The best YouTube channels for learning, by subject: measured video by video (2026)

La nostra libreria è curata, non un campione casuale di YouTube. La riempiamo di proposito con video didattici di qualità per l'app Gistil, quindi qui l'oro è più frequente che su YouTube nel suo insieme. Questi dati descrivono la nostra libreria, non tutto YouTube.

A note on the numbers. Every score on this page is the one the channel’s page on gistil.com shows, taken from one snapshot of the library on 19 October 2026. Channel pages are recalculated as we measure more videos, and this report is recalculated from the same data, so the two agree. The first edition of this report, measured 15 October 2026, is frozen at that date.

Most lists of educational YouTube channels are a person’s favourites. They are useful, and they share a blind spot: they rate the channel by its best video, the one that made the list-maker subscribe. We rate a channel by all of its videos we have measured, and we use the median, so one viral hit or one weak upload cannot move the result.

YouTube is where a lot of people already go to learn. 84% of American adults say they use it ⁵, and in a 2018 Pew survey half of its US users said it was very important for figuring out how to do new things ¹. The problem is not finding a channel. It is knowing which channel will keep paying back the time you give it.

How the ranking works

Every video below was scored by Gistil from 0 to 100 for how much it teaches: depth, accuracy, what is new, what you can use. The score comes from the full spoken content of the video, not from its title, thumbnail or view count. A video at 70 or above is gold, 40 to 69 is silver, below 40 is copper.

A channel’s score is the median of its measured learning videos, the same number its channel page on gistil.com shows. Channels with at least five measured videos are ranked. Two more numbers sit beside the score:

• Videos that reach gold: the share of the channel’s measured videos scoring 70 or more, also as on its page.
• Useful part: the median share of a video’s runtime that falls in the part worth playing.

The last one tells you how a channel spends your time. Two channels can have the same score while one gives you a dense twenty minutes and the other a good hour spread over two.

The highest-scoring channels

The top of the table is explanation-first channels. 3Blue1Brown scores 90, and 94% of its measured videos reach gold; 77% of a typical video’s runtime is in its useful part. PBS Space Time scores 80 with 100% of its videos gold. Veritasium scores 81 and is gold in 86% of its videos.

The strongest channels share one habit: they explain a mechanism. A machine part by part, an argument built from sources, why something happened rather than a retelling of what did.

Andrej Karpathy is the outlier on length. His videos run for about 2.1 h, and 100% of those we count here are gold: each step builds on the one before it, so little of the length is padding. Two of his longest from-scratch coding builds are classed as tutorials, which channel pages leave out (see the methodology), so his score of 90 rests on his explainers, talks and the remaining builds.

The best channels in each subject

Each channel name opens its page on gistil.com, with its three best videos to start with and how its videos usually spend their time. In a few subjects even a top-three channel has no gold video among those we measured; the subject still has gold videos, they sit in other channels. For the best single videos rather than channels, the measured shelves on /best are sorted by subject and topic and open without an account.

Subjects without a gold channel

Science, engineering, AI & machine learning, news & politics, history, programming, health & fitness, languages, nutrition, business, interview podcasts, finance and psychology each have at least one gold channel. Design & UX, productivity and learning and tech news do not: their best channels score below 70.

Part of this is our scale. It rewards depth and explanation, and some subjects are mostly something else: one sound rule with examples, advice for a particular situation, or a walkthrough of a tool. All of these can help, and all of them score lower on a scale built for depth. In such subjects the gold videos are mostly inside good silver channels, so look for single videos rather than for a gold channel.

What the top channels do differently

We split the ranked channels into three groups by their median score and compared them.

Four things differ between the groups. The outer groups are smaller than the middle one, so read the contrast as a direction rather than a precise measurement.

• They waste less of each video. In a typical video from a gold channel, 66% of the runtime is in the useful part. In channels under 55 it is 32%.
• Their titles keep their promises. The median title honesty score is 87 for gold channels and 72 for channels under 55.
• They are more consistent. The gap between the best and worst video is about 19 points in a typical gold channel and 39 points under 55.
• Length is not the main difference. The typical video runs about 18 minutes in a gold channel, 16 in the middle group and 11 in the weakest group.

Subscribing is a weaker signal than it looks

A subscription is a bet that the next video will be as good as the one that made you subscribe. For many channels the spread of their videos makes that bet weaker than it feels.

Among the ranked channels, the median gap between a channel’s best and worst video is 23 points. 30 points is the distance between a strong gold video and a middling silver one, and 32% of channels have a gap that wide or wider.

Even gold channels are not gold every time. Across the channels whose median is gold, on average 22% of the videos fall below 70, and for the least consistent of them it is 43%. In the other direction, 72% of silver channels have at least one gold video, and 22% of ranked channels have at least one copper video.

Our earlier guide on how to evaluate a channel before you subscribe lists the signals worth checking: whether the good video is typical of the channel, whether new videos are new ideas or the same idea with a new hook, and whether the titles across the channel hold up.

A smooth presenter is not one of them. People rate a fluent, confident lecturer as more effective and believe they learned more, while measured learning is the same ². The same goes for classrooms: students in active lessons learn more and feel they learned less ⁶. A channel that feels effortless to watch is worth checking against what it actually teaches.

How to use a channel well

1. Start a new channel with its best videos, not its newest. For most channels the range of quality is wide, so the newest upload may not be typical. Every channel page on gistil.com lists three videos to start with.
2. Match the channel to how you learn. A channel with most of its runtime in the useful part rewards watching start to finish. A channel at 40% rewards jumping to the strong parts.
3. Use conference channels for depth, in segments. Conference talks run close to an hour. Jump to the part of the talk that matches your question.
4. In long interviews, look for the strong stretches. Dwarkesh Patel and Peter Attia have some of the strongest long conversations in the library, and even there the best parts are stretches inside a long episode.
5. Watch actively. Pausing to predict what comes next, or to explain the last step back to yourself, helps you learn more from any channel ⁴. Videos that highlight what matters on screen also teach better ³. Our guide on how to learn from YouTube has the method.

The channel pages linked above come from the same measurements Gistil uses everywhere. In Chrome and Firefox, the extension shows the score, the metal and the useful minutes of a single video beside the player, for any video already measured or one you ask it to analyze. The Gistil app for iPhone and Android turns channels like these into feeds by topic, with gold videos first.

Methodology and limits

• Data. Gistil’s measurement of explanatory YouTube videos in English, as of 19 October 2026. Hidden videos and test entries are excluded.
• Score. Each video is scored 0 to 100 for what it teaches by one AI model, checked on a reference set against three AI judges (different models) following a written rubric, with no human panel. The channel score is the median of its measured videos, computed the way the channel’s page on gistil.com computes it, and the score and gold share in this report are the ones those pages showed on 19 October 2026. A channel page counts all of the channel’s measured videos, including those not yet re-measured with the current scoring model.
• Which videos count. As on the channel pages: lessons, explainers, talks and news. Hands-on tutorials and craft walkthroughs are left out of channel scores, and so are subjects we do not cover yet (music, comedy, gaming, travel, DIY). That is why a few strong channels are missing here: maker and DIY channels such as NightHawkInLight, and channels whose best videos are mostly tutorials. The useful-part share and typical length are computed from the same videos.
• Eligibility. Channels with at least five measured videos that count. Channels we have not measured enough of are absent, not ranked low; a channel page exists from three measured videos, so some channels have a page but no place in this ranking yet.
• Subjects are Gistil’s categories, assigned by the videos we measured. A channel sits in the subject most of its measured videos fall into, which puts a few channels in surprising places.
• Not a popularity ranking. Subscriber counts, views and likes play no part.
• A snapshot. The library leans toward channels people already use to learn. More caveats are in where our numbers do not apply.

Gistil Research. Cite as: Gistil, “The best YouTube channels for learning, by subject”, October 2026, gistil.com.