GPTBot
Whether your pages can be used to train OpenAI models
What GPTBot does
GPTBot collects pages that may be used to train OpenAI's models. It is the token to disallow if you do not want your writing in a training set, and disallowing it has no effect on whether ChatGPT can cite you in an answer.
It is the most-blocked OpenAI crawler in our survey, and that is a legitimate outcome rather than a mistake: publishers who sell access to their archive have an obvious reason to refuse. The mistake is what tends to travel with it.
OpenAI publishes the IP ranges GPTBot fetches from, which matters because a user agent string is a claim anyone can make. Traffic calling itself GPTBot from an address outside that list is something else wearing the name, and it is not obliged to respect anything in your robots.txt.
GPTBot is not OAI-SearchBot
Blocking GPTBot is a training opt-out and costs no citations. Blocking OAI-SearchBot alongside it, which most snippets do, costs all of them.
What OAI-SearchBot does insteadHow to block GPTBot
Add this to the robots.txt at the root of your domain, alongside what is already there rather than replacing it.
robots.txt
User-agent: GPTBot
Disallow: /What that costs you: Nothing, for citations. Blocking this is a legitimate choice and does not remove you from ChatGPT search.
To allow it instead, write nothing. A crawler with no group of its own follows your User-agent: * rules, and giving it a named group with Allow: / detaches it from those rules entirely — including any path you disallow there.
Find it in your logs
The short token is what robots.txt matches. This is what actually arrives in an access log, copied from OpenAI’s documentation. Version numbers move, so match on the token rather than the whole string.
User-agent
Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko); compatible; GPTBot/1.4; +https://openai.com/gptbotTo find every visit it has made, and the addresses it came from:
Terminal
grep -F 'GPTBot' /var/log/nginx/access.log | awk '{print $1}' | sort | uniq -c | sort -rnCheck that it is really GPTBot
A user agent is a string the visitor chooses, so on its own it proves nothing — anyone can send yours. OpenAI publishes the addresses GPTBot fetches from: 21 prefixes when we read the file, last changed on 30 October 2025. Take an address out of your logs and check it against that list.
Terminal
curl -s https://openai.com/gptbot.json | python3 -c "
import sys, json, ipaddress
addr = ipaddress.ip_address('203.0.113.10') # the address from your logs
nets = json.load(sys.stdin)['prefixes']
print(any(addr in ipaddress.ip_network(n.get('ipv4Prefix') or n['ipv6Prefix']) for n in nets))
"False means something else is wearing the name, and it is under no obligation to respect anything in your robots.txt. That is the traffic worth blocking at the edge, rather than the crawler that was asking politely.
What OpenAI says about it
Changes take about a day
OpenAI states it can take roughly 24 hours from a robots.txt change for their systems to adjust. If you fix a block and nothing happens immediately, that is why. Check again tomorrow before changing anything else.
It labels its own robots.txt fetches
When OpenAI fetches robots.txt rather than a page, it may add a robots.txt marker to the user agent, giving … compatible; OAI-SearchBot/1.4; robots.txt; +https://openai.com/searchbot. That lets you tell a policy read from a content fetch in logs that do not record paths.
Read from https://platform.openai.com/docs/bots on 14 September 2026.
Should you block GPTBot?
A genuine choice with no citation cost, which makes it the easiest decision on this list. If you would rather your writing were not in a training set, disallow it, and nothing about your visibility in ChatGPT changes.
The argument for allowing it anyway is weaker than people assume. Being in a training set does not get you cited: models do not reliably attribute what they learned in training, and the citation path runs through OAI-SearchBot instead.
What matters far more than your answer here is that you do not accidentally take OAI-SearchBot with it.
Questions
- Does blocking GPTBot remove me from ChatGPT?
- No. ChatGPT's citations are governed by OAI-SearchBot. Sites that disappeared from ChatGPT after editing robots.txt almost always blocked both.
- Does blocking GPTBot remove content already used in training?
- No. robots.txt governs future crawls. It is a forward-looking instruction and cannot retract what was collected before it was written.
The user agent and IP figures on this page were read from OpenAI’s own URLs on 14 September 2026. Blocking figures come from our survey of 170 well-known sites across 10 categories on 14 September 2026, 148 of which returned a readable robots.txt.
Is GPTBot allowed on your site right now?
The checker reads your live robots.txt and reports every token in this directory, with the exact directive that decided each one.