Your Server Logs Are the Only Place You Can See AI Crawlers
Analytics cannot see AI assistants fetching your site, because they run no JavaScript. What the server log shows, and what it genuinely cannot.
If you want to know how often ChatGPT, Claude or Perplexity are fetching your site, your analytics will not tell you. They cannot. The whole measurement has to move down a layer.
Analytics runs JavaScript, and crawlers do not
Google Analytics and every tool like it works by running a script in a browser. An AI assistant fetching your page makes an HTTP request and reads the HTML. No browser, no script, no event. As far as your analytics is concerned the visit did not happen, and no amount of configuration changes that.
The request is in your web server's access log, though, with a user agent that says who it was. That log is the only complete record you have.
What you are looking for
The vendors publish their user agents and they are reasonably honest about them. Broadly they fall into two groups, and the difference matters more than the volume.
Some fetches are training or index crawls, running on a schedule, working through your site whether or not anybody asked. Others happen because a person asked a question right then and the assistant went to look. The second group is the interesting one: each of those fetches is roughly one question somebody asked that your page was considered an answer to.
What it cannot tell you
It cannot tell you how many people. The request comes from the vendor's servers, not from the person asking, so the address in your log is a data centre. One fetch may serve one person or be cached and serve many. Anyone promising you a headcount for AI traffic is guessing.
It also cannot tell you that you were cited. You know your page was fetched. Whether the answer used it, quoted it or ignored it is not in the log.
Two things that will skew the numbers
A firewall or bot protection in front of your site may be blocking these fetches before they ever reach your server, in which case your log shows a floor rather than a total. Anything that requires JavaScript to render is particularly affected, since assistants do not run it. Check what your protection layer is dropping before you read anything into a low number.
The second is counting the wrong rows. Sitemaps, feeds, images and 404s are all in the same log. If you want "pages served to an assistant", filter to successful requests for actual pages, or your numbers are flattering and wrong.
Why bother
Because it is the only view anyone has, and most sites have none. Knowing which of your pages assistants keep coming back to tells you what they consider authoritative, and that is a more honest signal than a ranking position. It is also a conversation nobody else is having with your clients yet.