Live report ยท updated 2026-08-27T13:19:55Z

The State of AI Crawler Permissions

How the web's most-visited sites treat AI crawlers, measured daily by BotWitness from robots.txt / llms.txt.

21% of 640257 tracked sites block GPTBot from their whole site.

Who blocks which AI crawler

AI crawlerBlocking at rootShare
GPTBot132855 of 640257 (21%)
ClaudeBot126543 of 640257 (20%)
Amazonbot125776 of 640257 (20%)
Google-Extended122214 of 640257 (19%)
Meta-ExternalAgent119676 of 640257 (19%)
Applebot-Extended119400 of 640257 (19%)
CCBot96724 of 640257 (15%)
Bytespider93066 of 640257 (14%)
anthropic-ai68149 of 640257 (11%)
ChatGPT-User67762 of 640257 (11%)
PerplexityBot64906 of 640257 (10%)
cohere-ai64816 of 640257 (10%)
ImagesiftBot64471 of 640257 (10%)
Omgilibot64340 of 640257 (10%)
Diffbot64249 of 640257 (10%)
FacebookBot63871 of 640257 (10%)
YouBot62628 of 640257 (10%)
TimpiBot61528 of 640257 (10%)
OAI-SearchBot59874 of 640257 (9%)
Claude-SearchBot57288 of 640257 (9%)
Perplexity-User56647 of 640257 (9%)

Momentum: the last 30 days

58642 tracked domains materially changed their robots.txt AI directives in the last 30 days.

AI crawlerNewly blockedNewly unblocked
GPTBot39562534
Bytespider40762292
ClaudeBot38922427
CCBot38742382
Amazonbot38382346
Google-Extended36842375
Applebot-Extended37592207
Meta-ExternalAgent39281730

The permission stack beyond robots.txt

FileSites publishing itWhat it is
llms.txt236185AI-readable site guide
/.well-known/ai.txt115451AI usage policy file
RSL license230machine-readable license terms (robots.txt License: directive)

RSL count is sites whose declared license document currently resolves; declaration history is in the archive from the day we first saw it.

Beyond domains: the AI ecosystem's own record

Also captured daily, through the same timestamped pipeline: EU GPAI training-data disclosures (4 surfaces); Web Bot Auth key directories (12 operators); MCP registry entries (13852 servers); live MCP tool manifests (777 probed); Hugging Face model license/gating state (22 models).

Data licensing

Researchers, journalists and quantitative teams can license the underlying dataset: the longitudinal permission record across the tracked web, with per-capture hashes and timestamps. Citation of this page is free with a link.

[email protected]

Recently changed AI directives

Coverage: 1014574 domains, 3735087 snapshots, 11192436 checks; archive since 2026-07-09T19:41:24Z. Each capture is SHA-256 hashed and RFC-3161 timestamped. Point-in-time lookups and certified records via the API.