How Long Are Short-Form Clips, Really? Data From 177 AI-Picked Clips
Most advice about clip length is a number somebody liked. This page is different in one way: every figure on it comes from real clips that an AI actually selected from real videos, and the sample size sits next to every number so you can decide how much weight it deserves.
The sample. 177 clips from 50 finished AI Clipper jobs on EzClip, cut from 32.2 hours of source video between August 8 and September 3, 2026. Our own test account is excluded. Shorts made with the Video to Short tool — where the creator picks the segment — are counted separately in the last section, because they answer a different question. It is a small sample. We will re-run the numbers monthly and update this page; the date at the top tells you which run you are reading.
The median AI-picked clip is 61 seconds
When the AI is free to cut where the moment ends rather than pad to a target, this is where it lands:
| Percentile | Clip length |
|---|---|
| 10th | 40 s |
| 25th | 47 s |
| Median | 61 s |
| Mean | 66.7 s |
| 75th | 78 s |
| 90th | 98 s |
Bucketed, the shape is clearer:
| Clip length | Share of clips |
|---|---|
| 30–45 s | 19.8% |
| 45–60 s | 28.2% |
| 60–90 s | 33.9% |
| Over 90 s | 18.1% |
One rule shapes the bottom of that table, so read it correctly. EzClip will not return a spoken clip shorter than 30 seconds — when a moment is shorter, the AI is told to extend it with the surrounding context that keeps it coherent rather than cut it off mid-thought. The absence of clips under 30 seconds is therefore a rule, not a finding. Above that floor there is no target length: the instruction is to cut at the natural beat and never pad. What the AI chooses within that freedom is the finding — 62% of clips land between 45 and 90 seconds, and nearly one in five runs past 90.
The practical reading: the moments that score highest for engagement are rarely sound-bites. A hook, the turn, and the payoff take about a minute to say.
Nearly half of the best moments come from the first third of the video
We recorded where each clip started as a fraction of the source video's length:
| Position in the source | Share of clips |
|---|---|
| First third | 48.6% |
| Middle third | 31.6% |
| Last third | 19.8% |
Two explanations, and both are probably true. Creators front-load: the strongest material in a podcast or an interview tends to be near the top, after the pleasantries and before the energy dips. And openings cut cleanly: a moment that starts a conversation carries its own context, while a moment from the middle of one often needs a run-up that drags the clip over the length where it works.
If you only have time to clip one thing from an episode, start at the top.
A video yields about 3.5 clips, or 5.5 clips per hour of footage
| Value | |
|---|---|
| Median source length | 22 minutes (25th–75th percentile: 9 to 61 minutes) |
| Median clips per video | 3.5 |
| 90th percentile clips per video | 5 |
| Clips per hour of source | 5.5 |
That 5.5-per-hour figure is the number to plan a calendar around: a weekly one-hour show produces roughly five or six posts a week from the AI alone, before any manual cuts. Custom-prompt jobs — where the creator tells the AI what kind of moment to look for — ran shorter (median 40 seconds), but there were only ten such clips, so treat that as a hint rather than a result.
When creators choose the length themselves, they go longer
The Video to Short tool asks the creator to pick the segment, capped at two minutes. Across 26 shorts made that way in the same window:
| Creator-chosen shorts | |
|---|---|
| Median length | 63 s |
| 75th percentile | 103 s |
| Share between 90 and 120 s | 42% |
The median is almost identical to the AI's 61 seconds — but the top end is not. More than four in ten creator-chosen shorts ran between 90 seconds and the cap, where only 18% of AI-picked clips did. Left to themselves, creators cut long. Whether that is right depends on the platform: YouTube Shorts and Reels reward retention more than raw length, and a 100-second short that people leave at 40 seconds performs like a 40-second clip with a bad completion rate. The AI's habit of stopping at the beat is worth copying at least some of the time.
How to use this
- Aim for 45–90 seconds, then cut at the beat. That is where the AI lands most often when it is free to choose, and it matches how much room a hook, a turn and a payoff actually need.
- Clip the opening first. Half of the strongest moments in this sample came from the first third of the video.
- Plan on about five clips per hour of source. More than that usually means posting the AI's weaker candidates.
- If you cut by hand, check your completion rate before defaulting to the cap. The two-minute short is the most common creator choice and the least common AI choice.
Method and caveats
The sample is every finished AI Clipper job on EzClip between August 8 and September 3, 2026, excluding our own test account and any clip row with a nonsensical duration (under 3 seconds or over 10 minutes — none were found). Clip length is the difference between each clip's start and end time in the source. Position is the clip's start time divided by the source duration. "Creator-chosen shorts" are Video to Short renders in the same window. Content types are self-reported by the creator at upload; almost every job in this window used the automatic preset, which is why the breakdown by content type is not shown yet. Nothing here identifies a creator, a video or a channel, and no figure on this page was estimated, modelled or rounded beyond one decimal place.