Understand the difference between GPTBot, OAI-SearchBot, and ChatGPT-User, how IPBot verifies them, and how site owners should treat each one.
OpenAI publishes more than one crawler family, and that is the main source of operator confusion. A site owner who only sees 20.x.x.x plus an OpenAI-looking range still needs to ask what kind of request it is: training crawl, search retrieval, or a user-triggered fetch.
IPBot uses the source inventory below to decide whether a request is merely known or fully verified.
Swipe sideways to review every verification column.
| Crawler | Intent | Verification | CIDRs | Freshness |
|---|---|---|---|---|
| ChatGPT-User
Official source
| user triggered | User-agent + official IP range | 247
IPv4 247 / IPv6 0 | 2026-05-21 |
| GPTBot
Official source
| training | User-agent + official IP range | 21
IPv4 21 / IPv6 0 | 2025-10-30 |
| OAI-SearchBot
Official source
| search | User-agent + official IP range | 35
IPv4 35 / IPv6 0 | 2026-01-02 |
Separate these three families. GPTBot is the training crawler, OAI-SearchBot is the search crawler, and ChatGPT-User is user-triggered fetch traffic. They should not share the same allow or block policy by default.
Use explicit crawler tokens and document which OpenAI family each rule targets. Blocking GPTBot does not mean you want to block search or user-triggered retrieval.
Keep OpenAI policies granular. Allowing or challenging one family should not automatically apply to the other two, especially if your product depends on search visibility or shared-link fetches.
This should be governed separately from GPTBot because it represents user-triggered fetches rather than autonomous training crawls.
Useful as its own ruleset when you want to preserve shared-link or user retrieval behavior without allowing broader crawling.
Use the `GPTBot` token in `robots.txt` when deciding whether to allow OpenAI training crawl access.
Do not rely on IP range alone; pair the IP with the expected user-agent token before treating it as verified.
Treat separately from GPTBot if your concern is search visibility rather than model-training access.
Pair IP range membership with the `OAI-SearchBot` token before marking it verified.
GPTBot is the crawler family associated with training crawl access. It should be evaluated separately from search or user-triggered fetch traffic.
Official IP range membership alone is not enough for OpenAI-family traffic. IPBot marks these requests as verified only when the supplied user-agent matches the expected crawler token for the known OpenAI range.