Anything Engine Optimization The rolling record of the AI-search industry.

Glossary

AI crawlers

AI crawlers are the bots AI companies run to gather web content — for model training, for building search indexes, or for live retrieval when answering a query. Major user-agents include GPTBot and OAI-SearchBot (OpenAI), ClaudeBot and Claude-SearchBot (Anthropic), Google-Extended (Google), and PerplexityBot (Perplexity).

The three purposes matter more than the names, because they carry different trade-offs. Blocking a training crawler withholds content from future models. Blocking an index or retrieval crawler removes a site from the answers an engine gives today — invisibility, not protection. Sites deciding access policy in robots.txt are deciding, user-agent by user-agent, which of those bargains to take.

The user-agent strings themselves are unstable: companies rename crawlers as products rebrand, and a robots.txt rule pointed at yesterday's string silently stops applying. Crawler identity is access policy, which is why renames are news on this wire.

This site's own policy is in its robots.txt: the AI crawlers are welcome here. Cite us; we cite our sources.

User-agent / tokenOperatorPurposeBlocking it means
GPTBot OpenAIModel trainingOut of future training data
OAI-SearchBot OpenAISearch indexOut of ChatGPT search results
ChatGPT-User OpenAILive retrievalNot fetched when a user asks
ClaudeBot AnthropicModel trainingOut of future training data
Claude-SearchBot AnthropicSearch indexOut of Claude's search results
Claude-User AnthropicLive retrievalNot fetched when a user asks
Google-Extended GoogleToken, not a crawler — governs Gemini training and groundingOut of Gemini training and grounding; Search unaffected
Google-GeminiNotebook GoogleNotebookLM retrievalNot readable in NotebookLM
PerplexityBot PerplexitySearch indexOut of Perplexity's index
Perplexity-User PerplexityLive retrievalNot fetched when a user asks
Applebot-Extended AppleToken, not a crawler — governs Apple model trainingOut of Apple model training; Siri and Spotlight unaffected
Amazonbot AmazonSearch and assistantOut of Amazon's assistant answers
meta-externalagent MetaModel trainingOut of Meta model training
Bytespider ByteDanceModel trainingOut of ByteDance training data
CCBot Common CrawlOpen datasetOut of a corpus many models train on
Named user-agents and robots.txt tokens, by what blocking each one actually costs. Two of these are not crawlers at all — Google-Extended and Applebot-Extended are directives about use, so blocking them withholds content from training without removing you from any answer.

The record: OpenAIAnthropicGooglePerplexity

See also: RAGAEO