Guide
How to track the traffic ChatGPT and other assistants send you
Two different things get called AI traffic: the assistants reading your pages, and the people those assistants send. They are measured in completely different ways, and most tools show neither.
Updated
The visits are the easy half
When somebody clicks a link inside ChatGPT, Perplexity, Claude or Copilot, the visit arrives with a referrer naming it, exactly like a visit from a forum. Any analytics tool that classifies referrers can show you those, and this one groups them as their own channel so they are not buried under "other".
What makes the number worth looking at is the column beside it. These visits convert differently from search visits, sometimes better because the assistant has already answered the comparison question, sometimes worse because the reader is still orienting. Revenue per visitor by source is the only way to know which is true for you, and it is the same report as for every other channel.
The crawlers need your server, not your script
A crawler does not run JavaScript, so a tracking script never sees it. Anything claiming to show you GPTBot in a dashboard built on a script alone is showing you something else.
What it takes is your own server telling us what it served: a few lines of middleware posting the user agent, path and status to an endpoint, authenticated with your site key. From that comes the half nobody else has, which is how many pages an operator read against how many people it then sent you, and what those people were worth.
The agents are also separated by what they came for, because it matters. Training crawlers take your text for a model. Indexing crawlers build the thing an assistant searches. Answer-time fetchers are reading your page right now because somebody is waiting for a reply, and that is the one that turns into a visit today.
ChatGPT read 230 pages and sent 61 visitors is a sentence about a business. A bot hit your site 230 times is not.
What nobody can measure, and what can be inferred
An answer that quotes your page without linking it is invisible to everyone, including to us. There is no referrer, no visit and no record, and any product that claims to count those is modelling, not measuring.
Google's AI Overviews are a related gap: Search Console counts their impressions and clicks inside ordinary web search with no separate dimension, so there is no filter to apply. What can be spotted is the signature of one, which is a query holding its position with steady impressions while clicks fall away. That list exists here and is labelled as an inference everywhere it appears, including in the API.
What to do with it
Check which pages the answer-time fetchers read most. Those are the pages assistants reach for when somebody asks about your category, and they are the ones worth keeping accurate and current.
Check whether your llms.txt and your plain markdown are being taken, which is reported separately from HTML. If the crawlers never touch them, the file is decoration.
Then compare the AI side against Google on the same chart, with revenue on both. The interesting answer is usually not "AI is taking my traffic", it is that one assistant sends a tenth of the visits and a third of the revenue, which changes what you write next.
The dashboard this is describing
Live demo, real queries, generated traffic. No sign-up.
- Organic Social2,412$3,120
- Direct1,072$1,240
- Referral771$960
- Organic Search720$580
- AI246$310
- 🇺🇸United States885$2,410
- 🇮🇳India887$140
- 🇬🇧United Kingdom427$980
- 🇩🇪Germany427$760
- 🇧🇷Brazil451$210
Questions
Do I have to install middleware to see any of this?
No. The visits from assistants are visible from the tracking script alone. The crawler half, which is what the operators read rather than what they sent, needs your server to report it, because a crawler runs no JavaScript and there is no honest way around that.
Should I block AI crawlers?
That is a judgement about your content, not a technical setting, and the numbers here are the input to it: how many pages an operator takes, against how many people it sends back. Our own robots.txt does not block them, which is the same position we would defend for a customer whose business is being found.
Is a visit from an assistant worth more than a search visit?
Sometimes, and it is site-specific. The reader arriving from an assistant has often had the comparison done for them, which can mean a shorter path to a payment. The way to know is the revenue per visitor beside each source, over enough payments to mean something.
