resulthack

All AI crawlers /

GPTBot

OpenAI · Training · robots.txt: Yes

Crawls public web pages whose content may be used to train OpenAI's generative AI foundation models.

Block GPTBot

User-agent: GPTBot
Disallow: /

Blocking GPTBot only opts your content out of training. It does not remove you from ChatGPT search, which is controlled separately by OAI-SearchBot. If you allow both, OpenAI may use one crawl for both purposes.

Allow GPTBot

User-agent: GPTBot
Allow: /

An allow group only changes anything if a broader rule would block GPTBot. Under the robots.txt standard (RFC 9309, section 2.2.1) a crawler follows the group that names it and uses the User-agent: * group only when no group does — so this group lets it in even if your * group says Disallow: /. Token matching is case-insensitive.

Put these lines in /robots.txt at the root of each host (each subdomain has its own file). robots.txt is a request to well-behaved crawlers, not access control.

Block or allow it at the server

robots.txt asks; a server rule refuses. Every rule here matches GPTBot, spelled as in the User-Agent OpenAI documents (“GPTBot/1.4”), and matches it case-sensitively. The server answers 403.

What a User-Agent rule cannot do: any client can send any User-Agent. Access control by User-Agent is an unreliable technique, since the User-Agent header can be set to anything at all, at the whim of the end user. (Apache HTTP Server 2.4 documentation). A fake GPTBot is refused by the block rule along with the real one, and a crawler that hides its name passes. The “only the real one” rule is the check: it lets the name in only from OpenAI's published addresses.

Your server

To block it: refuse every request that says GPTBot

In the server block (or a location)

if ($http_user_agent ~ GPTBot) {
    return 403;
}

To allow only the real one: refuse GPTBot claims from outside OpenAI's list

In the http block

geo $gptbot_unlisted {
    default 1;
    132.196.86.0/24 0;
    172.182.202.0/25 0;
    172.182.207.0/25 0;
    20.125.66.80/28 0;
    20.171.206.0/24 0;
    20.171.207.0/24 0;
    4.227.36.0/25 0;
    52.230.152.0/24 0;
    74.7.175.128/25 0;
    74.7.227.0/25 0;
    74.7.227.128/25 0;
    74.7.228.0/25 0;
    74.7.230.0/25 0;
    74.7.241.0/25 0;
    74.7.241.128/25 0;
    74.7.242.0/25 0;
    74.7.243.128/25 0;
    74.7.244.0/25 0;
}
map $http_user_agent $gptbot_fake {
    default 0;
    "~GPTBot" $gptbot_unlisted;
}

In the server block (or a location)

if ($gptbot_fake) {
    return 403;
}

The geo and map blocks go at http level; nginx's docs give Context: http for both, and for if Context: server, location. By default, the address is taken from the $remote_addr variable (nginx documentation).

To block it: refuse every request that says GPTBot

In the virtual host, a <Directory> block or .htaccess

<If "%{HTTP_USER_AGENT} =~ /GPTBot/">
    Require all denied
</If>

To allow only the real one: refuse GPTBot claims from outside OpenAI's list

In the virtual host, a <Directory> block or .htaccess

<If "%{HTTP_USER_AGENT} =~ /GPTBot/">
    Require ip 132.196.86.0/24 172.182.202.0/25 172.182.207.0/25 20.125.66.80/28 20.171.206.0/24 20.171.207.0/24
    Require ip 4.227.36.0/25 52.230.152.0/24 74.7.175.128/25 74.7.227.0/25 74.7.227.128/25 74.7.228.0/25
    Require ip 74.7.230.0/25 74.7.241.0/25 74.7.241.128/25 74.7.242.0/25 74.7.243.128/25 74.7.244.0/25
</If>

Apache httpd 2.4. Syntax: <If expression> ... </If> Context: server config, virtual host, directory, .htaccess (Apache HTTP Server 2.4 documentation). Several Require ip lines in one block: When multiple Require directives are used in a single configuration section and are not contained in another authorization directive like <RequireAll>, they are implicitly contained within a <RequireAny> directive. (Apache HTTP Server 2.4 documentation)

To block it: refuse every request that says GPTBot

Inside the site block

@gptbot header User-Agent *GPTBot*
respond @gptbot 403

To allow only the real one: refuse GPTBot claims from outside OpenAI's list

Inside the site block

@fake-gptbot {
    header User-Agent *GPTBot*
    not {
        client_ip 132.196.86.0/24 172.182.202.0/25 172.182.207.0/25 20.125.66.80/28 20.171.206.0/24 20.171.207.0/24
        client_ip 4.227.36.0/25 52.230.152.0/24 74.7.175.128/25 74.7.227.0/25 74.7.227.128/25 74.7.228.0/25
        client_ip 74.7.230.0/25 74.7.241.0/25 74.7.241.128/25 74.7.242.0/25 74.7.243.128/25 74.7.244.0/25
    }
}
respond @fake-gptbot 403

Like directives, named matcher definitions must go inside the site blocks that use them. (Caddy documentation) On client_ip: This matcher is best used when the trusted_proxies global option is configured, otherwise it acts identically to the remote_ip matcher. (Caddy documentation)

To block it: refuse every request that says GPTBot

Expression of a custom rule; action Block

(http.user_agent contains "GPTBot")

To allow only the real one: refuse GPTBot claims from outside OpenAI's list

Expression of a custom rule; action Block

(http.user_agent contains "GPTBot" and not ip.src in {132.196.86.0/24 172.182.202.0/25 172.182.207.0/25 20.125.66.80/28 20.171.206.0/24 20.171.207.0/24 4.227.36.0/25 52.230.152.0/24 74.7.175.128/25 74.7.227.0/25 74.7.227.128/25 74.7.228.0/25 74.7.230.0/25 74.7.241.0/25 74.7.241.128/25 74.7.242.0/25 74.7.243.128/25 74.7.244.0/25})

Each expression is one custom rule with the action Block (Cloudflare's own example: Expression: (cf.bot_management.score lt 2 and http.user_agent contains "App_Name 2.0") Action: Block).

Addresses in the “only the real one” rules: OpenAI's list openai.com/gptbot.json, 18 ranges, fetched 8 Oct 2026, list dated 22 Sep 2026. Operators change their lists; compare the date with the live list before you rely on it.

The IP rules compare the address your server sees. Behind a CDN or load balancer that is the proxy's address, so every claim would be refused: restore the client address first, or put the rule at the CDN.

Syntax from each server's own documentation, fetched 9 Oct 2026: nginx (rewrite module, map module, geo module); Apache (Access Control, <If>, expressions, mod_authz_core, mod_authz_host); Caddy (request matchers, respond); Cloudflare (custom rule example, operators, values and lists, http.user_agent). Every line form above is listed with its quote in server-rules-sources.json.

Who blocks it

Of the 1,000 most-visited domains on Tranco list PY35J (generated 2026-10-08), we read 508 robots.txt files on 2026-10-09 and ran each through the same RFC 9309 matcher as the checker, for the path /. 81 block GPTBot from the whole site by name, 22 block it through their User-agent: * group, and 28 name it with partial rules (some paths closed, the home page open). 17 more name it only to let it in.

81

block it by name 15.9%

22

block it through * 4.3%

28

name it, partial rules 5.5%

Percentages are of the 508 files we could read, not of all 1,000 sites.

Highest-ranked sites that block GPTBot by name

  1. fbcdn.net rank 12
  2. amazon.com rank 27
  3. whatsapp.net rank 32
  4. whatsapp.com rank 57
  5. tiktok.com rank 58
  6. yahoo.com rank 61
  7. msn.com rank 72
  8. snapchat.com rank 134
  9. gravatar.com rank 154
  10. nytimes.com rank 157

Each links to the site's live robots.txt, which may have changed since 2026-10-09.

robots.txt is only what a site asks for. A site can also turn a crawler away at its firewall or CDN, or let an operator in under a licensing deal, and neither shows in its robots.txt. A block through * stops every crawler the file does not name, so it says nothing about this one in particular. The files we could not read are not counted: 222 of the 1,000 domains are infrastructure names with no website at the bare domain (CDN and DNS zones, for example), 59 answered our request with 401 or 403, 93 failed (timeout, TLS or server error), 59 have no robots.txt (404 and similar) and 59 answered with an HTML page instead of a robots.txt file. Every agent's count, side by side.

Facts

Operator
OpenAI
Purpose
Model training
Obeys robots.txt
Yes. OpenAI documents GPTBot as a robots.txt token: disallowing it tells OpenAI the site's content should not be used for training. When it fetches robots.txt it may add a robots.txt marker to its user-agent string.
In your server logs
The user-agent string contains "GPTBot/1.4"; OpenAI notes the version number may change.
How to verify it
Match the request's IP address against OpenAI's published GPTBot IP list.

Published list: openai.com/gptbot.json — 18 IPv4 ranges when we fetched it, list dated 22 Sep 2026. Always match against the live list; operators update them. To count real and fake GPTBot requests in your own server log against this list, drop the log into the log reader.

What OpenAI says

Quoted word for word from the operator's documentation.

“GPTBot is used to make our generative AI foundation models more useful and safe. It is used to crawl content that may be used in training our generative AI foundation models.”

“Disallowing GPTBot indicates a site’s content should not be used in training generative AI foundation models.”

“Each setting is independent of the others – for example, a webmaster can allow OAI-SearchBot in order to appear in search results while disallowing GPTBot to indicate that crawled content should not be used for training OpenAI’s generative AI foundation models.”

“If your site has allowed both bots, we may use the results from just one crawl for both use cases to avoid duplicative crawling.”

“When fetching robots.txt files, we may add a robots.txt marker to the user-agent string to help site owners distinguish those requests from requests for other resources, especially when logs do not include paths.”

“Example user-agent string (the version number may change): Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko); compatible; GPTBot/1.4;”

“Published IP addresses:”

Sources

Other crawlers from OpenAI

Same purpose, other operators

← All 49 AI crawlers

Made with AI tools. · Privacy