Answertrail links the AI fetch on your site to the human session that follows — two events that objectively happened, in your own server logs. That link runs all the way to conversion: revenue credited to the AI answer that sent the visitor, including visits your analytics books as direct or organic.
In client data, 40 to 60% of AI search visits were invisible in Google Analytics.
Some AI search traffic is visible today: assistants sometimes add tagged links, and plenty of referrals do come through. But not all of them. Ad-blockers, consent banners and the in-app browsers of mobile apps lose referrals along the way. Many visitors arrive without any referrer at all — a new tab, or a link stripped along the way — and get booked as direct. And others double-check the AI answer on Google first — booked as organic. Those last two groups are invisible in every analytics package. And what's invisible doesn't get budget.
Extra AI search visits arrived via Google organic alone — people validating the AI answer before clicking through. On top of that come the visits booked as direct.
measured at a client · re-attribution based on linked sessions
Three steps, all based on events that actually happened on your site. No querying models, no sampling — measurement.
Every time an AI assistant fetches live content to answer a question, we record which URLs it consults. Nobody outside the AI companies knows the exact question — but the URLs fetched for a single answer reveal what it must have been about.
AI fetch · fingerprintWhen a visitor lands on the same URL shortly after — via a click-through, or via a double-check on Google — we link that session to the AI fetch. The tighter the time window, the stronger the attribution.
session · Δ time windowIf your site has a clear URL structure, we classify the AI question by search intent: informational, commercial, transactional. So you see not just that AI search sends traffic, but where in the funnel it lands.
intent · URL structurePrompt monitoring tests whether your brand gets mentioned in answers to invented questions. But AI answers are non-deterministic and personal: the same question gets a different answer tomorrow, and your customer's answer is not a test script's answer. A sample of fictional prompts is not a valid measurement.
Answertrail turns it around. We measure only events that are objectively verifiable on your own infrastructure: the fetch, the session, and the time between them.
Sessions currently booked as direct or organic get attributed to AI search — including the assistant the visit came from. Finally a solid answer to what GEO is delivering.
The exact question stays inside the AI — but the URLs fetched for one answer reveal its topic. Fingerprints automate that inference: clustered fetches show which themes AI assistants consult your content for, and which content is actually used in answers. Available as an add-on.
With a clear URL structure, we classify the intent behind every AI question. So you know whether AI search mostly sends you researchers — or visitors with buying intent.
Built into every report: how your site responds when AI bots come knocking. Many sites unknowingly refuse the very fetches that answer real user questions — and vanish from those answers. We show every refusal, per bot and per URL.
Here's what a refused user fetch means: someone asked a real question in ChatGPT or Claude, the assistant tried to open your pages to answer it — and your site said no. That answer went out without you in it. Not because your content wasn't good enough, but because a CDN rule, firewall or robots.txt got in the way. And nine times out of ten, nobody at your company ever decided that: your hosting provider or CDN set those rules by default, to keep pesky bots from straining the server.
Most tools can't see this happening. The AI statuscode report makes every refusal visible, per AI actor and per URL — and separates the live user fetches that answer real questions from routine training-crawler traffic. It's part of every Answertrail report, and we run it as a free audit on your own server logs.
Get your free auditNo install, no tracking script, no cookies — Answertrail runs entirely on the raw server logs you already have, so it even counts the visitors who decline your consent banner. Here's how it works:
An export straight from your web server. Server logs contain personal data, so agree the handling with us up front — we process them for the audit only.
The software identifies every AI actor, clusters the fetches and flags every blocked or throttled request — per bot, per URL.
A working session on your own data: what AI bots see, where your site pushes them away, and what full attribution would add on top.