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

Published: 12 October 2026| Version 1 | DOI: 10.17632/vv5v5x8h52.1
Contributor:
RanAI Man18

Description

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 are kept in the file with fetched_ok = no. Headline results (from the report): 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%) and OAI-SearchBot least (30 sites, 11.5%). 81 of 241 sites with a definite answer (34%) serve a valid llms.txt. Categories: B2B SaaS, News & media, Reference & education, E-commerce & marketplaces, Tech & developer, Social & community, Government & public sector, Search & portals, Video & streaming, Travel & local. Method: HTTPS fetch of /robots.txt and /llms.txt for each domain. robots.txt was parsed per RFC 9309 (groups merged, fallback to the * group, longest match wins, Allow wins ties, * and $ wildcards) and each user agent was evaluated against the path /. Limitations: a single October 2026 snapshot, homepage path only, and a hand-built list of well-known sites rather than a random sample of the web. Full report and method: https://airanked.ai/research/ai-crawler-access-report Source CSV: https://airanked.ai/research/ai-crawler-access-data.csv

Files

Steps to reproduce

For each domain, fetch https://DOMAIN/robots.txt and https://DOMAIN/llms.txt over HTTPS. Parse robots.txt per RFC 9309 (groups merged, fallback to the * group, longest match wins, Allow wins ties, * and $ wildcards) and evaluate each of the 15 user agents against the path /. Record allowed or blocked per agent, plus whether a valid llms.txt is served. Sites whose robots.txt could not be read are kept with fetched_ok = no. Full method: https://airanked.ai/research/ai-crawler-access-report

Categories

Artificial Intelligence, Information Retrieval, World Wide Web, Internet, Web Search Engine

Licence