What is fill rate?

What it measures

Fill rate tells you how often a request for an ad produced one. A publisher making a million requests and serving 800,000 impressions has an 80% fill rate.

Low fill can mean several different things: floors set above what buyers will pay, inventory nobody wants, geographies with thin demand, or technical failures in the request path. The number alone does not distinguish them, which is why it is a starting point rather than a diagnosis.

Why 100% fill is a warning

The instinct is to maximise fill, and it is wrong. Fill rate can always be driven to 100% by dropping floors to zero, at which point every impression sells for almost nothing.

A publisher at 100% fill is almost certainly leaving money on the table — the floors are low enough that no bid is ever rejected, which means bids that would have cleared higher are being accepted at the floor. Some unfilled inventory is the sign of a floor doing its job.

The metric that should be optimised is eCPM or total revenue, and fill rate is one of the inputs. Raising floors reduces fill and frequently increases revenue.

Fill rate on AI surfaces

• The denominator is different. Requests are only made for eligible conversations, so the base excludes most activity by design.
• Low fill is expected and acceptable. Relevance thresholds mean many eligible moments legitimately find no suitable advertiser.
• An unfilled moment costs nothing. There is no empty slot to look at — the assistant simply answers, which is the neutral outcome.
• Revenue per query is the better metric, since it captures both fill and eligibility — see revenue per query.