---
title: "GPTBot"
description: "Ewelina Podrez-Siama · bot profiles GPTBot OpenAI’s training crawler: user-agent, IP addresses and robots.txt GPTBot is OpenAI’s crawler for model training. It collects pages that may be used to train OpenAI’s generative AI foundation models, and it honors robots.txt. Below you’ll find the user-agent GPTBot actually sends, OpenAI’s IP list, ready-made robots.txt rules and what"
date: 2026-11-12
author: "Ewelina Podrez-Siama"
url: "https://podrez.pl/en/bots/gptbot/"
---

  ![Paper collage: a bot directory with crawler icons, a spider, a magnifying glass and servers](https://podrez.pl/wp-content/uploads/2026/08/co-to-za-bot.jpg "GPTBot")Ewelina Podrez-Siama · bot profiles

GPTBot OpenAI’s training crawler: user-agent, IP addresses and robots.txt
=========================================================================

GPTBot is OpenAI’s crawler for model training. It collects pages that may be used to train OpenAI’s generative AI foundation models, and it honors robots.txt.

Below you’ll find the user-agent GPTBot actually sends, OpenAI’s IP list, ready-made robots.txt rules and what GPTBot does on podrez.pl. I checked the facts against OpenAI’s documentation and the sources below on November 12, 2026.

[Block or allow GPTBot ](#robots)[All bots in my list](https://podrez.pl/en/bots/)

 

 

[At a glance](#at-a-glance)[User-agent](#user-agent)[robots.txt](#robots)[IP check](#verification)[JavaScript](#javascript)[My logs](#logs)

**278**confirmed GPTBot requests
Oct 7 – Nov 11, 2026



**8**requests a day now,
40 in Feb–Oct



0robots.txt requests
under its own name





 

At a glance

GPTBot at a glance
------------------

QuestionAnswerOperatorOpenAIPurposeCrawls content that may be used to train OpenAI’s generative AI foundation models. Disallowing GPTBot tells OpenAI not to use your content for that training.robots.txt token`GPTBot`Without a GPTBot groupThe `User-agent: *` group applies. On October 30, 2026, OpenAI said it plans to support the shared training signal `__usage_ai-training` in the coming months.Honors robots.txtYes, according to OpenAI, which doesn’t mention `Crawl-delay`.IP addresses`openai.com/gptbot.json`: 18 IPv4 prefixes, file dated September 22, 2026.Reverse DNSNot documented. Use the IP list.Runs JavaScriptNot documented. A 2024 measurement found that OpenAI’s crawlers don’t render JavaScript.Other OpenAI bots`OAI-SearchBot` (ChatGPT search), `ChatGPT-User` (user-triggered: fetches a page someone asks about) and `OAI-AdsBot` (checks ad landing pages). OpenAI says each setting is independent.In my listAI: training ([see the entry in “What bot is this?”](https://podrez.pl/en/bots/#gptbot))

 

User-agent

The GPTBot user-agent
---------------------

OpenAI documents an example string and warns that the version number may change. Confirmed GPTBot requests on my site used a slightly different format:

RequestsUser-agent string278`Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; GPTBot/1.4; +https://openai.com/gptbot)`Example in OpenAI’s docs`Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko); compatible; GPTBot/1.4; +https://openai.com/gptbot`

Number: requests from OpenAI’s published addresses to podrez.pl, Oct 7 – Nov 11, 2026.

Note the parentheses. In confirmed requests, `compatible; GPTBot/1.4; +https://openai.com/gptbot` sits inside the parentheses after `like Gecko`; in OpenAI’s example the parenthesis closes right after `like Gecko`. All 63 requests that used the documented format came from outside OpenAI’s list, 57 of them from Google Cloud.

Treat this as a hint, not proof: OpenAI can change the string at any time, and the IP address is what decides. When OpenAI fetches robots.txt, it may add a `robots.txt` marker to the user-agent. On my site those requests came under the OAI-SearchBot name – more on that below.

 

robots.txt

How to block or allow GPTBot in robots.txt
------------------------------------------

GPTBot follows the robots.txt group that names it (a group is a block of rules that starts with `User-agent`). OpenAI’s tokens are independent, so you can switch off training and keep ChatGPT search.

### Variant 1 · No training

*ChatGPT search stays*

I keep OpenAI’s training crawler out but let ChatGPT search in.

 ```
User-agent: GPTBot
Disallow: /

User-agent: OAI-SearchBot
Allow: /
```

OpenAI recommends allowing OAI-SearchBot if you want your site to appear in ChatGPT search answers.

 

 

### Variant 2 · Training, with limits

*GPTBot stays in*

I let GPTBot in, except for one directory.

 ```
User-agent: GPTBot
Disallow: /private/
Allow: /
```

A GPTBot group replaces your `User-agent: *` rules for this bot, so copy over every Disallow line you still need.

 

 



OpenAI says that if your site allows both GPTBot and OAI-SearchBot, it may use the results of a single crawl for both. My logs fit: between October 7 and November 11, 2026, GPTBot never requested robots.txt under its own name. The 115 robots.txt requests from addresses on the GPTBot list carried the OAI-SearchBot name and the robots.txt marker, and those addresses are on the OAI-SearchBot list too.

On October 30, 2026, OpenAI announced that it plans to support the shared training signal `__usage_ai-training` in the coming months. Its existing robots.txt controls stay in place. I explain the token in [a separate article](https://podrez.pl/en/usage-ai-training/).

Check your file

Not sure what your robots.txt tells GPTBot? Enter your domain or paste the file into [Can AI bots read your site?](https://podrez.pl/en/robots-txt-checker/), my free robots.txt checker. It shows whether GPTBot may crawl your pages and which line decides it.



 

IP check

Is it really GPTBot? How to check the IP address
------------------------------------------------

OpenAI publishes GPTBot’s addresses at `openai.com/gptbot.json`. On November 12, 2026, the file, dated September 22, 2026, listed 18 IPv4 prefixes. OAI-SearchBot and ChatGPT-User have lists of their own; six of GPTBot’s 18 prefixes are also on the OAI-SearchBot list. OpenAI documents no reverse DNS name for its bots.

Quick glossary: the user-agent is the name a bot gives itself, and anyone can type any name. A prefix is a block of IP addresses. Reverse DNS turns an address back into a host name. A WAF (web application firewall) is the filter in front of your site, such as Cloudflare, your host’s firewall or a security plugin.

1. Take the IP address from the log line.
2. Check whether it falls within one of the prefixes in `gptbot.json`. A match confirms OpenAI; the user-agent tells you which bot.
3. No match means an impersonator. If it’s causing trouble, block it on the server or in your WAF, not in robots.txt.

Between October 7 and November 11, 2026, 63 of the 341 requests claiming to be GPTBot (18.5%) came from outside OpenAI’s list, 57 of them from Google Cloud. My server refused 28 of them with a 403; confirmed GPTBot requests never got one.

To check a whole log file at once, load it into [Check AI bots in your logs](https://podrez.pl/en/ai-bots-in-logs/), my free log checker. The file never leaves your browser, and the tool checks every address against OpenAI’s list.

 

JavaScript

Does GPTBot run JavaScript?
---------------------------

**Probably not. OpenAI doesn’t say.**

A widely cited measurement by Vercel and MERJ (published December 17, 2024) found that none of OpenAI’s crawlers render JavaScript. OpenAI’s crawler, which Vercel calls ChatGPT, did fetch script files – 11.50% of its requests in that sample – but didn’t execute them.

A log only shows downloads: between October 7 and November 11, 2026, 2 of 278 confirmed GPTBot requests were for scripts.

The practical rule: text that appears only after JavaScript runs is probably invisible to GPTBot. If it matters, put it in the HTML your server sends.

 

Data from my logs

What GPTBot does on my site
---------------------------

GPTBot has slowed down a lot on my site. Between February 2 and October 6, 2026, it made 40 confirmed requests a day; between October 7 and November 11, 2026, only 8, on 30 of 36 days, peaking at 57 on November 9, 2026. The drop began in summer: 1,812 requests under the GPTBot name in June, 733 in July and about 400 a month in August and September.

What it fetches now is unusual for a training crawler: 89 requests to the WordPress REST API (`/wp-json/`), 59 to images, 33 to sitemaps and only 58 to pages. It also fetched 22 Markdown versions of my pages, which podrez.pl has offered since October 16, 2026. Only 22% of its successful requests landed on content.

podrez.pl logsFeb 2 – Oct 6, 2026Oct 7 – Nov 11, 2026Requests claiming to be GPTBot10,218341Confirmed requests9,925 (97%)278 (82%)Confirmed requests per day408

“Confirmed” means the IP address is on OpenAI’s published list. “Content” means any page other than the home page, robots.txt, sitemaps, WordPress system paths (/wp-content/, /wp-includes/, /wp-json/) and .php files. The February–October figures are the ones in my list [What bot is this?](https://podrez.pl/en/bots/#verification); October–November is a fresh count against the list as it stood on November 12, 2026. Identical log lines are counted once, and each request is assigned to a single bot: the one whose name appears first in the user-agent.

 

Questions

Frequently asked questions about GPTBot
---------------------------------------

  ### Does blocking GPTBot remove my site from ChatGPT?

No. GPTBot is only about training. ChatGPT search relies on OAI-SearchBot, and a page someone asks about is fetched by ChatGPT-User. OpenAI says the settings are independent, so you can disallow GPTBot and still appear in ChatGPT search.



   ### What is the difference between GPTBot and OAI-SearchBot?

GPTBot collects content that may be used to train OpenAI’s foundation models. OAI-SearchBot finds pages for ChatGPT’s search features: sites that block it won’t be shown in ChatGPT search answers, though they can still appear as navigational links. I describe it in my [OAI-SearchBot profile](https://podrez.pl/en/bots/oai-searchbot/).



   ### Should I block GPTBot?

Only if you don’t want your future content used to train OpenAI’s models. Blocking it doesn’t take you out of ChatGPT search – that’s OAI-SearchBot – but your pages stop feeding the data future models learn from. On podrez.pl I let all AI bots in, GPTBot included.



   ### Why don’t I see GPTBot requesting robots.txt?

On my site it never did under its own name. OpenAI says it may use one crawl for both GPTBot and OAI-SearchBot, and the robots.txt requests from GPTBot’s addresses came under the OAI-SearchBot name, with a robots.txt marker. A missing robots.txt request doesn’t mean GPTBot ignores the file.



   ### Does GPTBot honor the \_\_usage\_ai-training token?

OpenAI has announced that it will. On October 30, 2026, it said it plans to roll out support for the shared training signal in the coming months and will publish details when it does. Until then, use a rule that names GPTBot.



   ### Is every request with GPTBot in the user-agent from OpenAI?

No. Between October 7 and November 11, 2026, 63 of 341 weren’t, and all of them copied the example string from OpenAI’s documentation. Check the IP address against `openai.com/gptbot.json`.



 

 

Sources

Sources and method
------------------

I read all the sources below on November 12, 2026. Figures about my site come from the podrez.pl server logs for February 2 – October 6 and October 7 – November 11, 2026.

1. [OpenAI: Overview of OpenAI Crawlers](https://developers.openai.com/api/docs/bots) – GPTBot, OAI-SearchBot, ChatGPT-User, OAI-AdsBot, example user-agent strings, the robots.txt marker, IP lists.
2. [OpenAI: GPTBot IP address list (gptbot.json)](https://openai.com/gptbot.json) – 18 prefixes, file dated September 22, 2026.
3. [OpenAI: Helping websites express AI training preferences](https://openai.com/index/helping-websites-express-ai-training-preferences/) – support for the shared training signal planned “in the coming months”, October 30, 2026 (I read the Polish edition).
4. [IETF: Machine-Readable Training Preference Signal (individual Internet-Draft -00)](https://datatracker.ietf.org/doc/draft-kelley-aipref-training-preference-signal/) – the \_\_usage\_ai-training token, published October 30, 2026, expires May 3, 2027.
5. [Vercel: The rise of the AI crawler](https://vercel.com/blog/the-rise-of-the-ai-crawler) – measurement with MERJ, published December 17, 2024: which AI crawlers render JavaScript.

Compiled by Ewelina Podrez-Siama, working in SEO since 2009. Facts last checked: November 12, 2026.

**Other bot profiles:** [ClaudeBot](https://podrez.pl/en/bots/claudebot/) · [OAI-SearchBot](https://podrez.pl/en/bots/oai-searchbot/) · [Bytespider](https://podrez.pl/en/bots/bytespider/) · [Applebot](https://podrez.pl/en/bots/applebot/) · [Amazonbot](https://podrez.pl/en/bots/amazonbot/) · [Meta-ExternalAgent](https://podrez.pl/en/bots/meta-externalagent/) · [CCBot](https://podrez.pl/en/bots/ccbot/) · [PerplexityBot](https://podrez.pl/en/bots/perplexitybot/). Or go back to [What bot is this?](https://podrez.pl/en/bots/), my searchable list of bots and user-agents.

 

Check your site

What does your site tell GPTBot?
--------------------------------

Two free tools: one reads your robots.txt the way GPTBot would, the other shows whether GPTBot really visits you.

[Check robots.txt](https://podrez.pl/en/robots-txt-checker/)[Check your logs](https://podrez.pl/en/ai-bots-in-logs/)