The eight-minute pile-up: where runtimes cluster, and what it costs

Sort the library by runtime and count how many videos land in each half-minute, and one band stands out: the half-minute that begins at 8:00 holds 107 measured videos, more than any other half-minute in the first thirty minutes of runtime. The two bands immediately before it hold 55 and 64.

That is the finding people expect. The second finding is the one we did not expect, and it is the reason this report exists rather than a headline: the videos in the pile are not measurably worse than the videos on either side of it.

A note on the numbers. Unlike the rest of this site, the figures on this page are frozen at the date above rather than recalculated on each visit. A report people cite should not move under them. When we run this measurement again it will be published as a new edition with a new date.

The pile is real and it is sharp

Half-minute bands, in runtime order, from five minutes to twelve.

The shape of that chart is not a gentle hump. Runtime counts fall steadily from 5:30 to a trough of 55 videos at 7:00, then jump to 107 at 8:00 and stay elevated. At ten-second resolution the step is sharper still: the three bands covering 8:00 to 8:29 hold 36, 35 and 36 videos, against an average of 20.9 across the eight ten-second bands covering 6:40 to 7:59.

A video in this library is more likely to run for eight minutes and change than for any other single stretch under half an hour. There is a smaller second bump at 10:00 — 86 videos, against 69 and 56 in the bands on either side — which is what you would expect if round numbers attract runtimes generally, but the eight-minute step is larger and it arrives after a trough rather than on a plateau.

We are describing a distribution, not a motive. Eight minutes is a threshold that appears constantly in advice given to people making videos, and the obvious reading is that some videos are being built to clear it. We do not measure how a video is monetised and cannot confirm that reading from our own data. What we can show is the shape.

The pile is not worse than its neighbours

If videos were being stretched to reach eight minutes, the stretched material would have to come from somewhere — and the thing we measure is exactly that: the share of runtime that restates, recaps or sets up rather than adds. So the prediction is easy to state and easy to check. Videos just over eight minutes should carry more padding than videos just under.

They do not. The 8:00–8:59 band carries 43.0% padding. The band immediately below it carries 42.5%, and the band below that carries 46.0%. Mean value across the three bands moves by less than a point. On the measurement that would have to move if a video were being padded out to a target, nothing moves.

There is one number in that table that leans the other way. The share of videos in the 8:00–8:59 band we would tell you to skip is 25.8%, against 20.2% just below and 21.7% just above, and against 20.3% for the library as a whole. That is a five-point gap in the direction the theory predicts.

We are not going to sell it to you as a result. It rests on 190 videos and 49 skip verdicts; the sampling error on a proportion that size is around three points, so a five-point gap against neighbours that are themselves noisy is suggestive at best. If the effect is real it is small, and this library is not large enough to prove it is there.

What actually grows with runtime is the wind-up

The measurement that does move across this range is not padding. It is how long a video takes to start.

The opening stretch of a video grows from 23.5 seconds under eight minutes to 35.8 seconds in the eight-to-ten-minute band — half again as long, for a runtime that grew by a quarter. The point at which the video stops setting up and starts delivering moves further still, from a mean of 41.8 seconds to 64.5 seconds.

That is a real cost and it is worth naming precisely, because it is smaller than it sounds. Twenty-three extra seconds of front matter is not a stolen afternoon. It is, though, the part of a longer video you pay first and the part you cannot skip without knowing where it ends — which is the whole reason we publish that number next to the video rather than making you discover it by scrubbing.

Why this does not contradict the length result

We have published a result that sounds like the opposite of this one: longer videos carry less padding, not more, falling from 45.0% under ten minutes to 27.6% over an hour.

Both are true, because they are answers to different questions. Across the whole range of runtimes, length is a symptom of a subject that genuinely takes time, and the long tail of the library is where the dense material lives. Inside one narrow band — the two minutes on either side of eight — length is not doing that work, and the padding figures sit flat.

The honest summary of the two together: runtime tells you something at the scale of “six minutes versus an hour” and almost nothing at the scale of “seven minutes versus nine”.

What this measurement does not say

It does not identify which videos were stretched. A cluster in a distribution is a property of the library. It cannot tell you that any particular eight-minute video was built to be eight minutes long, and we make no such claim about anybody’s work.

It does not measure monetisation. We measure the video. Whether a runtime clears a threshold that matters to the person who published it is outside anything we look at.

It is not a verdict on short videos. The 6:00–6:59 band carries the highest padding in the table at 46.0% and the highest skip share at 28.9%. If there is a runtime to be wary of in this range, the data points below eight minutes rather than above it.

How this was measured, and what it does not say

Every measured video has a runtime the platform has confirmed, and every one of the 3,243 videos in the library carries one; the histogram is a straight count over those runtimes. Padding, value and the verdict are our own measurements, described in what value density means.

2,681 of the 3,243 videos run under thirty minutes, which is the range the histogram covers. The bands above that are too sparse to show a half-minute count meaningfully and are left out of the chart rather than drawn as noise.

The sample is not YouTube. It is English-language explanatory video across twenty subjects, assembled to build a recommendation library rather than to be representative of the platform. A different library — one built around a different kind of video — could easily show the pile-up somewhere else, or not at all.

The figures are frozen at 1 September 2026. Scores are recalculated as the underlying measurement improves, so the live numbers on the shelves will already have moved slightly from these. This edition will not be edited; the next measurement will be published separately.