Impression rank percentile - where an ad sits in its own brand's delivery

Impression rank percentile is an ad's position by delivery inside its own brand's ads. It is not a number Meta publishes: the Ad Library returns per-ad impressions only for ads about social issues, elections or politics, and only as a range, so a rank like this can only be an estimate derived from public Ad Library signals: how long the ad has run, how many variants of it are live, where it is placed, how persistently it reappears. It is also not one of Meta's own rankings, and it should not be read as one: Meta's ad relevance diagnostics compare an ad against other advertisers competing for the same audience and land an ad in one of five bands (above average, average, or below average at the bottom 35%, 20% or 10%), where this number compares an ad against its own brand's other ads and reads from the top down. We weight it highest anyway, because it is the only signal that is about what actually got served rather than about what the advertiser decided to do.

Why the Curve Is Not Linear

The distance between top 1% and top 15% is a different kind of gap than between top 50% and top 64%, and a straight inversion priced them the same. Admetra raises the inverted percentile to a power so that being at the very top of a brand's delivery is worth far more than being merely above average.

Worked Example

Three ads from the same brand. One sits at top 1%, one at top 15%, one at top 50%. A straight inversion would score them 99, 85 and 50. Squaring the headroom scores them 98, 72 and 25. The ad at the very top of its brand's delivery pulls nearly four times the score of the merely above-average one, instead of twice.

Rank Alone Is Not Direction

A high rank today does not say whether an ad is climbing or fading, and raw direction is misleading on its own. An ad can climb ten places while sitting far below the peak it once held. Admetra scores trajectory as current rank measured against the ad's own all-time best percentile, which separates an ad at its peak from a dead-cat bounce.