How Do I Report False Information About My Company in Claude?
To report false information about your company in Claude, use the thumbs-down button on the response and describe the error. Give the correct fact and a public source that proves it. Anthropic's help centre describes the thumbs-down button as the way to flag responses that were wrong or unhelpful. If the error concerns personal information about an individual, use Anthropic's Privacy Center. For serious false allegations, take legal advice. Reporting is only part of the fix. Check whether the answer came from a web search or from what the model learned in training. Correct the sources behind search-based answers. Make sure Anthropic's search and user crawlers can read your site, since Anthropic says blocking them can reduce your site's visibility and accuracy in search results.
How do you report the response?
- Open the conversation with the false statement.
- Click thumbs down on the response.
- Explain the error in the feedback box: quote the false statement, give the correct fact, and include a link to a public source such as your website or an official register.
- Keep a copy of the conversation and your feedback, with the date.
A clear example: "The response says [company] was acquired by [other company] in 2024. It was not; [company] is independent, as stated at [link]."
Did the error come from search or from training?
It matters, because the fix is different:
- From web search: the response shows sources or links. The error usually traces back to a page that says the wrong thing, or to a misreading of a page. Fix or correct that page.
- From training: no sources are shown, and the answer reflects what the model learned before its training cutoff. Feedback helps Anthropic improve its models, but the answer may persist until a newer model is released. Clear, consistent sources across the web are what future models learn from.
Where Claude shows a Web search option under the "+" button, ask the same question with it on and off. In the new Claude experience there is no toggle and Claude searches when it judges it useful, so ask it to answer from what it already knows, without searching the web, for the memory answer; sources and links under a response show that search was used. If the searched answer is correct and the unsearched one is wrong, the problem is in model memory. Consistent sources give later models the best chance of learning the right fact, but you cannot control when or whether that happens. If both are wrong, look at the sources the searched answer cites. Why Claude can give vague or incorrect answers about a website is covered in why Claude gives vague or incorrect answers about your website.
Which crawler settings affect accuracy?
Anthropic describes three agents:
| Agent | What it does | Effect of blocking it |
|---|---|---|
| ClaudeBot | Collects web content that may be used to train models | Your future content is excluded from training data |
| Claude-SearchBot | Indexes content for search results | Anthropic says this may reduce your visibility and accuracy in search results |
| Claude-User | Fetches pages when a user's question needs them | Claude cannot retrieve your pages to answer questions about you |
Many businesses block ClaudeBot for training reasons but leave Claude-SearchBot and Claude-User allowed, so answers can draw on their current pages. If your firewall or bot protection challenges automated traffic, check that it lets these agents through as well.
What should you fix at the source?
- Your own pages: state the correct facts plainly in text, near the top of the relevant page.
- Third-party pages that carry the error: directories, review sites, news articles and comparison pages. Request corrections with evidence.
- Your profiles and listings: keep them consistent with your website.
- Old pages and files on your site that still carry outdated facts: update, redirect or remove them.
Getting your website mentioned in Claude's answers at all is covered in how to get your website mentioned in Claude answers.
When should you use privacy or legal routes?
- Personal information about an individual, such as a founder's home address or a false statement about an employee: Anthropic's privacy policy describes rights for individuals, including correction, depending on where they live. Requests go through its Privacy Center.
- Defamatory claims, such as invented lawsuits, fraud or crimes: document the responses, collect evidence, and take legal advice before escalating.
- Your content used without permission: raise it through Anthropic's published channels for intellectual property concerns.
These rights belong to individuals and rights holders, not to companies correcting ordinary business facts. For business facts, fixing sources is the main route.
How do you check whether the fix worked?
Ask the same questions in new conversations, with web search on and off, a few weeks after fixing sources. Record the answers and any sources. Search-based answers usually improve once corrected pages are indexed. Answers from model memory can take much longer. Track both separately, so you know which kind of error remains.
What might an illustrative case look like?
Consider an illustrative analytics company that Claude described as "acquired by a larger competitor in 2024". The company had in fact only signed a partnership. With web search on, Claude's answer cited a news headline that had misreported the partnership. With search off, it repeated the acquisition claim from memory. The company gave thumbs-down feedback with a link to its announcement and asked the publisher to correct the headline, which it did. It also added "independent and privately owned" to its about page. It had been blocking Claude's search and user crawlers along with its training crawler, so it unblocked the first two. Searched answers were correct within weeks. The unsearched answer changed after a later model update.
What do people ask about reporting errors in Claude?
Will Anthropic reply to my feedback?
Generally not individually. Feedback is used to improve responses and models.
Should we block ClaudeBot?
That is a policy choice about training. Blocking the search and user agents is different, and can make answers about you less accurate.
How long until Claude stops repeating the error?
Search-based answers can change within weeks of sources being fixed. Answers from training can persist until a newer model is released.
Published by The Kovmere Team · updated Sep 29, 2026 · More resources