What is GPTBot?
GPTBot is OpenAI's web crawler used to gather content for training and improving its foundation models. It is distinct from OpenAI's answer/retrieval crawlers (OAI-SearchBot and ChatGPT-User), which fetch pages to build cited answers rather than to train — a distinction that matters enormously for AI-search access control. Blocking GPTBot in robots.txt is a low-cost intellectual-property and brand-control decision with essentially no effect on your AI-search visibility, because a model does not need to have trained on your site to cite it live through search. The common mistake is assuming that blocking GPTBot removes you from ChatGPT answers — it does not; that decision belongs to OAI-SearchBot and ChatGPT-User. GPTBot respects robots.txt, so opting out of training is as simple as a Disallow directive for its user-agent, and it can be applied without touching your answer-surface eligibility.
Key statistics
- GPTBot trains models; OAI-SearchBot and ChatGPT-User fetch pages to build cited answers — three separate user-agents, three separate decisions.
- Blocking GPTBot has near-zero cost to AI-search citation because live citation does not require prior training.
- GPTBot respects robots.txt, so training opt-out is a single Disallow directive that leaves answer-surface eligibility intact.