AI Traffic Analytics: From the Crawler to the Customer

Visits from ChatGPT, Perplexity, Claude and Gemini are now a channel of their own: GA4 added a native AI Assistant channel in May 2026 (Google's definition). That covers the people who click. The other half, the crawlers that read your pages so an assistant can cite them, does not show up in any page snippet, because AI crawlers fetch the HTML without running your JavaScript. Datalenk measures both halves and ties them to revenue.

The whole funnel: crawl, cite, refer, pay

1. Crawl: which AI engines fetch your pages, how often, and which pages they keep coming back for.
2. Cite: the visits those engines send back, attributed to the engine that sent them.
3. Refer: the AI channel broken down by engine, ChatGPT, Perplexity, Claude, Gemini, Grok, DeepSeek and the long tail, classified automatically, no regex homework.
4. Pay: the revenue behind each engine. "Perplexity sent 212 visits" becomes "Perplexity sent $1,840 in new MRR", so your GEO effort goes where the money is instead of where the volume is.

Put end to end, that gives you the crawl-to-refer ratio per engine: what each one takes from your bandwidth versus what it gives back in humans and in dollars.

Training crawlers vs user-triggered crawlers, and why the difference is the whole game

Not every AI bot means the same thing. Datalenk separates them, because they are two different businesses.

• Training crawlers (GPTBot, ClaudeBot, CCBot and friends) are hoovering your content to train a model. They may pay you back one day, or never.
• User-triggered fetches (ChatGPT-User, Perplexity-User, OAI-SearchBot) mean a human asked a question right now and the assistant went to read your page to answer it. That is not training, that is demand. It is the leading indicator of the visit and the sale that may follow.

Reading those two as one number tells you nothing. Split, they tell you whether you are feeding a model or being consulted as a source.

And the verification matters. Anybody can put "GPTBot" in a user agent string. We check every claimed AI bot against the published IP ranges the engines document (OpenAI, Perplexity) and against reverse DNS for Google, Bing and Apple. Only the IP decides. The catalog covers 30+ crawlers and grows as engines ship new ones, on our side, not yours.

How it works

1. Add the script. AI referrers are classified into the AI channel automatically from day one, split by engine.
2. Add the server beacon to see the crawler side. Bots do not run JavaScript, so no snippet can see them. You add a few lines in your middleware or edge function that POST the request to /api/server. That is what surfaces GPTBot and the rest. Not server logs, not a proxy, a beacon you drop in once and forget.
3. Connect your payments (Stripe with a read-only key, or any of a dozen native providers, or the payments API) and every AI engine gets a revenue column.

How this differs from GA4 and GEO tools

GA4's AI Assistant channel counts the visits that carry an assistant's referrer, and ties them to the purchases recorded on your site. It never sees a crawler, because crawlers do not run its tag. GEO tools such as Profound or Otterly work from the other side: they query the assistants and report whether your brand is mentioned, without seeing your site. Datalenk measures what reaches your site, bots and humans, verifies the bots against their published IP ranges, and ties both to the revenue in your payment provider, renewals included.

Honest scope

Live today: the 30+ crawler catalog with IP and reverse-DNS verification, the training vs user-triggered split, the AI channel broken down by engine, the crawl-to-refer ratio, and revenue per engine.

Not built, and we will not pretend otherwise: we do not ingest ad spend from anyone, including the ad products now appearing inside AI assistants. If you tag your paid placements with UTMs we will report their traffic and their revenue like any other campaign, but we will never show you a ROAS we cannot compute.

Go deeper

The full money-side guide: ↗. How to track ChatGPT specifically: ↗. The take-versus-give metric, engine by engine: ↗. What ads inside AI assistants change for this channel: ↗.

See what the AI era actually sends you, and what it takes. 14-day trial, card required, cancel in two clicks. Start your trial.

FAQ

Can I see ChatGPT traffic in Datalenk? Yes, classified into the AI channel automatically and split by engine, with conversions and revenue attached, plus the crawler side that a JavaScript tag cannot see.

Why do I need a server-side beacon to see AI crawlers? Because crawlers do not execute JavaScript. A snippet in your page is invisible to them by definition. A few lines in your middleware or edge function that post to /api/server is the only honest way to see them, so that is what we ask for.

Can a bot fake being GPTBot? It can claim to be, in its user agent, and plenty do. We do not take its word for it: we verify against the IP ranges the engines publish and, for Google, Bing and Apple, against reverse DNS. What you see in your crawler report has proven who it is.

What is a crawl-to-refer ratio? AI crawler requests divided by the human visits an engine refers back, per engine. It tells you whether an AI engine is buying distribution with your bandwidth, or just taking.