Satisfaction-per-Impression

Most AI clippers surface candidates as a flat list. Without a ranking signal, a creator scrolls through 30 candidates and picks subjectively — which is slow and inconsistent. SpI sorts the list so the strongest candidates are at the top, and the weakest can be archived without review.

The metric is also stable across source types. A well-scored interview clip and a well-scored monologue clip share the same SpI band, so the same threshold ("only review clips above 0.7") works regardless of what the source video looked like.

• Hook strength (opening 1.5 seconds — does it create a question or tension).
• Predicted retention curve (does engagement hold across the clip's duration).
• Payoff (does the segment resolve the question, complete a thought, or land the emotional beat).

Sort candidates by SpI descending. Start with the top 5. If the top 5 cover the topics you want to publish, you're done with selection in under 2 minutes. If you need more variety, drop the threshold to top 10 and pick 2–3 additional clips that cover topics the top 5 missed.

Don't review past the top 15 unless you specifically need 10+ clips from one source. The drop-off in SpI between rank 10 and rank 30 is usually steep enough that the time spent reviewing isn't repaid by the clips selected.