Google-Extended vs GPTBot
In the census run that finished 2026-10-24, robots.txt closed the homepage to Google-Extended on 516 of the 4,501 top sites that serve a robots.txt (11.5%), and to GPTBot on 644 (14.3%).
| Google-Extended | GPTBot | |
|---|---|---|
| Operator | OpenAI | |
| What it is | robots.txt control token (never sent) | training crawler |
| What it does | Gemini training | OpenAI training + index |
| How a request can be verified | It is a robots.txt control token, not a crawler: no request ever carries this name, so any request that claims it is not from Google. | Its operator publishes the IP ranges it crawls from (https://openai.com/gptbot.json), so a request carrying the name can be checked against that list. |
| robots.txt closes the homepage (of sites with a robots.txt) | 516 (11.5%) | 644 (14.3%) |
| Refused at the door (of sites that answered a browser) | not requested as this identity | 1,108 (20.0%) |
Top 10,000 sites, census of 2026-10-24. “Refused at the door” is a non-2xx answer to a client declaring that name, from CrawlCheck’s own addresses, where the browser got 2xx. Every crawler on the radar.
Let Google-Extended in and keep GPTBot out
User-agent: Google-Extended
Allow: /
User-agent: GPTBot
Disallow: /A crawler obeys only the most specific group that names it and ignores * entirely. If your User-agent: * group has Disallow rules, repeat them under Google-Extended, or naming it drops them. Swap the two names to do the reverse.
Check what your own site tells both: scan it free. Catalogue data CC BY 4.0 at /ai-crawlers.json.