AI companies
CompanyTo train its modelsTo answer people and search
- OpenAIChatGPTTo train its modelsKept out
GPTBotline 7To answer people and searchLet inOAI-SearchBotChatGPT-User - AnthropicClaudeTo train its modelsPartly
ClaudeBotline 4, kept outanthropic-ailet inClaude-Weblet inTo answer people and searchLet inClaude-SearchBotClaude-User - GoogleGeminiTo train its modelsKept out
Google-Extendedline 8To answer people and searchLet inGoogle-CloudVertexBot - PerplexityTo train its modelsNone namedTo answer people and searchLet in
PerplexityBotPerplexity-User - MetaMeta AITo train its modelsPartly
meta-externalagentline 10, kept outFacebookBotlet inTo answer people and searchLet inmeta-externalfetchermeta-webindexer - AppleApple IntelligenceTo train its modelsKept out
Applebot-Extendedline 9To answer people and searchNone named - AmazonTo train its modelsKept out
Amazonbotline 12To answer people and searchLet inAmzn-SearchBotAmzn-User - Common Crawlopen data for trainingTo train its modelsKept out
CCBotline 11To answer people and searchNone named - ByteDanceTo train its modelsPartly
Bytespiderline 6, kept outTikTokSpiderlet inTo answer people and searchNone named - MistralLe ChatTo train its modelsNone namedTo answer people and searchLet in
MistralAI-User
1 of the 27 other AI crawlers we know is kept out: AI2Bot.
What it tells AI crawlers
It gives no Content Signals and no “noai” tag.
Like 30% of the 408 popular sites whose robots.txt we’ve read since 29 Oct 2026, it keeps at least one AI training crawler out. Of the 517 popular sites we’ve checked since 29 Oct 2026, 9% have a page drawn by scripts, 2% send a Markdown copy, 9% have an llms.txt and 3% offer a file for AI agents. How many websites block AI crawlers?
We read the site once, as our named bot: its robots.txt, read as crawlers read it, its home page and the files AI tools ask sites for. What we couldn’t read says so. robots.txt asks and doesn’t enforce: a firewall can treat AI crawlers differently from us.