What is ad inventory?

How inventory is described

A publisher's inventory is not one number. It is a set of segments, because a buyer is not purchasing 'impressions' but a particular kind of exposure.

The usual dimensions are format, placement position, device, audience and context. The same page can carry premium inventory above the fold and near-worthless inventory below it, and pricing them identically is how publishers give away yield.

Quantity, quality, and the temptation

Web inventory is expandable. More pages, more placements per page, refreshed slots, infinite scroll — all of them create more impressions from the same audience.

All of them also dilute. Each additional unit is worth less than the last, and past a point the additional supply lowers the clearing price of everything the publisher sells. The pattern is consistent enough to be a rule: publishers who expand inventory to hit revenue targets usually raise volume and lower total revenue within a year.

The metric that reveals it is revenue per user or per session rather than total impressions.

Why AI inventory is different

It cannot be expanded at all. The publisher does not create the inventory — users create it by asking questions — and adding more advertising per conversation does not produce more supply, it produces a worse product.

Supply is therefore bounded by two things the publisher only partly controls: how many conversations happen, and what proportion of them reach a commercially relevant moment. Eligibility rules set the second, and loosening them is the one lever available. It is also the lever that damages the asset.

This is why AI inventory clears high and why prompt inventory is described by intent rather than by placement — there is no position on a page to sell.