The shape of a YouTube video: 3,179 value curves overlaid

Averaged across 3,179 measured videos, the value of a YouTube video is not spread evenly across its runtime. It climbs fast in the first fifth, holds nearly flat through the middle two-thirds, and falls off a cliff in the last few percent.

That is one number standing in for a shape, and the shape is the actual finding. Nobody publishes this curve because almost nobody measures a video densely enough to draw it. We score value continuously across a video’s runtime for a different reason — to tell you where to start watching — and averaging the result over thousands of videos turns out to draw a picture of how explanatory video is built.

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 census again it will be published as a new edition with a new date.

The curve climbs fast, and most of the climb is over by the 20% mark

Every video’s runtime is normalized to 0–100%, regardless of how long it actually runs, and value is sampled at twenty-one evenly spaced points along it.

By the fifth of the way through a video, the average curve has completed 90% of its climb from start to plateau. It opens at 23.7, and by the 20% mark it is already at 61.6 — within four points of where it will spend most of the runtime. On an eight-minute video that is roughly the first minute and a half; on a thirty-minute video, roughly the first six.

This is not really a story about introductions being short or long. It is a story about how quickly a video, once it starts, gets to a level it holds — and the answer is: quickly, on average, relative to everything that comes after.

The middle two-thirds barely move

From the 20% mark to the 85% mark — well over half the runtime of a typical video — the curve moves inside a four-point band.

Sixty percentage points of runtime, four points of value. The plateau runs from 61.6 up to a peak of 65.8 at the 55% mark and back down to 60.5 at 85%, and nothing in between strays far from that range. If a video’s value were a single number, this stretch is where that number would come from.

That flatness is itself informative. It says the videos in this library are not typically front-loaded or back-loaded internally — once they get going, they hold a fairly constant level of value until they are ready to stop. The interesting motion is entirely at the two ends.

The last 5% of runtime accounts for nearly half the entire fall

The curve does not decline gently into its ending. It holds until the 85% mark, and then drops.

Value falls from 60.5 at 85% to 21.8 at 100% — and 42% of that entire decline happens in the final 5% of runtime alone, the step from 40.3 down to 21.8. On a thirty-minute video that last stretch is under a minute and a half; on an eight-minute video, under thirty seconds. It is a small slice of the runtime carrying a disproportionate share of the weakness.

The video ends almost exactly as low as it began: 21.8 at the close against 23.7 at the open, a gap of under two points. A video’s two weakest moments, on average, are its first few seconds and its last few — which matches what those stretches usually are: an open before there is anything to deliver, and a close after there no longer is.

The shape holds across kinds of video, at different heights

Splitting the curve by the kind of video it belongs to does not change the shape. It changes how high the plateau sits.

Educational and hands-on videos plateau in the mid-sixties; mixed-format videos plateau roughly six points lower and their final stretch falls less steeply, ending at 42.3 rather than the high 30s. All three still open low, climb, hold, and fall — the same three-part shape, at three different altitudes.

We are not claiming that a lower plateau makes mixed-format video worse. It means the material inside it varies more from minute to minute by design, so a single averaged curve for the category sits closer to its own middle than a more uniform kind of video does.

What the shape is not telling you

It is not a claim about any one video. This is 3,179 individual curves averaged together, and any single video can look nothing like it — start strong, end strong, or wander. The shaded band around the line is the honest version of that: at the 95% mark the middle half of all measured videos scores anywhere from 15 to 62, which is a wide spread on a 0–100 scale.

It is not a verdict on introductions or endings. Some of what happens in the first and last few percent of a runtime is doing real work — framing what is about to be covered, or wrapping it up so a viewer who followed along can leave with a clean sense of the whole. The curve says those stretches score lower on average, not that they should be cut.

It is not the same measurement as padding. The padding census counts a different thing — restatement and setup, wherever in the runtime it falls. This curve is about where value sits along a video’s timeline, at every point, not about how much of it is filler.

How this was measured, and what it does not say

Each video’s runtime is scored across its full length, and that per-second reading is what produces a curve rather than a single figure. For this report, every video’s own curve is stretched or compressed onto a common 0–100% scale — position 50% means the midpoint of that particular video’s own runtime, whatever its length — and sampled at twenty-one points, then averaged across the library. Method and its limits, including how the underlying score is built, are in what value density means.

64 of the 3,243 measured videos are not in this curve. A curve needs a runtime the platform has confirmed, and 64 measured videos do not currently have one. The remaining 3,179 are the basis for every figure on this page.

Sampling at twenty-one points is a summary, not the full curve. A video’s actual value can move between the sampled points in ways the chart does not show; twenty-one points is enough to see the shape without pretending to show every second of every video.

The sample is not YouTube. It is 3,179 English-language explanatory videos across twenty subjects, assembled to build a recommendation library rather than to be representative of the platform.

The figures are frozen at 26 August 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 census will be published separately.