How does Claude's web search cite sources?

When Claude's web_search tool returns results, the response text carries a citations array. Each citation records the source URL, page title, the exact quoted text (up to 150 characters), and a marker for where in the response it was used. Citations stay on by default whenever web search is active and can't be turned off.

Can site owners block Claude's crawlers?

Yes, through robots.txt. ClaudeBot (training data collection), Claude-User (real-time fetches triggered by a user's request in a Claude conversation), and Claude-SearchBot (search-result indexing) can each be controlled independently. Blocking Claude-SearchBot, for example, can reduce a site's visibility in Claude's search results without affecting the other two bots.

What's the difference between ClaudeBot and Claude-User?

ClaudeBot periodically crawls the web to collect training data. Claude-User is a separate agent that fetches a specific page in real time only when a user references it during a Claude conversation, so its traffic scales with user requests rather than running on a schedule.

Which Sites Does Claude Cite?

Claude marks its sources through a citations array attached to web search responses — each entry lists a source's URL, title, and the exact text quoted — while site owners can independently control ClaudeBot, Claude-User, and Claude-SearchBot through robots.txt.

Updated 2026-07-07

Frequently asked questions

Which Sites Does Claude Cite?: Claude marks its sources through a citations array attached to web search responses — each entry lists a source's URL, title, and the exact text quoted — while site owners can independently control ClaudeBot, Claude-User, and Claude-SearchBot through robots.txt.
How does Claude's web search cite sources?: When Claude's web_search tool returns results, the response text carries a citations array. Each citation records the source URL, page title, the exact quoted text (up to 150 characters), and a marker for where in the response it was used. Citations stay on by default whenever web search is active and can't be turned off.
Can site owners block Claude's crawlers?: Yes, through robots.txt. ClaudeBot (training data collection), Claude-User (real-time fetches triggered by a user's request in a Claude conversation), and Claude-SearchBot (search-result indexing) can each be controlled independently. Blocking Claude-SearchBot, for example, can reduce a site's visibility in Claude's search results without affecting the other two bots.
What's the difference between ClaudeBot and Claude-User?: ClaudeBot periodically crawls the web to collect training data. Claude-User is a separate agent that fetches a specific page in real time only when a user references it during a Claude conversation, so its traffic scales with user requests rather than running on a schedule.