GPTBot
Model training. OpenAI's automatic crawler for content that may be used to improve and train generative AI foundation models.
OpenAI · Model training
AI crawler directory / GPTBot
GPTBot robots.txt reference.
OpenAI's automatic crawler for content that may be used to improve and train generative AI foundation models.
20observed domains
7disallowed
8partial policies
4allowed by matched rules
What GPTBot is documented to do
This summary is tied to a first-party operator page reviewed on Sep 20, 2026. It should be rechecked when the registry version changes.
- Operator
- OpenAI
- Documented purpose
- Model training
- Separate HTTP fetcher
- Documented as a fetcher or crawler
- Official documentation
- https://developers.openai.com/api/docs/bots
GPTBot vs OAI-SearchBot vs OAI-AdsBot vs ChatGPT-User
OpenAI documents these tokens for different purposes. A rule for one does not automatically control the others.
OAI-SearchBot
AI search. OpenAI's automatic search crawler, used to surface and link websites in ChatGPT search results.
OAI-AdsBot
Ad landing-page validation. OpenAI's fetcher for validating the safety and relevance of landing pages that advertisers submit for ChatGPT ads.
ChatGPT-User
User-requested fetch. OpenAI's user-initiated fetcher for certain ChatGPT, Custom GPT, and GPT Action requests.
Common full-site robots.txt block pattern
This example states a crawl preference for the named token. It does not authenticate requests, enforce blocking, or automatically control another token from the same operator.
User-agent: GPTBot
Disallow: /Check the official operator source before publishing a policy. User-requested fetchers and non-fetcher control tokens can have different behavior from automatic crawlers.
Latest observed pilot policies
These states are calculated from each domain's latest available robots.txt record using the versioned CrawlLedger parser. They are observations of published text, not proof of crawler behavior.
| Domain | Parsed state | Matched group | Last observed |
|---|---|---|---|
| anthropic.com | allowed | * | |
| apnews.com | disallowed | GPTBot | |
| apple.com | partial | * | |
| cloudflare.com | allowed | GPTBot | |
| commoncrawl.org | partial | * | |
| google.com | partial | *, Yandex | |
| medium.com | disallowed | Amazonbot, Applebot-Extended, Bytespider, ClaudeBot, FacebookBot, GoogleOther, GPTBot, meta-externalagent | |
| openai.com | partial | * | |
| oreilly.com | disallowed | GPTBot, anthropic-ai, cohere-ai, CCBot, AI2Bot, Amazonbot, Amazonbot-Video, Bytespider, meta-externalagent, Diffbot, omgili, TimpiBot, SeznamBot, Exabot, YandexBot, Sogou, 360Spider, YisouSpider, Baiduspider-Render/2.0, MJ12bot, DataForSeoBot, AliyunSecBot, ArchiveTeam, ArchiveTeam ArchiveBot, AwarioBot, ZoominfoBot, ali-implementer, Blueno, BIGO-baiguoyuan, WorksOgCrawler, OnPageBot, DoCoMo, SAMSUNG-SGH-E250, TA-Googlebot | |
| perplexity.ai | partial | * | |
| quora.com | disallowed | GPTBot | |
| reddit.com | disallowed | * | |
| rslcollective.org | allowed | * | |
| rslstandard.org | allowed | * | |
| stackoverflow.com | unverified | No matching group observed | |
| theguardian.com | partial | * | |
| usatoday.com | disallowed | GPTBot | |
| voxmedia.com | partial | * | |
| yahoo.com | disallowed | ADmantX, AlphaBot, anthropic-ai, AwarioRssBot, AwarioSmartBot, BLEXBot, Buzzbot, Bytespider, CCBot, ChatGPT-User, claritybot, Claude-Web, ClaudeBot, cohere-ai, Diffbot, FacebookBot, FriendlyCrawler, Google-Extended, GPTBot, huggingface, ImagesiftBot, img2dataset, magpie-crawler, Meltwater, Neevabot, news-please, NewsNow, Nutch, omgili, omgilibot, panscient.com, Perplexity-ai, PerplexityBot, PetalBot, PiplBot, scoop.it, Scrapy, Seekr, SentiBot, SeznamBot, TurnitinBot, YouBot, ZumBot | |
| ziffdavis.com | partial | * |
Interpret the state carefully
“Allowed” means a selected group was observed without a blocking rule for the tested policy shape. “Partial” means at least one non-empty disallow path was observed. “No matching group” is silence, not an affirmative grant. None of these states authenticate the requester or establish legal permission.