11 September 2026 · 5 min read
What makes a clip work on its own
The difference between a good moment and a good clip is whether it survives being taken out of the conversation. Here is what that actually requires.
Every long conversation contains moments that feel significant while you are in it and land flat the moment you take them out. Understanding why is most of the skill in clipping.
A clip is not an excerpt
An excerpt is a piece of something larger, and it reads like one. A clip is a small complete thing. The test is simple and unforgiving: play it to somebody who has not heard the episode and watch their face in the first three seconds.
Most moments fail that test for one of three reasons. They refer back to something said earlier. They answer a question nobody in the clip asked. Or they depend on knowing who is speaking and why that person saying it is surprising.
Self-contained means it carries its own setup
The strongest clips tend to include their own question. Somebody says "the thing people get wrong about this is" and the setup arrives with the payoff, wrapped together.
This is why direct answers to a host question often make poor clips even when the answer is excellent. The question was four minutes and one topic change ago. Inside the clip it is invisible.
You can sometimes fix this by moving the start earlier to include the question. If the question is short, do it. If including it means starting ninety seconds back, the moment is an excerpt and you should let it go.
A turn is what keeps people watching
Self-contained is necessary, not sufficient. A complete, coherent forty seconds where nothing changes is a complete, coherent forty seconds nobody watches.
What holds attention is a turn: an expectation set and then bent. A concession before a claim. A number that is smaller or larger than it should be. Someone disagreeing with themselves halfway through a sentence.
Setup, turn, landing. If you cannot point at all three in a passage, it is information rather than a clip.
The best clips are rarely the funniest thirty seconds
This surprises people. Scanning an episode for the biggest laugh usually produces clips that are funny in context and puzzling out of it, because humour leans hardest on shared setup.
Whereas a passage where somebody explains something they clearly believe, completely, in under a minute, tends to travel. Not because it is more entertaining, but because it needs nothing from outside itself.
What this means for scoring
When Keptbits scores candidate moments, the heaviest weight is on whether a passage opens and resolves inside its own boundaries — not on how emphatic it is or how many keywords it contains. A passage that needs the ten minutes before it scores near zero no matter how good it sounds.
That is also why it sometimes returns fewer clips than you asked for. Given the choice between handing back a filler clip and saying nothing cleared the bar, the second is more useful, if less satisfying.
Keptbits cuts on sentence boundaries.
Paste a link or upload an episode, say how many clips you want, and get them back captioned and vertical. Seven days free on any plan.
Start your free trialRead next
The first three seconds decide the rest
Almost every clip that fails, fails at the start. What the opening has to do, why most auto-generated openings do not do it, and how to fix one without re-editing.
How long should a clip be?
The honest answer is that length is a consequence, not a setting. What the number should actually be decided by, and why the popular advice is backwards.
Clipping a conversation with two people in it
Interviews produce the best clips and are the hardest to cut. Where the boundaries go, what to do about interruptions, and why the question usually belongs in the clip.