Quick Answer: What Do People Actually Learn from YouTube?
Mostly code and data - and increasingly, AI. In 453 learning paths built on LearnPath as of June 2026, Python leads at about 24% of paths, with AI skills as a group close behind. 90% of learners start as beginners, the ideal teaching video is 14 minutes, and the best tutorials are around 3.5 years old.
The rest of this report breaks down the full dataset: 453 learning paths, 1,936 curated video slots, 1,023 distinct videos, and 604 YouTube channels - plus the uncomfortable numbers about what people actually remember afterwards. (This is an updated snapshot; our first cut of this report covered 333 paths in June 2026, and the numbers below replace it.)
Where This Data Comes From
LearnPath builds structured learning paths out of YouTube videos: a learner says what they want to learn and at what level, AI designs a curriculum, searches YouTube, ranks candidate videos against the curriculum, and assembles a step-by-step path with a quiz after each video, generated from that video's transcript.
That process leaves behind an unusual dataset. Every path records what someone chose to learn, their self-declared starting level, the goal they typed in, which videos were selected from which channels, how long those videos are, when they were published - and, because every video has a transcript-based quiz, whether the learner could actually answer questions about what they just watched.
This report covers every successfully built, non-deleted learning path on the platform as of June 30, 2026:
| Dataset | Count |
|---|---|
| Learning paths analyzed | 453 |
| Learners represented | 381 |
| Curated video slots | 1,936 |
| Distinct YouTube videos | 1,023 |
| Distinct YouTube channels | 604 |
| Submitted quizzes | 138 |
It is a young dataset from one platform, and our audience skews technical - the caveats section near the end spells out exactly what that means for interpretation. Every number in this report is a real aggregate from production data, not an estimate.
The Skills People Choose: Python Still Leads, AI Right Behind
The single most popular skill is no surprise. Python appears in 109 of 453 paths - about 24% of everything people set out to learn.
| Rank | Topic | Paths | Share |
|---|---|---|---|
| 1 | Python | 109 | 24% |
| 2 | SQL and databases | 54 | 12% |
| 3 | Machine Learning | 34 | 8% |
| 4 | Cybersecurity | 30 | 7% |
| 5 | System Design | 24 | 5% |
| 6 | DevOps | 15 | 3% |
| 7 | AI Engineering | 14 | 3% |
| 8 | JavaScript | 10 | 2% |
| 9 | AI and Automation | 10 | 2% |
| 10 | React | 7 | 2% |
Three observations from the full list:
Python, SQL, and machine learning are more than 40% of everything. Together they account for 43% of all learning paths. The "learn to code" era has consolidated into a "learn to work with data and AI" era, and these three are its entry points.
AI as a category is right behind Python. Add up the AI-flavored topics - machine learning, AI and automation, AI agents, AI engineering, AI-assisted coding tools, LLM training, prompt engineering - and you get about 92 paths, or 20% of the total. That is close behind Python's 24%, but the gap is real: in our June snapshot the two were roughly tied, and Python has since edged back ahead. In 2026, "I want to learn AI" is nearly as common a starting point as "I want to learn to code."
Tech dominates, but not exclusively. Well over 90% of paths are technical (programming, data, AI, DevOps, security). The remainder is a genuinely mixed bag: music, video editing, SEO, soft skills. People do use YouTube to learn everything - but when they want enough structure to build a path, it is overwhelmingly for career-shaped technical skills.
Almost Everyone Starts as a Beginner - and Aims High
The level data is the clearest single finding in the report:
- 90% of learners declare themselves beginners in the topic they choose (409 of 453 paths).
- 39% of all paths go from beginner straight to advanced - the most ambitious jump the platform offers.
- Only 2 paths out of 453 were started by someone who called themselves advanced.
Two things seem true at once. First, YouTube learning is overwhelmingly a beginner's activity - people with experience either search for one-off answers or read documentation, but they rarely build a structured path. Second, beginners are not modest: nearly 4 in 10 want to go from zero to advanced in one path. The typed goals show the same energy - "become an ML engineer," "land a junior data analyst role within 6 months," "pass system design interviews."
The time they give themselves is more sober than the ambition: learners set a median budget of just 3 hours a week. That mismatch - zero-to-advanced ambition on a few hours a week - is exactly the gap a structured path needs to manage. The distance between "I am at zero" and "I want a job doing this" is where most self-learning quietly dies.
The Ideal Teaching Video Is 14 Minutes, Not 4 Hours
If you picture "learning from YouTube" as grinding through a 10-hour full-course video, the data disagrees. Across all 1,936 curated video slots:
- The median video is 14 minutes long.
- Half of all picks fall between 9 and 21 minutes.
- Only about 4% are longer than an hour.
- 30% are under 10 minutes.
Remember how these videos are chosen: an AI ranks real YouTube candidates against a specific curriculum step, using transcripts. When the unit of learning is "one concept, then a quiz," the winning format is a focused 10-20 minute explanation - not a monolithic course. Long courses bundle sequencing, explanation, and pacing into one take-it-or-leave-it package. A path does the sequencing itself, so each slot just needs the clearest single explanation available.
A practical implication for self-learners: stop defaulting to the longest, most "complete" video in the search results. Completeness is the path's job. Clarity is the video's job.
The Best Tutorials Are Years Old
The recency numbers surprised us most:
- The median curated video was published about 3.5 years ago.
- 64% of selected videos with a known publish date are more than two years old.
- Only 18% were published in the last 12 months.
These videos were selected by a ranking process that could have chosen anything on YouTube, including videos uploaded last week. It mostly did not. For fundamentals - Python syntax, SQL joins, how neural networks work, design patterns - a five-year-old explanation by a great teacher beats a three-week-old explanation by an average one, and the ranking reflects that.
The exception is fast-moving tooling. Paths about AI agents, AI-assisted coding, and specific frameworks pull much newer videos, because last year's interface genuinely no longer exists. The rule of thumb the data suggests: judge fundamentals by clarity, judge tooling by date.
Half the Videos Come from Channels You Have Never Heard Of
The most-selected channels read like a who's who of YouTube education: Bro Code, Tech With Tim, codebasics, Programming with Mosh, freeCodeCamp, IBM Technology, NetworkChuck, Fireship, Alex The Analyst, Web Dev Simplified.
But the concentration is lower than we expected:
- The top 10 channels supply about 30% of all curated videos.
- The top 25 channels supply 45%.
- The remaining 55% comes from 580+ channels, most of which appear in only one or two paths. Roughly two-thirds of all channels in the dataset show up exactly once.
This is the strongest argument in the dataset for curation as a distinct job. The famous mega-channels are genuinely excellent - they earn their share of the picks. But for any specific step of any specific path, the best explanation is, more than half the time, on a channel with a fraction of the subscribers, which you would never find by browsing the front page. Subscriber count predicts production quality; it only loosely predicts whether this video teaches this concept best.
If you are picking videos by hand, that long tail is where the search cost lives - and it is the part our best YouTube channels guides can only partially cover, because they rank channels, not individual explanations.
The Uncomfortable Part: Watching Is Not Learning
Every video in a path is followed by a quiz generated from that video's transcript - questions about what the video itself taught, taken minutes after watching. These are motivated learners, tested on material they chose, immediately after consuming it. The results:
- About 4 in 10 quizzes are failed. Of 138 submitted quizzes, the pass rate is 59%.
- The average score is 69%, the median 75%.
- On first attempts specifically, the average is 65%, and nearly half of first attempts score below 70%.
This is the illusion of competence, measured in production: watching a clear explanation feels like understanding, and a few minutes later, a third of it is gone. Decades of cognitive science predicted exactly this - retrieval practice research, going back to Roediger and Karpicke's 2006 testing-effect studies, shows that being tested on material beats re-watching it for retention. Our data just shows how big the gap is for video specifically.
And no, shorter videos do not fix it. We checked whether the length of a video predicts how well people score on its quiz. It does not: pass rates were essentially flat across sub-10-minute videos (64%), 10-to-30-minute videos (59%), and longer ones (60%), and the correlation between video length and quiz score was effectively zero. Length is not the lever. Testing is.
The follow-through numbers point the same way. Completion in our data clusters almost entirely at the very start of a path - the hardest video in a learning journey is rarely the advanced one at the end, it is the one that requires coming back a second time. That matches what we found when we examined why chat-generated study plans collapse: the plan is never the problem; the follow-through is.
What This Means If You Learn from YouTube
Five takeaways the data supports directly:
- Pick 10-20 minute videos over multi-hour courses. The best-fit explanation for a single concept is almost never the longest video. Sequence short videos yourself - or use a tool that does.
- Do not filter by upload date for fundamentals. A several-year-old video with a great explanation outranks this month's upload most of the time in our data. Save the recency filter for fast-moving tools.
- Search beyond the channels you know. More than half the best-fit videos come from small channels. Search for the concept, not the creator.
- Test yourself after every video, immediately. Motivated learners fail about 4 in 10 quizzes on material they just watched. If you cannot answer questions about a video, you have not learned it yet - re-watch the section you missed, not the whole thing.
- Plan for video #2, not video #20. Drop-off happens at the second session, not the tenth. Whatever gets you to come back tomorrow - a streak, a calendar block, an unfinished quiz - matters more than the perfect curriculum.
If you would rather have all five handled for you - the curation, the sequencing, the quizzes, the coming-back part - that is literally what LearnPath does, and it is free to start: type a topic, get a structured path of curated videos with a quiz after each one. You can also browse ready-made paths by topic and clone one in a click.
Methodology and Caveats
What was measured. All aggregates were computed on June 30, 2026 from LearnPath production data: every successfully built, non-deleted learning path (n=453), the videos curated into them (1,936 slots; 1,023 distinct videos after deduplication across paths), and all submitted quizzes attached to those paths (n=138). Topic shares use the platform's canonical topic mapping, with obvious variants merged (for example "SQL" with "SQL and databases," "cybersecurity" with "cyber security," "DevOps" with "DevOps and Docker"). Video age is measured from YouTube publish date to June 2026, and could only be computed for the 660 slots (about a third) that carry a recorded publish date. No individual user data appears in this report; every number is an aggregate.
Honest caveats. This is a young dataset from a single platform. 453 paths and 381 learners is enough to see clear patterns but not enough for fine-grained slicing, which is why we report medians and shares rather than decimals of precision. Our audience skews technical, so the tech-heavy share describes structured YouTube learning by our users, not all of YouTube. Quiz results describe learners who submitted quizzes - drop-off means the true retention picture is, if anything, worse than reported. Video selections reflect our AI ranking process as well as learner choices; a different curator would produce overlapping but not identical picks.
Citing this report. Feel free to cite or quote any statistic with attribution: "LearnPath, What People Actually Learn from YouTube (June 2026 update), learnwithpath.com". We re-run these numbers as the dataset grows; this page states its data snapshot date at the top whenever it is updated.
Frequently Asked Questions
What do people learn most from YouTube in 2026?
Python is the single most popular skill, chosen for about 24% of learning paths in our June 2026 dataset. AI skills as a group - machine learning, AI agents, AI-assisted coding, prompt engineering - are close behind at roughly 20%. Python, SQL, and machine learning together account for more than 40% of all paths.
How long is the ideal YouTube tutorial?
Shorter than most people expect. Across 1,936 videos curated into learning paths, the median video is 14 minutes, and half of all picks fall between 9 and 21 minutes. Only about 4% are longer than an hour. Focused, single-concept videos beat multi-hour courses for step-by-step learning.
Are newer YouTube tutorials better for learning?
Usually not. The median video selected into a learning path was published about 3.5 years ago, and 64% of all curated videos with a known publish date are more than two years old. Only 18% were published in the last 12 months. For fundamentals, a clear explanation ages well - recency matters mainly for fast-moving tools.
How many videos does it take to learn a skill from YouTube?
A focused learning stage in our data is about 4 videos and 1 to 1.5 hours of watch time - one coherent step, not a whole journey. Learning a full skill takes several such stages with practice and testing between them, rather than one 40-hour course consumed passively.
Do people actually remember what they watch on YouTube?
Less than they think. In our dataset, learners took quizzes immediately after watching a video they had just chosen to study - and still failed about 4 out of 10 of them. The average score was 69%. Watching feels like learning, but without testing, much of it does not stick.
Which YouTube channels do people learn from the most?
Big educators dominate the top - channels like Bro Code, Tech With Tim, Programming with Mosh, freeCodeCamp, and NetworkChuck appear in dozens of paths. But concentration is lower than expected - the top 10 channels supply only about 30% of curated videos. More than half come from a long tail of 580+ smaller channels.