A number that shouldn't need explaining
Publishers talk constantly about CPM.
Header bidding configurations. Floor prices. SSP strategy. Yield optimisation.
Entire teams exist to move the needle from $4.20 to $4.80.
$15.02 is the average CPM from AI agent monetisation in live publisher accounts on our network.
That number is sitting on traffic they were previously earning $0 from.
For context
Display advertising across most publisher verticals runs between $3 and $8 CPM.
Premium programmatic, direct-sold inventory, can reach $20 to $30 in the best-performing categories.
The AI agent CPM of $15.02 sits at the top end of programmatic.
From traffic that currently generates nothing.
To be clear about what that means: publishers who have integrated are earning more per AI agent request than most of them earn per human page view.
Why the CPM is high
The CPM isn't high because AI agents are particularly valuable as impressions.
It's high because of what they represent.
Live Search Agents are dispatched in real time, in direct response to a user question.
When one visits your automotive pages, someone is asking about buying a car. Right now.
When it reads your travel content, someone is planning a trip. Right now.
The content an agent retrieves is tightly correlated with the intent behind the query that sent it.
Most digital advertising is bought against audience proxies: cookie segments, contextual categories, demographic estimates.
AI agent traffic is the intent itself, expressed at the moment of the query.
That's an unusual commercial signal. Advertisers pay for it.
The optimisation problem that distracts from this
Publishers have spent years working to improve their programmatic CPM by fractions.
Testing header bidding configurations.
Adding SSP partners to increase auction competition.
Testing floor prices across segments.
All to move from $4.20 to $4.50.
The AI agent CPM benchmark sits at $15.02.
From traffic they're currently earning nothing from.
And that traffic is growing. Personal AI agent traffic grew 15x in 2025 alone (Cloudflare).
51% of internet traffic is now automated, projected to reach 90% by 2030 (IAB Tech Lab).
The trajectory matters more than the current number
The $2 billion annual loss figure from IAB Tech Lab describes what has already happened.
The CPM benchmark describes what's available now.
Those are different conversations.
The loss figure requires industry coordination, licensing negotiation, and in some cases legislation.
The CPM benchmark requires different infrastructure: one that operates in the same layer as the agents themselves, before JavaScript fires, inside the content they actually read.
Publishers who establish their AI agent CPM baseline now will have a benchmark to optimise against as traffic grows.
The publishers who don't are optimising a shrinking human audience while an unmonetised one builds on the other side.
The CPM is already there.
The question is whether the infrastructure is.