Every article on this subject opens with a number, and none of them can tell you where the number came from. What is actually true is smaller and more useful: a shorter clip is more likely to be finished and re-watched, a longer one holds somebody for more total time, and the feeds care about both. The right length is the one where your point lands and the clip stops. Everything after that is measured as people leaving.
Limits as of September 2026. They move, and each platform publishes its own: YouTube, Instagram, TikTok.
Note what the caps are not: a target. Three minutes is the point at which YouTube stops calling it a Short, not a length anybody is asking you to fill.
Two things get measured and they pull in opposite directions.
Completion and repeat. A twenty-second clip that people watch to the end - and, on a feed that loops, straight into again - produces a very strong signal, because almost nobody leaves early and the watch time is counted twice. This is why short clips punch above their length.
Total time held. A ninety-second clip that keeps people to the end holds each person four times as long as that twenty-second one. When it works, it works harder. When it does not, it fails in public: the drop-off happens in the middle and is measured there.
So the question is never "what is the best length". It is how long can this particular material hold somebody, and the answer is a property of the material, not of the format. A single line with a punchline has twenty seconds in it. A story with a setup and a payoff has ninety. Padding the first to ninety and cutting the second to twenty both destroy the thing that was working.
Three shapes, and the length each one wants.
A single strong statement and the sentence that proves it. It is over when the proof lands, which is usually well under half a minute.
Setup, turn, payoff. It cannot be cut short without losing the payoff, and the setup is what earns the time. This is the shape that survives past a minute.
Each item buys a little more attention, and it can stop at any of them. Cut it where the items stop being good, not where the list ends.
The practical test is the same for all three: watch your own clip and mark the second where you would have swiped if it were somebody else's. That second is the end of the clip. Almost everything sitting after it - a wind-down, a thank you, the next question being asked - is costing you the completion the whole thing was built on.
SnapDub cuts on that principle rather than to a fixed duration: it finds where a self contained moment starts and ends in a long recording, so each clip runs the length its own material justifies instead of a number set in advance. You can set a length if you want one, and you can move both ends afterwards.
If the clips are the right length and still going nowhere, length was not the problem - why a short gets no views covers what else it could be, and how many clips one recording gives you covers where the next ten come from.
Start free with 500 credits a month - no card required. Drop in a long recording and see where the clips fall.
Start clipping freeAs long as the material holds and no longer. Short clips get finished and re-watched, which is a strong signal; longer ones hold each person for more total time, which is also a strong signal. Neither wins in general, so the length is decided by where your point lands.
Only if the thing you are showing is done in 30. A padded 30-second clip loses to a 60-second one that earns its time, and a 60-second clip that made its point at 25 is 35 seconds of people leaving.
Three minutes for both, as of September 2026. Past that, YouTube stops treating the upload as a Short and it leaves the Shorts feed; on Instagram it simply stops being a Reel. TikTok's limit is much higher and is rarely the real constraint.
The platforms count repeat watching as watch time, which is part of why a short clip that ends cleanly and starts again does well. It is not a trick to exploit with an abrupt cut - a clip that ends confusingly gets abandoned rather than re-watched.
No. A clip that ends mid-thought reads as broken, and people leave rather than re-watch. If it is too long, the fix is cutting the setup or the tail, not the point.
Not for being longer. It gets less reach if people stop watching it, which is more likely the longer it runs. Length is a risk, not a penalty.