Gistil Researchデータレポート
数字で見る YouTube：データレポート
YouTube 動画が何をどれだけ教えているかを1分ごとに採点し、学習動画のライブラリ全体で数字が示すことを報告します。各レポートには日付と方法を明記し、数値はその日付のまま固定しています。
最新の調査
-
データレポート2026/10/1512分
57%of a typical explanatory video's runtime is the part worth playing.
How much of a YouTube video is actually worth watching? We measured, minute by minute
We scored every minute of a library of learning videos. The median video earns 57% of its runtime, the first and last tenth earn least, and the title predicts a skip.
レポートを読む → -
データレポート2026/10/158分
38%of ranked learning channels have a gold median score; AndrejKarpathy leads at 90.
The best YouTube channels for learning, by subject: measured video by video (2026)
The top three educational YouTube channels in each of 16 subjects, ranked by the median score of their measured videos, and why even the best channels are not gold every time.
レポートを読む →
その他のレポート
-
データレポート2026/9/17分
A chapter list tells you where things are, not which one is worth watching
Section by section across the library. The parts a chapter list would call content vary by 34 points inside a single video, and length barely predicts which is which.
-
データレポート2026/9/16分
Does a longer video teach you more? We checked, and the answer is yes
Measured depth rises by 31 points from the shortest videos to the longest, holds inside every subject we can test, and buys less usefulness than you would expect.
-
データレポート2026/9/16分
Shelf life: how long the material in a video stays true, by subject
Measured shelf life runs from 84.8 in science to 37.9 in tech news — a 47-point gap. Runtime does not predict it and neither does difficulty. Subject does.
-
データレポート2026/9/16分
The eight-minute pile-up: where runtimes cluster, and what it costs
Runtimes pile up the moment a video passes eight minutes. The pile is real and easy to see. The penalty everyone assumes comes with it is not.
-
データレポート2026/8/265分
Do YouTube titles tell the truth? Honesty measured across twenty subjects
Average title honesty is 81 out of 100, from 87 in science down to 73 in travel. The same subjects that oversell titles also carry the most padding.
-
データレポート2026/8/265分
How long is a lesson, really: runtime measured across twenty subjects
Median runtime runs from 18:16 in history to 9:30 in productivity, nearly two to one. It has almost no relationship to which subjects score highest on value.
-
データレポート2026/8/257分
We measured twenty subjects. Three percent of the videos were advanced.
Three percent of the videos we measured were pitched above intermediate. Here is which subjects have none at all, and the one signal that finds the rest.
以前の版
これらのレポートは以前のスケールを使っています。数値はそのまま残しています。最新のデータはこちら： How much of a YouTube video is actually worth watching? We measured, minute by minute.
-
データレポート2026/9/1新しいデータあり
Sponsor reads take 1.6% of the runtime. Padding takes 40%.
One in five measured videos carries a sponsor read, and it runs about sixty-seven seconds. The thing that actually spends your time is somewhere else entirely.
-
データレポート2026/8/26新しいデータあり
Which subjects are worth your time: measured value across twenty subjects
Science scores highest of twenty measured subjects at 66.9; travel lowest at 38.4. The same subjects also carry the least padding and the most honest titles.
-
データレポート2026/8/26新しいデータあり
Runtime is a useless filter: longer videos carry less padding, not more
Padding falls from 45% under ten minutes to 28% over an hour, and measured value rises the same way. Picking the shorter video is, on average, backwards.
-
データレポート2026/8/26新しいデータあり
The shape of a YouTube video: every value curve overlaid
Average measured value climbs fast, holds through the middle two-thirds of a video, then falls sharply near the end. Here is the curve, averaged across the library.
-
データレポート2026/8/25新しいデータあり
One in five measured videos was not worth the time it asked for
One in five measured videos was not worth its runtime. The share runs from 5% in science to 64% in travel, and the title predicts it before you click.
-
データレポート2026/8/25新しいデータあり
We measured the padding again, on a larger library. The number has not moved.
A second padding census, on a larger set of measured videos. The median is still 40%, the set has grown by three quarters, and the distribution barely moved.
-
データレポート2026/8/25新しいデータあり
The useful part of the median measured video starts at 1:02
Across the videos we have measured the useful part starts at 1:02, and one in ten makes you wait past 3:47. Here is how that splits across twenty subjects.
-
データレポート2026/8/17新しいデータあり
We measured YouTube across fifteen categories. This is how much of it was filler.
A padding census across fifteen categories: the median video is 40% structure, and the long ones are the least padded, not the most.