The home page shows 1.41M+ installs, 15,936+ ratings and an average rating of 4.55, dated 13 September 2026. Each is built from public store pages in a way anyone can repeat. This post explains the choices; the arithmetic is on the figures page.
Stores publish ranges, so the sum is a floor
Google Play does not show an install count. It shows a range: 1M+, 100K+, 10K+. We take the lower bound of each range across the six listings under our developer account and add them: 1,000,000 + 100,000 × 4 + 10,000 = 1,410,000. The true number is somewhere above that, which is what the plus sign means. Reporting the top of each range, or a midpoint, would be a guess dressed as a count.
The App Store is not counted
Apple publishes no install figure for an app. PDF AI's iOS installs are therefore not in the 1.41M+ at all. The listing's ratings are counted, because those are shown.
Third-party estimates are not used
Analytics sites publish install estimates for Play apps. We could not verify how any of them is derived, so none appears on this site, whether larger or smaller than our floor.
Ratings are summed; the average is weighted
The ratings figure is the number of ratings shown on the seven listings, six on Google Play and one on the App Store, added up. It carries a plus sign because ratings only accrue between counts. The average is each listing's average weighted by its number of ratings, so a listing with 12,946 ratings counts for more than one with 153. An average can fall as well as rise, so it carries no plus sign.
The latency figure is someone else's
The fourth number, 488 ms down from 876 ms, is Google DeepMind's measurement of one step of Cue, not ours. It is quoted with its conditions on the figures page and discussed in What the DeepMind case study says about Cue.
When the numbers change
The date beside the figures is the date of the last count. When they are recounted, the home page, the figures page and the corrections page change together.