What is ChatGPT-User?

How it identifies itself

Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko); compatible; ChatGPT-User/1.0; +https://openai.com/bot
Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; ChatGPT-User/1.0; +https://openai.com/bot

agenttraffic recognises it by the name ChatGPT-User.

What it does

A live fetch agent requests a page because a person just asked an AI assistant about it. Each hit is a real reader, arriving through the assistant instead of a browser. More on live fetch agents.

Is it really them?

OpenAI publishes the IP addresses this agent uses (204 address ranges, list dated 2026-08-14, specific to ChatGPT-User). agenttraffic checks every ChatGPT-User request against them and reports it as verified, spoofed, or unchecked when the log has no client IP.

• https://openai.com/chatgpt-user.json: 204 address ranges, published 2026-08-14, specific to this agent. Copy shipped with agenttraffic: 2026-08-28.

ChatGPT-User is one of the most impersonated names in server logs. Scanners and scrapers borrow it hoping to pass as assistant traffic, so the verified count and the spoofed count on its row are usually both non-zero. Only the verified number is a floor (at least this many) of real ChatGPT fetches.

How to block it

In robots.txt it goes by ChatGPT-User. Copy one of these:

# Allow
User-agent: ChatGPT-User
Allow: /

# Block
User-agent: ChatGPT-User
Disallow: /

What it means in your logs

A ChatGPT-User hit is the closest thing to a pageview an assistant produces. Someone was in a ChatGPT conversation, the model decided it needed your page to answer, and OpenAI fetched it. The person then read a summary, and possibly a link, inside ChatGPT. Your analytics script never ran, because the fetcher does not execute JavaScript, so this reader is invisible everywhere except the log line.

That is why agenttraffic files ChatGPT-User under live fetch and lists the pages it fetched. The list is a map of what people are asking ChatGPT about on your site. A page that shows up here with no matching referral traffic is being consumed inside the assistant, which is worth knowing before deciding whether the page is "underperforming".

The same user agent string is used by ChatGPT agent mode, where the model browses on the person's behalf across several pages. Those sessions look like a short burst of fetches from one verified IP.

Where the verification data comes from

OpenAI publishes a dedicated list of IP addresses for ChatGPT-User, separate from the GPTBot and OAI-SearchBot lists. agenttraffic ships a snapshot of that list, records when OpenAI last changed it, and marks each request as verified (IP inside the list), spoofed (IP outside it), or unchecked (no client IP in the log). npx agenttraffic update refreshes the snapshot. The published ranges are the only reliable way to tell a real fetch from a scraper wearing the same string; the user agent string alone proves nothing.

Should you block it?

Blocking ChatGPT-User does not remove you from any index or training set. It only stops ChatGPT from reading your page when a user asks about it, and the user typically sees a message that the page could not be fetched. For most sites that is a loss with no offsetting gain. If a page must not be summarised (paywalled, legal, embargoed), block the specific path rather than the whole site.

Telling the OpenAI agents apart

OpenAI runs three public crawlers plus an ads bot. GPTBot collects training data. OAI-SearchBot builds the index behind ChatGPT search and its citations. ChatGPT-User fetches live for a conversation. Each one has its own robots.txt name, its own IP list, and a different reason to allow or block it, so a robots.txt rule written for one should not be copied to the others.

Related agents

• OAI-SearchBot (OpenAI, ai search)
• GPTBot (OpenAI, training)

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