot and Claude all have similar in-product buttons. None of them promises to correct a specific answer, so use them as a record, and fix the sources the engines read.

Below is every platform's current route as of October 2026, what each can and can't do, and when a formal channel applies instead.

01What kinds of reporting routes exist?

It helps to separate four types of route, because they have very different powers.

02How do you report incorrect information in ChatGPT?

In-product feedback. OpenAI's help center describes reporting a response by tapping thumbs down under a message, then "Select an issue," then "Safety or Legal concern," and following the prompts.^[1] For an ordinary factual error, thumbs down with a short written explanation is the available route.

OpenAI's older web form for model-behavior feedback has been deprecated; the page now points users to the thumbs-down button, API dashboard feedback, content-reporting forms and customer support.^[2]

Privacy requests. If ChatGPT states false information about an individual, such as a named executive, that person can submit a request at privacy.openai.com or to [email protected]. OpenAI's privacy policy says it will consider correction or removal requests "based on applicable law and the technical capabilities of our models."^[3] Its help article on removing personal data says requests come from the individual or a legal representative, require identity verification, and are assessed case by case.^[4]

This limitation has been challenged. In 2024, privacy group noyb filed a complaint with Austria's data protection authority arguing that ChatGPT produced false information about a person and that OpenAI said it could not correct it.^[5]

03How do you report a wrong Google AI Overview?

At the bottom of each AI Overview are thumbs-up and thumbs-down icons. Google's help page says to choose thumbs down if the overview was "unhelpful, inaccurate, biased, or otherwise problematic," then click Report a problem, pick a category, add details and submit. You can also use the three-dot menu at the top right of the overview.^[6]

For content that violates the law or your rights, Google's Report content for legal reasons process asks you to choose the product, the reason and the specific content, and confirms with a reference number.^[7]

04Can you fix a Google Knowledge Panel?

Often, yes, and it matters because Google's entity data feeds into what its search features say about you. If your organization has a Knowledge Panel, you can click Claim this knowledge panel at the bottom and verify through an official site or profile, such as Search Console or YouTube. Not all panels are claimable yet.^[8]

Once verified, owners and managers can suggest changes and manage users. Anyone can also click Feedback under a panel and describe what should change, with links to pages that verify the correct fact.^[8] For local businesses, Google points owners to Google Business Profile instead.

05How do you report a wrong answer in Gemini?

Under a Gemini response, click Good response or Bad response, optionally choose a reason and add details. For legal issues, click More and then Report legal issue, select the reason and follow the instructions.^[9] Google notes that the conversation is attached to your feedback.

06How do you report incorrect answers in Perplexity?

Perplexity's help center says: for incorrect responses, click the three-dot menu below the response and select Report. For other issues, email [email protected] with your device, app version and a shareable thread link.^[10]

07How do you report a wrong answer in Microsoft Copilot or Bing?

Copilot collects feedback through thumbs up and thumbs down on responses, with a field to explain what went wrong. Microsoft's documentation describes this for Copilot in Microsoft 365 apps and notes that work or school admins can turn feedback off.^[11]

For Bing search results, Microsoft offers a route to report concerns about specific URLs or information, and a Feedback link at the bottom of results pages for factual errors. Microsoft states that reporting "will not necessarily result in removal" and that Bing doesn't control what websites publish.^[12]

08How do you report a wrong answer in Claude?

Anthropic's help center says you can use the thumbs down button to flag an unhelpful response, or write to the feedback email address listed in that article.^[13] The same article explains that Claude can produce incorrect or misleading responses and should not be the sole source for high-stakes facts.

09What can each route actually do?

The table summarizes the routes and their limits. "Changes the source" is the column that matters most.

10How should a company submit feedback so it is useful?

Feedback is cheap, so do it, but do it well. A vague complaint is less useful to a reviewer than a sourced correction.

1. 1
   Capture the evidence

   Screenshot the answer with date, prompt, engine, model if shown, location and logged-in status.
2. 2
   State the error neutrally

   One sentence: what the answer says, what is correct. Avoid repeating damaging wording more than needed.
3. 3
   Link a primary source

   Point to your dated facts page, a regulator filing or an official announcement.
4. 4
   Use the right route

   Feedback button for ordinary errors; legal or privacy forms only when the case fits them.
5. 5
   Log the submission

   Record the route, date and reference number in your error ledger.
6. 6
   Fix the source and re-test

   Update the pages engines cite, then re-run the prompt weekly until it flips.

11Which route should you use for which problem?

Matching the problem to the route saves time and avoids requests that get declined.

A wrong product fact, price or date goes through the in-product feedback button on each engine where you see it, plus source fixes. None of the platforms reviewed here offers a dedicated correction form for ordinary company facts.

A false claim about a named person, such as an executive accused of something they did not do, fits two routes: the platform's legal reporting option and, for OpenAI, the individual's own privacy request. Involve legal counsel before filing, because the wording of a legal complaint matters.

Wrong company details in Google's panels go through Knowledge Panel feedback or a verified claim, and through Google Business Profile for local details. These routes change entity data that Google's other features can draw on, which makes them more durable than a thumbs down.

A third-party page that has already been corrected but still appears in Google with old content fits Google's Refresh Outdated Content tool, which is for pages you do not own that no longer exist or have changed significantly.^[14]

12Why isn't feedback enough on its own?

Because none of these buttons edits what the engine reads. AI answers are built from model memory and from the sources the engine retrieves. If a directory, an old press release or a forum thread still states the wrong fact, the next answer can repeat it, whatever you reported.

The full source-correction workflow is in how to fix wrong AI information. If the error is damaging and spreading, treat it as an incident; our guide to AI hallucinations about your brand covers triage and escalation.

For enterprises with many markets and engines, an operating partner can run this loop continuously. OwnlyReal measures answers across engines, fixes errors at the source and works through your approval chain; see how it works.