Glossary
Retention curve
A retention curve plots the percentage of viewers still watching at each second of a clip. A flat or slightly-declining curve from second 0 to the end is the goal; sharp drop-offs in the middle indicate places where the clip lost the viewer's attention.
Part of the AI Video Clipping topic cluster.
What a healthy retention curve looks like
On a 60-second Short, the median creator on Clipperz sees about 60–75% retention at second 30 and 35–50% at second 60. Above those numbers, the recommender treats the clip as engaging and surfaces it more aggressively. Below those numbers, distribution slows.
What matters most is the shape, not the absolute number. A curve that drops 50% in the first 3 seconds and then stabilizes is a hook problem. A curve that holds for 30 seconds and then collapses is a length or pacing problem.
How AI clippers use retention prediction
Clipperz scores predicted retention before a clip is rendered by simulating how the segment would play out: language model assessment of pacing, sentence-boundary density, and emotional intensity over time. Clips with predicted curves that drop below 40% midway are usually rejected from the recommendations and marked for human review.
What to do when retention drops mid-clip
- Trim the clip earlier — most retention drops are caused by the segment running past the natural payoff.
- Switch to a more dynamic visual (B-roll, caption animation) at the drop point.
- Re-cut the clip to a stronger hook, then the same content reads as 'why is this still worth watching' instead of 'when does this get to the point'.
Related terms
Hook strength
How compelling the first 1–2 seconds of a clip are at making a viewer keep watching.
Satisfaction-per-Impression
Clipperz's composite metric for ranking clip candidates by likely outcome rather than raw segment availability.
Clip score
A per-clip ranking number representing predicted performance, used to sort candidate clips from a single source video.
See it in action
Paste a video URL into Clipperz and watch the concept play out on your own content.
Try free →
