How long should a short be? The useful answer is not a number you copy from a blog — it is the length at which the one idea you cut is still watched to the end. Platform limits set the ceiling, not the target, and the real question behind 'what length' is 'how long will my audience actually watch'. This article gives you the rule for finding your length from the source and from your retention data, and the honest split between the choice you make and the clip work a machine can do.
The platform cap is a ceiling, not a target
YouTube Shorts used to be limited to 60 seconds; since late 2024 it accepts vertical videos up to three minutes, and anything three minutes or under counts as a Short. TikTok allows longer clips still, and Instagram Reels sits close to YouTube's limit. Those numbers are easy to look up, and that is exactly why they are the wrong place to start: the platform limits only answer how long a clip is allowed to be, which is not the question anyone is really asking.
3 minMaximum length of a YouTube Short since September 2024 (it was 60 seconds before). Source: the Wikipedia page on YouTube Shorts.
The question hiding under 'how long should a short be' is 'how long will people actually watch mine'. A cap is the outer edge of the system; it tells you nothing about your audience. A clip that runs 2:55 because the platform lets you reach 3:00 is not twice as valuable as a 20-second clip that is fully watched and re-watched — usually it is the reverse. Treat the ceiling as a boundary you almost never need to touch, and start the decision somewhere else entirely.
A finished short beats a longer one that loses people
The thing that decides whether a short gets pushed is not its length but whether people finish it. Recommendation systems on YouTube and TikTok weigh early engagement and watch-through: a viewer who stays to the end signals that the clip is worth showing to more people; a viewer who leaves early signals the opposite. Completion is measured as a proportion, and that quietly changes the arithmetic of length.
Take two versions of the same idea. One runs 60 seconds and holds people for 35 of them — under 60 percent watched. The other cuts the same idea to 25 seconds and nearly everyone who starts finishes it. The platform reads the second as the stronger video even though it is shorter than half the first, because completion is a ratio and the short clip's ratio is high. That is why fixating on a 'good' number of seconds is backwards: what you are really trying to maximise is the share of viewers who reach the end, and the fastest way to hurt that share is to make a clip longer than its idea.
This is not a licence to make everything ten seconds long. A clip that is too short for its content leaves the payoff unearned, and viewers feel a clip that stops before the point. The mistake is never 'too short' on its own; it is a clip that outlives the idea holding it together.
Length comes from one complete idea, not from a stopwatch
When you cut a clip from a long video, the source already divides itself into natural units: one claim, one answer, one story beat, one reaction worth keeping. A good clip is one of those units, and its length is whatever that unit takes — not a number you impose and then stretch footage to fit. If the source is an interview or a talk, you find the unit by reading the transcript and marking where one complete idea starts and where it ends; the same reading that tells you which moments are worth clipping also tells you how long each clip should be.
Here is the practical test. Take the moment you chose and ask: does the clip still make sense, to a stranger, if it ends right here? If it does, you have found the natural end, and anything you add after it is padding. The clip's length becomes a consequence of the idea, not a decision you take in advance. You are not choosing between 'short' and 'long'; you are choosing which complete idea to cut and letting its boundaries decide the duration.
A useful habit when a clip feels too long is to check whether it is actually two ideas glued together. Clips that run on rarely do so because one idea is long; they run on because a second idea got bolted onto the first. Cut at the seam and you get two tighter clips instead of one rambling one — which is usually what 'make it shorter' meant all along.
Read watch-through to find the length your audience finishes
Length is only partly a creative decision; the rest is feedback, and the feedback lives in your retention report. The audience-retention graph shows the percentage of viewers still watching at each second, and the shape of that curve tells you whether a clip is the right length for the people it reached.
Look at where the line drops. If it stays high and only falls off in the last second or two, the clip is close to the right length for that audience — nearly everyone who started reached the end, which is exactly the signal you want. If it falls away in the middle, you have one of two problems: the clip is longer than the idea can hold, or the middle contains dead time (a pause, a false start, a rambling aside carried over from the source) that you should have cut. A mid-clip cliff is never a vote for 'post shorter clips'; it is a report that this clip had a stretch that did nothing, and the fix is to remove it, not to re-upload the same content shorter.
The larger point about why clips stall at a few hundred views is that early engagement decides reach, and retention is the biggest part of early engagement. So the length question and the reach question are the same question: a clip whose watch-through holds gets shown to more people, while a clip whose line dies in the middle never reaches the audience that might have liked it. Read the curve, find the drop, and cut to just before it.
Test length like a scientist: same source, several lengths
The honest truth is that the ideal length is not a universal number — it depends on your topic, your audience and the platform. A cooking tip and a ten-minute explainer cut into shorts will not hold people for the same time even if both are excellent. No rule of thumb replaces data, and the data you need is specific to you: which length does your audience finish?
The way to answer that cheaply is to stop deciding length once and instead test it as a batch. Cut the same source into clips that deliver the same idea at different lengths — a tight 15 seconds, a fuller 30, a longer 60 — and post them. You are comparing length alone, so everything else that would muddy the comparison has to stay identical: the same opening moment, the same caption style, the same framing. Then read which length holds watch-through and make more of those.
This is where a clip production line changes the economics of the test. Making three variants used to mean framing, captioning and exporting three clips by hand — real hours, so nobody ran the test. The honest boundary is the one drawn in the case for and against AI clip tools: the judgement of which moments and which lengths are worth trying stays yours, but the repetitive pass of turning each chosen moment into a finished, consistently captioned clip is exactly what ClipFinish's clip production line is built to remove. It does not tell you what length to test — that is your hypothesis about your audience — it simply makes shipping the three variants cheap enough that the test is worth running at all.
When the right length is shorter than you think
Much of the 'how long should shorts be' worry dissolves once you accept that most clips should be shorter than the creator first feels comfortable with. Cut footage carries the rhythm of the source video, and source videos are rarely tight: a podcast has pauses and tangents, an interview has throat-clearing, a talk has asides. None of that belongs in a short. When you trim a cold start, a rambling middle and a tail that repeats the point, the clip that remains is almost always shorter — and stronger — than the version you first marked.
Three ways to cut length honestly:
- Remove the cold start: open on the first line of the idea, not on the 'so, anyway' that preceded it.
- Cut the middle that does nothing: any pause, aside or repetition that survives from the source is dead time.
- End at the point: stop the moment the idea is delivered, before any sign-off or summary tail.
- What is the best length for a short?
- As long as the one complete idea takes and watch-through still holds. There is no fixed best number; it changes with your topic and audience, so test a few lengths from the same source.
- Does a short's length affect whether it gets views?
- Indirectly, through watch-through. Platforms weigh early engagement and completion, so a clip that many people finish tends to reach more viewers than one that loses them early, whatever its length.
- How long can a YouTube Short actually be?
- Up to three minutes since September 2024; before that the limit was 60 seconds. Reaching the cap is not the goal — the goal is a clip that people finish.
- Should all my clips be the same length?
- No. Different ideas and audiences hold for different times. Only keep length-identical when you are deliberately testing, so the comparison stays clean.
- My retention drops halfway through. Is my clip too long?
- Probably the middle is dead time, or the clip is two ideas stuck together. Cut the stretch that does nothing or split it, rather than simply re-uploading it shorter.
A short should be as long as the one complete idea it carries takes — cut so it ends when the idea is delivered, then keep it only as long as your audience keeps watching. Start by removing the cold start, the dead middle and the tail, set each clip's length by the idea it holds, read your watch-through curve and cut to just before the drop. When you are unsure of the number for your audience, test it: cut the same source into a few lengths, keep the style identical, and ship the batch to see which one people finish. Choosing which length to try is a judgement only you can make. The repetitive framing, captioning and exporting of each candidate is what ClipFinish handles, so running the test stops costing you a day of editing.