How to Know if ChatGPT Crawled Your Website

Three OpenAI crawlers, one shorthand

The phrase “ChatGPT crawled my site” is useful shorthand, but OpenAI publishes separate identities for separate jobs.

Identity	What it does
OAI-SearchBot	Crawls public pages for ChatGPT Search
GPTBot	Collects content that may be used for model training
ChatGPT-User	Fetches some pages after a user asks ChatGPT to open them

Claude has the same kind of split: Claude-SearchBot, ClaudeBot, and Claude-User. Perplexity documents PerplexityBot and Perplexity-User.

A GPTBot visit doesn't tell you whether ChatGPT Search indexed the page, and an OAI-SearchBot request doesn't tell you whether GPTBot used the page for training. That is why the identity belongs in the log, not just the fact that a crawler visited.

What counts as evidence

Start with a request your own infrastructure observed. At minimum, keep:

the exact host and path
the request time
the full user-agent value
the HTTP status returned
whether the source could be verified against current provider evidence

A user-agent string is only a claim. It is easy to spoof. Where a provider publishes network ranges or supports verified-bot programs, verify the connection source with that evidence and record when you checked it.

The HTTP outcome changes the meaning of the event. A 200 means the page was served. A 301 or 302 means the crawler was redirected. A 403 means a security layer refused the request. A 429 means it was rate-limited.

Where to look

If the site uses Cloudflare, open AI Crawl Control and inspect crawler activity, paths, and outcomes. On Vercel, inspect Firewall events and request logs. On a server you manage, search the nginx or Apache access logs.

SearchSeal collects the same class of first-party evidence from connected infrastructure. It keeps crawler requests separate from human website analytics, so a bot request never becomes a visitor or conversion.

If your analytics tool only runs JavaScript in a browser, it is the wrong place to look. Many crawlers do not execute the analytics tag, and Google Analytics automatically excludes known bots.

If you find no visits

“No requests observed” is valid only when you know which source and time range were checked. If the CDN, host, or log stream was not connected, the result is unknown.

Start with coverage. Confirm that the source includes the right hostname and time range. Then fetch the public robots.txt file, inspect CDN, WAF, and rate-limit rules, and look for redirects or 4xx responses. Finally, confirm that internal links or a sitemap expose the page.

A crawl is not a citation

Keep four events separate:

A crawler requested a page.
A search system indexed or stored information about it.
An AI answer cited or mentioned it.
A person clicked through and converted.

Treat these as four separate observations. If you measured the first one, report the first one. Do not quietly promote it into the second, third, or fourth.

Frequently asked questions

Can Google Analytics show whether ChatGPT crawled my site?

Google Analytics cannot answer this. It measures browser events and automatically excludes known bot traffic. Use CDN or server request evidence instead.

Does an OAI-SearchBot visit mean ChatGPT cited the page?

The request record only shows that OAI-SearchBot reached the site and received the logged response. It says nothing about whether the page later appeared in an answer.

Which OpenAI crawler should I check for ChatGPT Search?

Check OAI-SearchBot for automatic ChatGPT Search crawling. GPTBot and ChatGPT-User have different purposes.