A chapter list tells you where things are, not which one is worth watching

A chapter list gives you three facts about each section of a video: what it is called, when it starts, and how long it runs. None of the three is a measurement of what is inside it — and when we measure the sections of 3,243 videos ourselves, the sections that a chapter list would treat as interchangeable turn out to differ by an average of 34 points.

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.

Almost all of a video is “the video”

We read every measured video in sections — 46,105 of them across the library, a median of thirteen per video — and label what each one is doing. Here is how the runtime divides.

86.9% of all measured runtime is the body of the video. Everything a viewer would recognise as a skippable structural block — the opening, the closing, the sponsor read, the ask to subscribe, the digression — comes to 13.1% between them, and the median video gives 13.8% of its runtime to those blocks.

Now hold that against the number this site publishes most often: the median measured video carries 40% padding, the share of runtime that restates, recaps or sets up rather than adds.

Those two figures are not in conflict; they are the finding. If only 13% of a video is structurally separable from its body, and 40% of it does not add, then most of the padding is inside the body, in sections that no boundary marks and no chapter list distinguishes from the sections that do the work. A chapter list is a map of where things are. It is not a map of what is worth anything.

The named blocks are exactly as weak as you assume — and they are small

There is a satisfying result in the second chart, and it is worth stating before the disappointing one.

The sections that look like filler measure like filler. A sponsor read averages 11.2 out of 100, an ask to subscribe 13.6, a closing 19.1, an opening 20.9. The body of a video averages 64.0, three times higher than anything else on that list. Intuition about which parts of a video are worth skipping is, at the level of named blocks, correct.

It just does not buy very much, because those blocks are 13% of the runtime. Removing every sponsor read from the library — a measure people would applaud — would recover 53,192 seconds out of 3,356,077. The sponsor segment is not where a video spends your time.

Inside the body, sections differ by a third of the scale

This is where a chapter list stops helping and where the numbers are most surprising.

Take a single video, look only at the sections that are the body of it, and score each one. The weakest averages 44.9 and the strongest averages 78.9 — a gap of 34 points on a 0–100 scale, inside one video, between sections a chapter list presents as peers. The median video’s own range is 33 points, so this is not an average dragged around by a few extreme cases.

Put differently: the average video’s own internal range is wider than the range across the whole library’s subjects. The best-scoring subject in the library averages 66.9 and the worst 38.4 — a gap of 28.5 points, against 34.0 inside a typical single video. Choosing well inside a video matters at least as much as choosing between videos, and a chapter list gives you nothing to choose with.

Length is a weak signal, and it is right about a quarter of the time

The one quantitative fact a chapter list does give you is how long each section runs. So we checked whether it works: is the longest section of a video also its best?

It is, in 728 of 3,138 videos — 23.2%. Those videos carry a median of eleven body sections each, so picking one at random would land on the best about nine percent of the time; length beats that comfortably and is genuinely better than nothing. It is also wrong three times in four, and on average the longest section scores 10.8 points below the best one in the same video.

The direction is at least consistent: in 1,778 videos, longer sections do tend to score higher, against 216 where the relationship runs the other way. A long section is more likely to be substantive than a thirty-second one. That is a tendency to keep in mind, not a rule to navigate by.

What a chapter list is still good for

We are not arguing against chapters. They solve a real problem and they solve it well.

Returning to something you have already watched. If you know the part you want, a labelled timeline is the fastest way back to it, and nothing here changes that.

Seeing the shape of an argument before committing. Ten chapter titles tell you whether a video is a list, a build-up or a comparison. That is genuine information about structure.

Skipping the blocks that are reliably weak. The chart above says a sponsor read and a subscribe ask really are worth jumping. A chapter list that labels them lets you do it in one click.

What a chapter list cannot do is tell you which of the ten body sections is the one you came for — because the label was written by the video about itself, and the answer varies by 34 points.

What this measurement does not say

It is not a measurement of anybody’s chapter markers. We measure our own reading of a video’s sections. We do not compare it against the chapters a creator wrote, and nothing here says a particular chapter list is inaccurate. The claim is narrower and harder to argue with: a title, a start time and a length do not encode value, whoever wrote them.

A weak section is not a wasted one. A section that scores low can be doing necessary work — establishing a definition the next ten minutes depend on, or giving somebody without the background a way in. Our number says how much a section adds for a viewer who already knows why they clicked.

It is not accuracy. We measure how much a section says, not whether what it says is true.

How this was measured, and what it does not say

Each measured video is read in sections; each section carries a label for what it is doing and a value score on the same 0–100 scale used everywhere on this site. The section boundaries are ours. Method and its limits are in what value density means.

The spread figures use fewer videos than the masthead. A within-video range needs several body sections to mean anything, so the 34-point gap is computed over the 3,199 videos with at least four, and the length comparison over the 3,138 with at least five. Videos read as one or two long sections are excluded rather than counted as having no variation.

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.

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.