Measure AI traffic
AI traffic has two halves. Your analytics sees one.
“How much is AI actually sending us, and what is it doing to the rest of our demand?”
The visits from ChatGPT, Perplexity, and Copilot show up in your analytics, small and undercounted. The answers that name you, cite you, or recommend a competitor never send a click and shape the decision anyway. We set up the first half in your analytics and measure the second every day, on every major AI engine, so both halves land in the same monthly report.
What stands in the way
Why nobody in the building can say what AI is doing to demand
The tools were built for a results page and a click. AI answers deliver neither, and the parts that do reach your analytics arrive mislabelled.
A sliver called “referral”
GA4 puts chatgpt.com, perplexity.ai, and copilot.microsoft.com in the same bucket as every other link on the web. The AI visits are in there, but no default report separates them, so nobody knows what the sliver contains.
Rankings hold while sessions fall, and no row says why
The head terms sit where they sat last year. The organic line slides anyway, because the answer above the results took the click. There is no row in Search Console or GA4 that says so.
AI Overviews and AI Mode are invisible
A click from a Google AI Overview or AI Mode arrives as google.com organic, the same referrer as a blue link. Nobody can pick the largest AI surface your buyers use out of referrer data.
Leadership asks, and the answer is a screenshot
The board asks what AI is doing to the business. Somebody types the category into ChatGPT, screenshots the answer, and pastes it into a deck. That is one engine’s answer on one day, and it becomes the whole measurement.
A vendor sells “AI traffic” and delivers a count
The pitch says AI traffic. What arrives is a count of mentions without a prompt list, an engine split, a window, or a method, so you cannot compare it with last month or with anything else you report.
Why now
The half you can count is small. The half you cannot is where the click went.
Third-party figures, linked to their sources.
68%
of Google searches ended without a click in the first four months of 2026
Up from 60% in 2024, on a US clickstream panel. The same post notes that AI tools send less than 1% of all outbound traffic.
61%
drop in organic CTR on queries that show an AI Overview
1.76% to 0.61% across 3,119 terms at 42 companies, June 2024 to September 2025. None of that loss carries an AI label in your analytics.
12.1%
of Ahrefs’ signups came from 0.5% of its visits, the ones referred by AI search
A 30-day window in June 2025, most of it from ChatGPT. A channel that small and that valuable deserves its own row, and the referrer is the only thing that gets it one.
26%
of enterprise marketing leaders cannot track an AI discovery through to a conversion
300 US leaders at companies of 500 or more, January 2026. Another 24% say their analytics tools cannot handle AI attribution at all.
How it works
Both halves, measured the way each one can be
Your analytics counts the visits, once they are set up as a channel. Our platform records the answers every day, on every major AI engine. The monthly report puts the two side by side.
The visits
AI referrals as their own channel, in your analytics
The visits AI assistants send you already sit in GA4, filed under referral beside every other link on the web. As part of an engagement we set up a channel for them: a grouping or segment that gathers chatgpt.com, perplexity.ai, copilot.microsoft.com, gemini.google.com, and the rest, with landing pages and conversions attached, so the row appears in the reports your team already opens.
This half lives in your analytics, not on our platform, and it stays undercounted. Referrers get stripped, some assistants open links without one, and a click from a Google AI Overview or AI Mode arrives as google.com organic, where nobody can tell it apart from a blue link. We say so on the report rather than pretend the row is complete.
- One channel for the AI assistant referrers, with landing pages and conversions
- Read beside the answer data every month, with what it cannot show stated on the page
The answers
What the engines say about you, recorded every day
Your buyers’ questions run once a day on ChatGPT, Gemini, Perplexity, Copilot, and Google AI Mode. The platform keeps every answer in full, with the sources it cited, the brands it named, and how early it named each one. Nothing here depends on a click happening, because most of the time none does.
From that record come the four numbers: mention rate, how often an answer names you; citation rate, how often it cites a page of yours; share of voice against the competitors you name; and position score, 0 to 100, for how early in the answer you appear.
- Brands detected by matching aliases and owned domains rather than by guessing
- Per engine and per country, with a persona on each prompt
The cuts
Cut the way the business is run, saved, and shared by a link
Nobody trusts a number they cannot slice, so every metric filters by engine, country, persona, tag, and prompt, and any cut can be saved as a named report: one for the board, one per market, one per buying persona. A saved report renders the same 24 metrics every time, so two months are comparable.
Every figure carries its delta against the previous window of the same length; nobody picks the comparison. To share a report you send its link, and a colleague with a login sees exactly what you saw. There is no PDF export; the monthly written report is the document for people who do not log in.
- Window, engine, country, prompt, tag, persona: leave one empty and it means all
- Every delta is the previous equal-length window
The reading
Movement tied to the work that was shipped
A number that moved for no reason you can name will move back. So the monthly report reads what changed beside what shipped that month: the fixes that went live, the pages published, and the sources earned. A prompt that gained twelve points of share arrives with the comparison page that went live two weeks earlier.
Movers rank on each window’s share of its total, in percentage points, so a month with more answers never reads as growth. When a window has too little data, the change shows a dash rather than a zero, and the report says so.
- Prompts, competitors, and cited domains that gained, lost, appeared, or dropped
- On Managed GEO the shipped list is ours; on consulting it is your teams’, and our strategist reads it
The limits
What it cannot tell you, stated on the page
The engines publish no impression counts, so nobody can tell you how many people saw an answer that named you. There is no per-user data, no sentiment score, and no conversion inside a chat. Anyone who reports those numbers is inventing them.
What can be measured is the daily record of what the engines answer over a window: how often you are named, how often you are cited, how early, and against whom. That record stands in for the half of AI traffic that never clicks. It sits beside the referral row, and the report says which is which.
- Daily runs, not real time: a report is at most a few hours behind the last run
- US, UK, Brazil, and Greece today
Engagement models
The measurement comes with both engagements
Both start with a free audit and both run daily tracking. What changes is who does the work the numbers are tied to.
For small and medium businesses
Managed GEO
We set up the AI channel in your analytics, run your buyers’ questions every day on every major AI engine, and ship the fixes and the content the numbers respond to. One monthly call reads what moved beside what we shipped, so every number comes with the reason it moved. From $990 a month, scoped after the audit.
For large companies
GEO Consulting & Trends
We set up the measurement across your markets, engines, and personas, with a report per question your leadership asks, and a strategist who reads the movement beside the work your teams shipped. Your teams execute; the monthly report and the trend briefing are ours. Quoted per engagement.
The platform behind the work
Where the answer half is measured
- A saved filter set over 24 metrics: window, engine, country, prompt, tag, persona, with every delta against the previous window. Shared by link.
Custom Reports
- Your buyers’ questions run daily on every major AI engine, and every answer is kept with its citations and the brands it named.
Prompt Tracking
- The same numbers split by who is asking, so a fall in one persona does not hide inside a flat average.
Visibility by Persona
- For SEO and content leads: citations and share of voice in the same deck as rankings, and a fix list per template.
SEO & Content
Questions we get about measurement
Before you put a number in the deck
Does the platform measure AI referral visits?
No. The platform records answers, not visits: which engines name you, cite you, and recommend a competitor, every day, for the prompts you track. Visits from AI assistants are already in your own analytics, filed under referral, and that is where they should stay counted. As part of an engagement we set that channel up in GA4 or whatever you run, with landing pages and conversions, and read it beside the answer data in the monthly report. Nobody should sell you a second analytics tool for a row your current one can show.
Why can’t we see AI Overviews or AI Mode in our referrer data?
Because Google sends them as google.com. A click from an AI Overview, an AI Mode answer, or a blue link all arrive with the same referrer, and neither Google Analytics nor Search Console labels which surface the click came from. No tool can separate them from referrer data, whatever its pitch says. What you can see is the effect: impressions holding while clicks fall on the queries that show an Overview. For those surfaces the answer record is the measurement. We run the prompts on Google AI Mode daily and report whether you are named and cited there.
What do we set up in our analytics, and who does it?
A channel for AI assistant referrers: chatgpt.com, perplexity.ai, copilot.microsoft.com, gemini.google.com, and the rest, defined as a channel grouping or a segment so it appears in the reports your team already opens, with landing pages and conversions attached. On Managed GEO we do it with whoever holds your analytics access. On consulting we specify it and your analytics owner ships it, a small job in most setups. Either way the row is yours, in your account, and it keeps working if we are no longer in the picture.
AI referrals are under 1% of our sessions. Why measure them at all?
The sessions are the small half. Most AI answers never send a click, and the ones that do are undercounted, so the referral row understates AI’s effect on demand by a wide margin. It is still worth a row: on the published studies those visits convert well above organic, and a small channel that converts is easy to lose in a bucket called referral. The larger measurement is the answer record, which does not need a click to happen. Put both in the report, say which is which, and the under-1% figure stops being the whole story.
How is measuring different from auditing what AI says?
An accuracy audit is a read of the truth at one moment: a person on our team goes through the recorded answers for your prompts, engine by engine, flags what is wrong or outdated, and traces each claim to the source the answer cited. Measurement is the ongoing record: the same prompts run every day, and each month the report sets mention rate, citation rate, share of voice, and position against the previous window. The audit tells you what to fix. The measurement tells you whether the fix, and everything else you shipped, moved the answers.
See what AI is already saying about your brand.
A free audit shows you exactly where you stand across ChatGPT, Gemini, Perplexity, Copilot, Claude, Grok, and AI Overviews, and what it would take to close the gap.