GPTBot (impostor)'s food bowl
the bowl, as bots see itThe bowl is empty.
Identification
Its full user-agent string, exactly as it arrives at the gate:
This is not a real species. It is a costume: requests that use GPTBot's name but fail reverse-DNS or IP-range checks against OpenAI's network.
How it was checked: the address must be in OpenAI's published IP ranges. Each address is checked when it first visits, and the result is kept for a day.
How to block GPTBot (impostor)
GPTBot (impostor) does not read robots.txt, so a polite sign is wasted on it. Refuse it at your web server or firewall instead. User agents are easy to fake, so pair this with rate limiting.
# robots.txt will not stop GPTBot (impostor). Block it at the server.
# nginx
if ($http_user_agent ~* "GPTBot") {
return 403;
}The same thing on Apache:
# Apache (.htaccess)
RewriteEngine On
RewriteCond %{HTTP_USER_AGENT} GPTBot [NC]
RewriteRule .* - [F,L]Observed behaviour
Most active around 12:00. Office hours, like a professional.
Requested 138 disallowed pages out of 173 requests. Read robots.txt 3 times.
Walked through the hidden /trap/ door. Last caught 26 days ago.
Where it comes from
Claims to be GPTBot. Real GPTBot traffic comes from OpenAI's own network; these visits came from:
Networks and countries come from the visitor's IP address, looked up in a local copy of the DB-IP database. The addresses themselves are never stored.
Keeper's field notes
Questions site owners ask
Does GPTBot (impostor) respect robots.txt?
Mostly. It reads robots.txt, but it has fetched a disallowed page 138 times out of 173 requests observed here.
Will blocking GPTBot (impostor) hurt my search rankings?
No. Nothing respectable will miss it.
How often does GPTBot (impostor) visit?
Here, about 2 requests a day over the last week. Visits to your site depend on its size, how often it changes, and how many links point to it.