What a retrieval event actually is
When a Live Search Agent visits a page, it was sent there for a reason.
A real user asked a question.
The agent was dispatched to find an answer.
Your content was judged relevant.
The agent is reading it to help form a response.
That's a tight chain of intent.
The user query reveals what they want to know.
The pages retrieved reveal what content addresses it.
The retrieval event connects your content to a specific, active intent: live, at the moment it's happening.
Every one of those events is a signal.
Almost all of them are currently disappearing.
The audience that's taking shape
The publishers on our network who have visibility into their AI traffic see consistent patterns emerge over time.
Clusters of pages being retrieved together, repeatedly, by the same type of query.
A travel publisher finds that hotel comparison pages and packing guides are retrieved together consistently, by queries that look like active trip planning.
An automotive publisher sees EV review pages retrieved alongside range calculators and charging infrastructure guides, by users in the decision stage of a car purchase.
These aren't demographic segments inferred from browsing behaviour.
They're intent segments, built from actual purchase research, at the moment it's happening.
What the segments look like
The Audience Measure Graph assigns each cluster a funnel stage, a commercial intent signal, and a CPM estimate.
What starts as a pattern in retrieval data becomes a named audience segment.
Auto Purchase Research. Decision stage. $49 CPM.
Travel Planning. Consideration stage. $22.50 CPM.
Tech Comparison. Evaluation stage. $31 CPM.
These are the audiences advertisers spend aggressively to reach.
Publishers are generating them out of AI retrieval events and, at the moment, leaving them unread.
Why this is different from what first-party data has been before
First-party data in publishing has historically meant cookie-based behavioural data.
Which pages someone visited. How often. In what sequence.
Useful, but noisy.
You don't know if the person reading your car review is shopping for a car or procrastinating at their desk.
AI retrieval events have context that cookies don't.
The query that sent the agent reveals the intent.
A retrieval triggered by "best EV under £40k" is unambiguously commercial.
A retrieval triggered by "what car should I buy" is decision-stage intent in explicit form.
This is not audience inference. It's audience observation.
The data advantage
The publishers who start reading their AI retrieval data now will have an advantage that compounds.
They'll understand which content genuinely serves high-intent audiences, not by measuring time on page, but by seeing which pages are retrieved during active purchasing research.
They'll be able to make editorial decisions based on commercial intent, not traffic volume.
They'll have audience segments built from behavioural signals that don't exist anywhere in the standard publisher data stack.
ChatGPT now generates over 50 million UK shopping visits a month (Retail Economics, Feb 2026).
The intent behind those visits is running through publisher content right now.
The question the data is already asking
The retrieval events are happening.
The intent signals are real.
The audience is building.
Every day without the infrastructure to read it is a day of audience intelligence that can't be recovered.
The data is being written on your pages right now.
Most publishers aren't reading it.