Two issues ago we told you the score is a shortlist, not a ranking. This week we finally had enough published posts to test that claim against real view counts instead of asserting it. The answer is worse than we expected, and we'd rather you hear it from us.
What we are not claiming
That clip scoring is useless. Forty-five posts from one small, new account is a thin sample, views on a new account are driven mostly by platform distribution rather than by the clip, and the same clip posted to three places appears three times in that number.
What it does show is narrower and still worth acting on: at our scale, the score did not rank clips the way the platforms did. If you have been choosing which clip to post by the highest number, our own data does not support that choice.
What to do instead, this week
- Use the score to cut the bottom half, then choose from the rest by what the clip is actually about.
- Post the clip that starts at 0:00 even when it scored last — the hook lives there because creators put it there.
- Publish the same clip to all three platforms before you conclude anything about it. Ours varied from 0 to 90 views for the same file.
- Judge a clip after a week, not a day. Our 1,517-view clip did nothing for its first two days.
Why we published a number that makes our own feature look bad
Because the alternative is letting you plan a workflow around a promise we cannot support. The score does one job well — ranking hook-like structure inside the window you hand it — and we would rather be precise about which job that is than be flattering about it.
The full breakdown, with the query behind every figure, is at clipember.com/guides/do-ai-clip-scores-predict-views. We will update it when the dataset is meaningfully larger, and say what changed.
Two of these a week, to everyone with a ClipEmber account. Make a clip or read the rest of the archive.