What is AI media?

What counts as AI media

The category is defined by the interaction, not by the technology. Plenty of advertising is bought and optimised using machine learning without being AI media; what makes a surface AI media is that the user is in a conversation with a model and the advertising appears inside that exchange.

In practice it spans three groups: the general assistants, where most attention sits; AI-enhanced search, where generated answers sit above or instead of links; and the long tail of AI-native applications built on foundation models for a specific job.

How it behaves differently from search and social

Three properties separate it from the channels a media plan is usually built around.

Intent arrives complete. A search query is a fragment; a conversation has already established the constraint, the budget and the use case. Targeting works on what was said rather than what was inferred.

Supply is scarce and non-expandable. Social feeds can carry more ads by inserting more units. An assistant that inserts more advertising degrades the product it is monetising, so publishers hold the ratio deliberately low. Reach is bounded by commercially relevant conversations, not by user numbers.

The click is not the event. Users act on recommendations without clicking them. Last-click measurement systematically understates the channel, which is the single most common reason early tests are read as failures.

What it means for planning

• Do not forecast from display or social CPMs. Clearing prices are higher and volumes are far lower; a plan built on transferred assumptions will miss on both.
• Budget for incrementality testing from the start. A holdout is the only instrument that reads the channel correctly — see measuring honestly.
• Treat organic and paid as one programme. Whether an assistant recommends you unpaid is a function of GEO / AEO, and it moves the same business outcome.
• Expect the creative to be data, not artwork. What renders is generated in the assistant's voice from structured inputs.

Where the category is going

The open structural question is whether AI media consolidates into walled gardens — a few surfaces owning inventory, tooling and measurement end to end — or develops into an interoperable market on shared protocols such as AdCP. Both futures have working examples today. Our comparison of the major assistants covers where each currently sits.