Skip to content

AI crawler directory

Every AI crawler, by name

49 AI crawlers, fetchers and agents: who runs each one, what it feeds, the exact User-agent token for robots.txt, and whether a hit can be verified. Updated .

Grouped by what the crawl is for, because that is the decision a site owner actually makes: block training, keep citations, or both.

Training crawlers

Build the corpus a model is trained on. Blocking one keeps future models from learning from your pages; it has no effect on whether an AI product can cite you today.

CrawlerOperatorFeedsrobots.txt tokenRobots.txt
GPTBotOpenAIChatGPTGPTBotHonours robots.txt
ClaudeBotAnthropicClaudeClaudeBotHonours robots.txt
GoogleOtherGoogleGoogle (other products)GoogleOtherHonours robots.txt
AmazonbotAmazonAlexaAmazonbotHonours robots.txt
Meta-ExternalAgentMetaMeta AIMeta-ExternalAgentHonours robots.txt
CCBotCommon CrawlCommon Crawl corpusCCBotHonours robots.txt
BytespiderByteDanceDoubao / TikTokBytespiderUndocumented
CohereCohereCoherecohere-aiUndocumented
DiffbotDiffbotDiffbot KGDiffbotUndocumented
PanguBotHuaweiHuawei PanGuPanguBotUndocumented
ImagesiftBotThe HiveImageSiftImagesiftBotHonours robots.txt
AI2BotAllen Institute for AIAllen InstituteAI2BotHonours robots.txt
OmgiliWebz.ioWebz.io corpusomgiliHonours robots.txt
YandexAdditionalYandexYandex AIYandexAdditionalHonours robots.txt
SBIntuitionsBotSB Intuitions (SoftBank)SB IntuitionsSBIntuitionsBotHonours robots.txt
ICC-CrawlerNICT (Japan)NICT corpusICC-CrawlerHonours robots.txt
VelenPublicWebCrawlerVelenVelen corpusVelenPublicWebCrawlerUndocumented
img2datasetOpen-source toolDataset builderimg2datasetMay ignore robots.txt
aiHitBotaiHitaiHit corpusaiHitBotUndocumented
CrawlspaceCrawlspaceCrawlspace corpusCrawlspaceUndocumented
Poseidon ResearchPoseidon ResearchPoseidon corpusPoseidon Research CrawlerUndocumented

User-triggered fetchers

Fetch a page because a person asked the assistant about it or pasted the link. Several operators state these fetchers act on the user's behalf and do not check robots.txt.

CrawlerOperatorFeedsrobots.txt tokenRobots.txt
ChatGPT-UserOpenAIChatGPT (live browse)ChatGPT-UserHonours robots.txt
Claude-UserAnthropicClaude (live browse)Claude-UserHonours robots.txt
Perplexity-UserPerplexityPerplexity (live browse)Perplexity-UserMay ignore robots.txt
Meta-ExternalFetcherMetaMeta AI (link fetch)Meta-ExternalFetcherMay ignore robots.txt
MistralAI-UserMistral AILe ChatMistralAI-UserHonours robots.txt
SemrushBot-OCOBSemrushSemrush AI corpusSemrushBot-OCOBHonours robots.txt

Agentic browsers

Browse and act on one person's instruction — comparing prices, filling forms, completing a task. Closer to a visit than a crawl, but they do not run analytics scripts.

CrawlerOperatorFeedsrobots.txt tokenRobots.txt
ChatGPT-AgentOpenAIChatGPT (agent)ChatGPT-AgentHonours robots.txt
GoogleAgent-MarinerGoogleProject Mariner (agent)GoogleAgent-MarinerUndocumented
NovaActAmazonAmazon Nova Act (agent)NovaActUndocumented
DevinCognitionDevin (agent)DevinUndocumented
FirecrawlAgentFirecrawlFirecrawl (agent scraping)FirecrawlAgentUndocumented

Robots.txt control tokens

Not a crawler you will see in logs. A token the operator reads in robots.txt to decide whether pages its ordinary crawler already fetched may be used for AI features.

CrawlerOperatorFeedsrobots.txt tokenRobots.txt
Google-ExtendedGoogleGemini / AI OverviewsGoogle-ExtendedHonours robots.txt
Applebot-ExtendedAppleApple IntelligenceApplebot-ExtendedHonours robots.txt
Webzio-ExtendedWebz.ioWebz.io corpusWebzio-ExtendedHonours robots.txt

How this list is maintained

Every entry corresponds to a signature in the classifier that runs on NumberHill's ingest endpoint, so the directory and the product cannot disagree about what a user agent is. A test fails the build if either side gains an entry the other lacks.

Robots.txt behaviour is recorded as the operator documents it, with a link where one exists. “Undocumented” means the operator has published nothing either way, not that the crawler misbehaves. Sample user-agent strings are representative of what appears in server logs; version numbers change.

To report a missing crawler or a correction, email [email protected] with the user-agent string and the operator's documentation page.

The whole list is also available as ai-crawlers.json with open CORS, and the generator's presets as a plain ai-robots.txt you can curl into a build step.

Common questions

What is the difference between an AI training crawler and an AI search crawler?

A training crawler (GPTBot, ClaudeBot, CCBot) collects pages to train future models; blocking it has no effect on whether a product cites you today. An AI search crawler (OAI-SearchBot, PerplexityBot, Claude-SearchBot) builds the index an answer engine retrieves from; blocking it removes your site from that product's citations and the referral traffic they send.

Does blocking Google-Extended hurt my Google rankings?

No. Google-Extended is a robots.txt control token, not a crawler. Disallowing it tells Google not to use pages Googlebot already fetched for Gemini training and grounding. Google Search indexing and AI Overviews follow the ordinary Googlebot rules and are unaffected.

Which AI crawlers ignore robots.txt?

By their operators' own documentation, user-triggered fetchers such as Perplexity-User and Meta-ExternalFetcher may bypass robots.txt because a person requested the specific page. Bytespider is widely reported by site operators to continue crawling after being disallowed. Where an operator publishes nothing, this directory says so rather than guessing.

How can I tell whether a hit claiming to be GPTBot is genuine?

OpenAI, Perplexity, Google, DuckDuckGo, Mistral and Kagi publish the IP ranges their crawlers use; a request from outside those ranges is spoofed. Amazon, Apple, Yandex and Huawei document reverse-DNS verification. Anthropic publishes no machine-readable list, so a ClaudeBot hit cannot be verified by address.

Should I block AI crawlers at all?

It depends on what the site is for. A publisher whose revenue is page views loses when models answer from its content without a click. A business that sells something usually benefits from being cited: an engine that cannot read a page cannot recommend it. Most commercial sites do best blocking training crawlers and allowing AI search and user-triggered fetchers.