OpenAITrainingIP ranges published

What is GPTBot?

GPTBot is OpenAI's training crawler. It fetches pages on its own schedule to build datasets for training OpenAI's models. It is not the agent that fetches a page when someone asks ChatGPT about it (that is ChatGPT-User), and it is not the search indexer (that is OAI-SearchBot). A GPTBot hit is a crawl event, not a reader.

User agent strings

Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko); compatible; GPTBot/1.0; +https://openai.com/gptbot
Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko); compatible; GPTBot/1.1; +https://openai.com/gptbot
Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko); compatible; GPTBot/1.2; +https://openai.com/gptbot

agenttraffic matches on the token GPTBot.

What kind of agent it is

A training crawler collects pages to train models; blocking it costs no citations and no readers.

Verification

OpenAI publishes IP ranges for this agent (21 prefixes, list dated 2025-10-30, specific to GPTBot). agenttraffic checks every GPTBot request against them and reports verified, spoofed, or unchecked when no client IP is available.

GPTBot is impersonated less often than ChatGPT-User because there is nothing to gain from looking like a training crawler, but SEO tools and log analysers still emit it during tests. The spoofed column catches those.

robots.txt

The robots.txt token is GPTBot. Copy one of these blocks:

# Allow
User-agent: GPTBot
Allow: /

# Block
User-agent: GPTBot
Disallow: /

What it means in your logs

GPTBot traffic is the one category on the report that carries no reader behind it. A person did not ask for your page; a scheduler did. agenttraffic therefore hides training crawlers from the default terminal report and shows them with --show-training, so the AI crawler section stays focused on fetches that map to people.

That does not make GPTBot uninteresting. Its page list tells you which of your content OpenAI considers worth collecting, and a sudden increase usually precedes a model training cycle. Sites that publish a lot of long-form text sometimes see GPTBot as their single largest bot by request count.

Verification details

OpenAI publishes a JSON file of GPTBot IP ranges, separate from the ChatGPT-User and OAI-SearchBot lists. agenttraffic ships a dated snapshot and checks every GPTBot request against it. Because the list is small and specific, verification is reliable: a GPTBot UA from an IP outside the list is almost always a tool pretending, and it lands in the spoofed column rather than being counted as OpenAI.

Blocking GPTBot

This is the one OpenAI agent where blocking costs nothing visible. Disallowing GPTBot in robots.txt removes your pages from future training crawls and leaves search citations and live fetches untouched, because those use different agents. OpenAI states that GPTBot honours the token. If you want to opt out of training but stay reachable in ChatGPT answers, block GPTBot and allow OAI-SearchBot and ChatGPT-User.

Versions

The version number in the UA has moved from 1.0 through 1.2. agenttraffic matches on the GPTBot token, so every version is recognised without a registry change; the exact strings above are listed so you can grep for them directly.

Reading the page list

The pages GPTBot fetches are a rough proxy for what OpenAI's collection pipeline treats as text worth having. Long articles, documentation and reference pages dominate; app shells, login pages and paginated archives appear rarely. If a page you consider important never shows up, the usual causes are a robots.txt rule broader than intended, a noindex header (which GPTBot does not use, but which often travels with a disallow), or the page simply not being linked from anywhere GPTBot has already been. Fix the linking first; GPTBot discovers pages the same way search crawlers do.

See it in your logs

npx agenttraffic ./logs --days 7 --show-training

Point it at CloudFront, nginx or Apache logs. GPTBot appears under its own row with verified, spoofed and unchecked counts kept separate.

Related agents

FAQ

What is GPTBot?

GPTBot is OpenAI's training crawler. It fetches pages on its own schedule to build datasets for training OpenAI's models. It is not the agent that fetches a page when someone asks ChatGPT about it (that is ChatGPT-User), and it is not the search indexer (that is OAI-SearchBot). A GPTBot hit is a crawl event, not a reader.

How do I verify GPTBot traffic is real?

OpenAI publishes IP ranges for this agent (21 prefixes, list dated 2025-10-30, specific to GPTBot). agenttraffic checks every GPTBot request against them and reports verified, spoofed, or unchecked when no client IP is available.

How do I block GPTBot in robots.txt?

Add "User-agent: GPTBot" followed by "Disallow: /". OpenAI documents this token; whether it is honoured is up to the operator, and blocking is only enforced by their crawler, not by your server.

Does blocking GPTBot affect ChatGPT citations?

No. Citations in ChatGPT come from OAI-SearchBot's index and from ChatGPT-User live fetches. Blocking GPTBot only opts your pages out of future training datasets.

How often does GPTBot crawl?

There is no published schedule. On the reference deployment it arrives in bursts, often re-fetching the same popular pages, and it respects crawl-delay hints in robots.txt according to OpenAI's documentation.

Weekly: which agents read your site

Get told when the hosted weekly report exists. One email, no list.

Published 2026-08-28, updated 2026-08-28. Facts come from the agenttraffic registry.