New: monitor which AI crawlers actually visit your site with Sona Agent Analytics  |  See the platform →

Claude-User

Fetches a page live when a Claude user asks Claude to read it. Notably, Anthropic says it respects do-not-crawl signals.

Reference reviewed by Sona

OperatorAnthropic
PowersLive page fetches when Claude acts for a user
PurposeUser-triggered fetch
User-agent tokenClaude-User
Respects robots.txtYes

Claude-User fires when a person using Claude requests web access - asking Claude to read, summarize, or check a specific page. Anthropic describes it as supporting Claude users when they request web access through their queries.

What makes it worth a second look is its robots.txt posture. Anthropic states Claude-User respects robots.txt do-not-crawl signals. That puts it in a small minority: the equivalent fetchers at OpenAI, Perplexity, Meta, Google, and Amazon are all documented as possibly or generally ignoring robots.txt because a human initiated the request. Anthropic reaches the opposite conclusion from the same premise.

The practical consequence is that a Disallow here is likely to work as written, which is not something you can assume for user-triggered fetchers generally. It also means an over-broad wildcard Disallow genuinely does stop Claude from reading pages users ask about - a self-inflicted block that the other vendors' fetchers would have ignored.

Anthropic's older Claude-Web and anthropic-ai tokens covered similar access and are retired. Claude-User is the current token for on-demand reads, and it is the one to name if you want to allow human-initiated visits while disallowing ClaudeBot's training crawl.

How Claude-User behaves

  • Single-page, on-demand fetches tied to one conversation - not a sweep of your site.
  • Documented as respecting robots.txt do-not-crawl signals, unlike most user-triggered fetchers at other vendors.
  • Traffic tracks Claude usage patterns, so it appears in unpredictable bursts rather than on a crawl schedule.
  • Shares Anthropic's published IP feed with ClaudeBot and Claude-SearchBot.

Full user-agent string

Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; Claude-User/1.0; [email protected])

Allow Claude-User

A user asking Claude about your page gets an answer grounded in the live page - and unlike most fetchers, your robots.txt preference here is actually honored either way.

User-agent: Claude-User
Allow: /

Block Claude-User

You want no programmatic access from Claude at all, including human-initiated reads - and here a robots.txt Disallow is documented to be sufficient.

User-agent: Claude-User
Disallow: /

How to verify Claude-User

Published IP range feed

Anthropic's shared feed at claude.com/crawling/bots.json covers Claude-User along with its two siblings. Because Anthropic documents this bot as honoring robots.txt, unexpected Claude-User hits on disallowed paths are a reason to check the IP against the feed - genuine Anthropic traffic should be following your rules, so a request that isn't is more likely a forgery than a policy change.

Check an IP against this bot

Commonly confused with Claude-User

ChatGPT-User

Same role, opposite documented policy: OpenAI says robots.txt may not apply to ChatGPT-User, while Anthropic says Claude-User respects do-not-crawl signals.

ClaudeBot

Claude-User reads one page a human asked about; ClaudeBot crawls in bulk for training. Allowing the first while blocking the second is a common posture.

Claude-Code

Claude-Code covers fetches from the Claude Code CLI, a developer tool, rather than from Claude's chat interface.

Claude-User FAQs

Does Claude-User actually honor robots.txt?

Anthropic states that it respects robots.txt do-not-crawl signals. That is a stronger commitment than OpenAI, Perplexity, Google, Meta, or Amazon make for their equivalent user-triggered fetchers, all of which are documented as possibly ignoring robots.txt.

Will a wildcard Disallow block Claude from reading pages users ask about?

Yes, if no Claude-User group overrides it. Because this fetcher genuinely honors robots.txt, an over-broad wildcard rule really does stop Claude from reading your pages on request - an outcome most other vendors' fetchers would have ignored.

How is Claude-User different from Claude-Web?

Claude-Web is a retired token. Claude-User is the current identifier for on-demand fetches made when a Claude user requests web access, and it is the one your robots.txt should name.

Should I allow Claude-User if I block ClaudeBot?

For most sites, yes. Blocking ClaudeBot keeps you out of training, while allowing Claude-User means a person who asks Claude about your page gets an answer based on what the page actually says.

Can Claude-User read your page right now?

Test any URL and see exactly what AI crawlers receive.

Check my site