Audience retention measures how much of a video people actually watch — shown as average view duration and, more usefully, as a curve mapping exactly where viewers leave.
Reading the curve
- The opening cliff: a steep drop in the first seconds means the intro broke the thumbnail's promise.
- Mid-video dips: the precise moments interest died — cut those segments next time.
- Bumps: rewatched sections; make more of whatever that was.
Why algorithms trust it
Views can be bought and clicked away from; minutes watched cannot. Retention is the closest thing recommendation systems have to a satisfaction sensor, which is why it powers watch time and, through it, distribution.