データレポート 2026年10月15日
How much of a YouTube video is actually worth watching? We measured, minute by minute
A minute-by-minute measurement of explanatory YouTube videos (October 2026): how much of each one earns your time, where that part sits, and which signals on the page predict it.
時点 2026/10/19 読了12分
このページは英語で書かれています。翻訳されているのはこのページの周りのサイト部分のみです。
Key findings (October 2026)
- 57%of a typical explanatory video's runtime is in the part worth playing. The middle half of videos sits between 41% and 72%.
- 45%of the videos in our library reach gold, a score of 70 or more: about one in two. 51% are silver and 3% copper.
- 24%of first tenths and 28% of last tenths are in the useful part, against 68% to 70% in the middle.
- Over 2 hoursis the length that is gold most often (61%). Videos under 10 minutes are skipped most often (18%).
- 49%of videos whose title scored below 60 for honesty got a skip verdict. At 90 and above, 2%.
このライブラリは厳選されたもので、YouTube全体からの無作為な抽出ではありません。Gistilアプリ向けに質の高い教育動画を意図的に集めているため、ゴールドの割合はYouTube全体よりも高くなっています。これらの数値はあくまで当ライブラリのものであり、YouTube全体を表すものではありません。
A note on the numbers. The figures on this page come from one snapshot of the library, taken on 19 October 2026. They count only the videos measured with our current scoring model (from 16 October 2026 on, 73% of the library on that date), so every number rests on one scale. The first edition of this report, measured 15 October 2026 across the whole library on the earlier scale (gold 20%, a median useful share of 42%), is frozen at that date. These numbers replace the ones in our earlier posts, which used an older version of the scale and an older definition of the useful part (for example, when videos get to the point and sponsor segments).
People ask whether a video is worth watching one video at a time, usually with the play button under their cursor. We asked the same question about a whole library at once. Gistil scores every stretch of a video for how much it teaches, so we can say, minute by minute, which part of a video earns the time it asks for. This is the picture as of 19 October 2026, after the scoring was recalibrated against a reference set judged by independent AI judges.
The library is explanatory video: science, programming, history, health, finance, languages, design, long-form interviews and the rest of what people open YouTube to learn from. It is a curated set, not a random sample of YouTube, and the limitations section says what that changes.
The short answer
Take any video in the library and ask what share of its runtime falls in the part worth playing. The median answer is 57%. Half the videos land between 41% and 72%.
The spread matters as much as the middle. One video in ten is under 32% useful. One in ten is over 83%.
Counted by minutes rather than by videos, 55% of all measured runtime is in the useful part.
“Useful part” has a precise meaning here. Gistil scores each segment of a video from 0 to 100 for how much it teaches. The useful part is the set of segments the product would play if you pressed Play the useful part: the best stretches of that particular video, never below 55, never above 70 as a bar. The rest is not all filler. Most of it is ordinary context, examples and transitions that a viewer in a hurry can skip.
One metal per video
The whole video also gets one score and one metal. Gold means 70 or above, silver 40 to 69, copper below 40.
- Gold45.2%
- Silver51.5%
- Copper3.3%
One metal per video. Gold 70 and above, silver 40 to 69, copper below 40.
Table view
| Category | % of videos |
|---|---|
| Gold | 45.2% |
| Silver | 51.4% |
| Copper | 3.3% |
Gistil measurement, 19 October 2026
Gold is about one in two video in our library (45%). Silver (51%) is a video with a real point that takes longer to make it than it has to. Copper is rare (3%), and when it appears it is concentrated in a few subjects.
A silver video still has its best moments. The metal describes the video, so those moments are drawn silver too, and their brightness shows where they are. Two examples from the library make the difference visible.
Transformers, the tech behind LLMs | Deep Learning Chapter 5 · 3Blue1Brown
Measured by Gistil. The whole video scores 91, gold. 87% of its runtime is in the useful part.
The 3Blue1Brown chapter on transformers spends 87% of its runtime in the useful part. The interview below is a good conversation, and it is spread thin. Most of its 99 minutes are pleasant and relevant, and about 20 of them are what you would replay.
What happens after coding is solved? | Fiona Fung (Claude Code & Cowork) · Lenny's Podcast
Measured by Gistil. The whole video scores 63, silver, so its useful segments are silver too. Good conversation, spread thin: 20% of its runtime is in the useful part.
Where the useful part sits
Split every video into ten equal parts and ask how often each part falls in the useful part.
Each video split into ten equal parts, from its first tenth to its last.
Table view
| Position in the video | % of videos |
|---|---|
| 0–10% | 24.2% |
| 10–20% | 55.2% |
| 20–30% | 67.2% |
| 30–40% | 69.8% |
| 40–50% | 69.8% |
| 50–60% | 68.1% |
| 60–70% | 69% |
| 70–80% | 64.2% |
| 80–90% | 55.9% |
| 90–100% | 27.7% |
Gistil measurement, 19 October 2026
The share rises fast, stays flat through the middle and falls at the end. The first tenth is the weakest part of a video: only 24% of first tenths are in the useful part. By the third tenth the share is 67%, it stays between 68% and 70% through the middle, and it falls to 28% at the end.
The opening is where the video introduces itself, greets the viewer, asks for a subscription or reads a sponsor. The median video gets to its subject after 48 seconds, and reaches its first segment worth playing at 1:50. One in ten makes you wait past 6:43 for that. Length stretches the wait: in a video under ten minutes the subject arrives after a median 20 seconds, in a one-to-two-hour video after 2:25, and over two hours after 2:50.
The ending fades more slowly than the opening rises. The last fifth carries calls to action, promos and outros, and by the final tenth only 28% of videos are still in their useful part.
So skipping the first minute or two usually loses little, and jumping to the second half usually loses a lot. YouTube itself offers related shortcuts. Chapters let creators mark sections 4, and the Most replayed graph over the progress bar shows where viewers rewind 5. Neither tells you which section is worth your time. Chapters describe what a part is about, and replays measure what people went back to, which is not the same as what taught them most.
Does length matter?
Length matters less than subject.
Bands in runtime order.
Table view
| Runtime | % of the band |
|---|---|
| Under 10 min | 34.7% |
| 10–20 min | 46.5% |
| 20–40 min | 52.5% |
| 40–60 min | 49.4% |
| 1–2 hours | 36.7% |
| Over 2 hours | 61.3% |
Gistil measurement, 19 October 2026
Under ten minutes, 34.7% reach gold and 18.3% get a skip verdict. A short video has little room for depth.
Videos over two hours are gold most often, 61.3%. The share of the runtime in the useful part is 41% for videos of one to two hours and 57% for videos over two hours. Interviews and long lectures wander, and the good parts are separated by long, decent stretches.
The useful minutes still rise with length.
Median minutes in the useful part, by runtime band.
Table view
| Runtime | minutes |
|---|---|
| Under 10 min | 3.5 |
| 10–20 min | 8.1 |
| 20–40 min | 16.4 |
| 40–60 min | 26.2 |
| 1–2 hours | 34.2 |
| Over 2 hours | 81.2 |
Gistil measurement, 19 October 2026
A short video gives a median 3.5 useful minutes. A one-to-two-hour video gives about 34, and one over two hours about 81. If you have an evening and a topic, the long video is usually the better deal, as long as you know where its good parts are. Our earlier analysis of padding by runtime found the same thing about minutes on an older version of the scale (runtime is a useless filter). The share figures of the two editions are not directly comparable, because the scale changed.
Interview podcasts are a special case. As a group they reach gold 32% of the time, are skipped 17% of the time and take a median 1:54 to get to the subject. Good episodes exist, and some of the strongest long videos in the library are interviews, but the format rewards rapport over density.
Which subjects earn the most of your time
Subject predicts the metal better than length does.
Music is at the top: 75% of its videos reach gold. The subjects near the top are the ones where a video exists to explain one mechanism, and the explanation is the content.
At the bottom of the chart, Design & UX reaches gold in 29% of cases. Low on this scale is a statement about what a video teaches, not about whether it is pleasant to watch. In advice-style subjects the useful idea is often real and short: one rule, stated early, followed by examples of the same rule. As a lesson that scores silver.
Subjects with fewer than 50 videos measured on the current scale are left out of the chart.
The title predicts the skip
Gistil compares what a title promises with what the video delivers and scores the match from 0 to 100. That one number is the strongest single predictor of a skip in the data.
Title honesty compares the promise in the title with what the video delivers, 0 to 100.
Table view
| Title honesty score | % of the band |
|---|---|
| Below 60 | 48.9% |
| 60–74 | 20.3% |
| 75–89 | 6.4% |
| 90 and above | 2.2% |
Gistil measurement, 19 October 2026
When the title scores below 60, the video gets a skip verdict 49% of the time and reaches gold in 9% of cases. When the title scores 90 or above, 2% are skips and 72% are gold. The honest-title videos also get to their subject sooner: a median 32 seconds against 66.
You can check this yourself without any tool. A title that promises a secret, a number of things “nobody tells you”, or a result in a time frame is making a claim the video has to pay off. Our earlier note on title honesty across subjects has the per-subject view (do YouTube titles tell the truth?).
The other strong signal is level. Videos pitched as advanced reach gold 70% of the time, against 37% for beginner material. Advanced videos are also rare, and we wrote about why.
Sponsor reads take little of the time
45% of the measured videos carry at least one sponsor or self-promotion segment. On average these take 3.2% of the runtime, about 29 seconds in a fifteen-minute video. Tools like SponsorBlock, where viewers mark sponsor segments by hand, deal with exactly this part 6. The time cost of a weak video sits elsewhere: in the slow opening, the long middle stretches at 40 to 55, and the recap at the end.
What to do with this
- Skip the first minute or two, not the second half. The useful part rarely starts at 0:00 and rarely sits in the last tenth.
- Prefer one longer video on a subject you care about over three short ones. A longer video gives more useful minutes per decision.
- Read the title as a claim. If it oversells, expect a skip.
- Treat a long interview as a source of a few great stretches, not as a lecture. Find the stretches and watch those.
- Trust the subject. A science explainer and a productivity video of the same length are not the same bet.
Research on lecture video adds a different lesson: how you watch matters as much as what. In a study of 6.9 million viewing sessions on edX, median engagement topped out at about six minutes, whatever the length of the video 1. That is an argument for watching a long video in segments rather than for picking a short one. Learners do better when a video is broken into segments they control 2, testing yourself beats watching again 3, short quizzes inside an online lecture halved mind wandering 9, and a smooth, confident presenter makes people feel they learned more without them learning more 8. Our guide to learning from YouTube turns these findings into a method.
Every number in this study is something Gistil shows for a single video. In Chrome and Firefox, beside the YouTube player, it shows the verdict and the metal, a map of where the useful minutes are, and a check of the title against what the video delivers, for any video it has measured or that you ask it to analyze. One button plays only the useful part. The Gistil app for iPhone and Android uses the same scores to build feeds on the topics you choose, without copper videos. Measured videos open free, with no account.
Methodology
What was measured. Explanatory YouTube videos in English, scored by Gistil and visible on the site on 19 October 2026. This edition counts only the videos measured with the current scoring model, from 16 October 2026 on (73% of the library on that date). Videos still on the earlier scale are left out until they are re-measured, so the two scales are never mixed in one number. Hidden videos, test entries and videos without a verified runtime were left out. The library grows by subject: channels and videos people learn from, plus every video a Gistil user asked for. It is not a random sample of YouTube.
How a video is scored. Gistil reads the full spoken content of the video with its timing and scores each segment from 0 to 100 for what it teaches: depth, accuracy, how much is new and how much can be used. The video’s score is not an average of the segments. It is a judgement of the whole, from which the metal follows: gold 70 and above, silver 40 to 69, copper below 40. The library is scored by an AI model calibrated to one scale. Details are in how we measure.
How we checked it. In September and October 2026 we compared the scores with a reference set of videos judged independently by three AI judges, each a different model, following a written rubric. The three judges agreed with each other on the metal in about 84% of videos, and the scoring model agreed with them in roughly four videos out of five. One judge belongs to the same model family that scores the library, and there was no human panel. Scores of individual videos move by a few points between runs, so a video at 69 or 71 sits on a boundary.
Definitions used here. - Useful part: the segments Play the useful part would play: for each video, the stretches at the top of that video’s own range, with a floor of 55 and a bar no higher than 70. - Skip, skim, watch: Gistil’s verdict. Skip means the video is not worth its runtime as a lesson. - Interview podcasts: the long-form interview shows we added to the library as a group. They have no subject label of their own, so they are reported as one group. - Title honesty: how well the promise in the title matches what the video delivers, 0 to 100. - Percentages are of videos unless the chart says minutes.
Limitations
- Curated, not random. The library leans toward channels people already use to learn, and YouTube is a common place to learn: in a 2018 Pew survey of US adults, half of YouTube users said it was very important for figuring out how to do new things 7. A random sample of YouTube would almost certainly score lower.
- English only. Videos in other languages are scored on the same scale, but there are too few to report.
- “Teaches” is the yardstick. Comedy, music and travel are scored as lessons here. A low score means the video teaches little, not that it is bad.
- One model, checked by models. Scores come from one AI model, checked against other AI models on a reference set, not from human raters.
- One pass, several readings. Title honesty, level, verdict and score come from the same reading of the video, so the link between an honest title and a good score is partly built into the method, not fully independent evidence.
- Some entertainment is included. Comedy, music and gaming are a small part of the library and are in the overall figures.
- A snapshot. These figures are from 19 October 2026. They will move as the library grows and is re-measured. The first edition (15 October 2026) is frozen at its date. Our earlier editions used an older version of the scale, so their numbers are not directly comparable: see the padding census and one in five.
Sources
- How video production affects student engagement: an empirical study of MOOC videos Guo, Kim, Rubin — ACM Learning @ Scale, 2014
- A meta-analysis of the segmenting effect Rey, Beege, Nebel et al. — Educational Psychology Review, 2019
- Test-enhanced learning: taking memory tests improves long-term retention Roediger, Karpicke — Psychological Science, 2006
- Video chapters YouTube Help, checked 15 October 2026
- YouTube video player updates: Most Replayed, video chapters, single loop and more TeamYouTube, YouTube Help Community, 18 May 2022
- SponsorBlock: crowdsourced skipping of sponsor segments SponsorBlock, checked 15 October 2026
- Many turn to YouTube for children's content, news, how-to lessons Pew Research Center, 7 November 2018
- Appearances can be deceiving: instructor fluency increases perceptions of learning without increasing actual learning Carpenter, Wilford, Kornell, Mullaney — Psychonomic Bulletin & Review, 2013
- Interpolated memory tests reduce mind wandering and improve learning of online lectures Szpunar, Khan, Schacter — PNAS, 2013
Gistil Research. Cite as: Gistil, “How much of a YouTube video is actually worth watching?”, October 2026, gistil.com.
よくある質問
How much of a typical YouTube video is worth watching?
In Gistil's measurement of explanatory YouTube videos (19 October 2026), the median video spends 57% of its runtime in the part worth playing. For the middle half of videos that share runs from 41% to 72%. One video in ten is under 32%, and one in ten is over 83%.
Where is the useful part of a YouTube video?
Mostly in the middle. Across the measured library only 24% of the first tenth of a video and 28% of the last tenth fall in the useful part, against 68% to 70% in the middle of the video. The median video reaches its first useful segment at 1:50.
Are longer YouTube videos worth watching less?
Not in the measured library. Videos over two hours are gold most often of any length (61%), and videos under 10 minutes are skipped most often (18%). The useful minutes grow with length, from a median 3.5 minutes in a video under 10 minutes to 81 minutes in one over two hours.
Can I tell if a video is worth watching from the title?
Partly. Gistil compares what a title promises with what the video delivers. Videos whose titles we rated below 60 for honesty were skipped 49% of the time and reached gold in 9% of cases. Videos rated 90 or above were skipped 2% of the time and gold 72% of the time.
How is this measured?
Gistil reads the full spoken content of each video and scores it segment by segment from 0 to 100 for how much it teaches. The segments a person would want to play are the useful part. The whole video gets one score and one metal: gold at 70 and above, silver from 40 to 69, copper below 40. The scale was checked against a reference set judged by three independent AI judges (different models) following a written rubric; the scoring model agreed with their metal in roughly four videos out of five. There was no human panel.