airanked/ai-crawler-access-2026

AI Crawler Access Study: robots.txt and llms.txt of 261 Major Websites (Oct 2026)

This dataset records whether 285 major websites allow or block 15 user agents in robots.txt (14 AI crawlers plus Googlebot as a control), and whether each site serves a valid llms.txt. Data was collected in October 2026 by AI Ranked. One row per domain. 261 of the 285 robots.txt files could be read; the other 24 stay in the file with fetched_ok = no.

Full report and method: Which AI crawlers do top websites block? A study of 261 sites

Headline results

• Of the 261 readable sites, 91 (35%) block at least one AI training crawler and 56 (21%) block at least one AI search crawler.
• CCBot and Bytespider are blocked most (75 sites each, 28.7%). OAI-SearchBot is blocked least (30 sites, 11.5%).
• 31 of the 59 sites that block GPTBot still allow OAI-SearchBot.
• 81 of the 241 sites with a definite answer (34%) serve a valid llms.txt.

Columns

Column	Meaning
domain	The website tested (one row per domain, 285 rows)
category	One of 10 groups: B2B SaaS (102), News & media (53), Reference & education (25), E-commerce & marketplaces (23), Tech & developer (22), Social & community (17), Government & public sector (14), Search & portals (11), Video & streaming (10), Travel & local (8)
fetched_ok	yes if robots.txt could be read (or the site has none), no if the fetch failed (24 rows)
robots_status	ok, no robots.txt (404), or failed (...) with the HTTP status or error
GPTBot, ClaudeBot, Google-Extended, Applebot-Extended, meta-externalagent, CCBot, Bytespider, Amazonbot	Training crawlers: allowed or blocked at the path /; empty when fetched_ok = no
OAI-SearchBot, Claude-SearchBot, PerplexityBot	AI search crawlers: allowed or blocked at /
ChatGPT-User, Claude-User, Perplexity-User	User-requested fetchers: allowed or blocked at /
Googlebot	Control: allowed or blocked at /
llms_txt	yes if /llms.txt returned HTTP 200, a text/plain or text/markdown content type and opened with a Markdown H1; no on 200 without that, 404 or 410; unknown on 403, 429, other errors or timeouts (44 rows)

Method

A Python script requested https://<domain>/robots.txt and https://<domain>/llms.txt with a 10-second timeout, redirects followed, and one retry (including the www. host). robots.txt was parsed per RFC 9309: consecutive User-agent lines form one group, groups naming the same agent are merged, an agent with no named group falls back to *, the longest matching rule wins, Allow wins a tie, and * and $ work as wildcards. Each agent was evaluated against the path /. A 404 or 410 for robots.txt counts as permission to crawl.

Limitations

A single October 2026 snapshot, homepage path only, one file per site as served to a self-identified research fetcher, and a hand-built list of well-known sites rather than a random sample of the web. robots.txt is a request, not enforcement: CDN or firewall blocks do not show up here.

License and citation

CC BY 4.0. Reuse freely with a link to the report.

AI Ranked Editorial (2026). AI Crawler Access Study. https://airanked.ai/research/ai-crawler-access-report