AI TrainerAnthropic
Last verified: 2026-09-11 · maintained by Unsourced
ClaudeBot is Anthropic's crawler that collects training data for Claude. It respects robots.txt. Blocking it removes your content from what Claude learns, which can reduce future mentions.
ClaudeBot gathers web text that contributes to training Anthropic's Claude models, so its interest is durable knowledge rather than answering any single query. It identifies clearly and follows a ClaudeBot rule in robots.txt, giving you a clean on/off switch. Disallowing it keeps your material out of future Claude training runs, which over time can lower how readily Claude recalls or names your site.
ClaudeBot can be checked two independent ways, and a real request passes both. Its source IP has to fall inside the 26 ranges Anthropic publishes, and it has to forward-confirm by reverse DNS: the IP resolves to anthropic.com, and that hostname resolves back to the same IP. Anything carrying the ClaudeBot user-agent that fails either test is an impostor, whatever the header says.
Recommended: keep. ClaudeBot gathers training text for Claude, so the trade is long-term recall rather than referral clicks: cut it off and your material stops feeding future Claude training, which over time erodes how readily Claude names you.
If you do choose to act in robots.txt (which crawlers honour but don't enforce):
# ClaudeBot: recommended to ALLOW — blocking can cost you AI visibility User-agent: ClaudeBot Disallow:
Will blocking ClaudeBot remove me from Claude's answers?
Not from Claude's live web results, which it pulls separately. Blocking ClaudeBot keeps your content out of Anthropic's model training, which over time can lower how readily Claude recalls or names you.
Does ClaudeBot publish IP ranges?
Yes. As of 2026 Anthropic publishes a machine-readable IP-range feed for its crawlers at claude.com/crawling/bots.json, so a ClaudeBot request can be checked two independent ways: the source IP must fall inside a published range, and it should reverse-resolve to anthropic.com and forward-confirm back. A request that matches neither is wearing the name, not using it.
Does ClaudeBot obey robots.txt?
Yes. A User-agent: ClaudeBot disallow is respected, giving you a clean on/off switch for whether your material feeds future Claude training.
Unsourced checks each crawler against published ranges andreverse DNS, and shows where AI search cites you instead of a competitor.
Check your site free →10-day free trial · no card required · cancel anytime