Key takeaways
- There is no length the algorithm prefers. YouTube's own position is that there is no ideal duration, and the system weights how satisfied viewers were — watch time, survey responses, likes, shares — rather than how many minutes you uploaded.
- Only three lengths are enforced by the platform itself: 3 minutes (the line that makes a vertical video a Short), 8 minutes (the minimum for mid-roll ads), and 12 hours or 256 GB (the upload ceiling). Everything else is convention.
- Metricool's 2026 study of 799,718 videos found long-form views up nearly 76% year on year while average view duration fell about 37%, from 3.98 minutes to 2.51. More people press play and fewer of them stay, which changes what an extra minute is worth.
- Watch time is a product, not a length: views × average view duration. Adding minutes only pays if those minutes hold, and a padded 12-minute video routinely produces less watch time than a tight 6-minute one.
- Stretching to 8 minutes for mid-rolls is a real trade. Mid-rolls need monetised playbacks, not runtime — if the padding drops your retention past the first ad break, you added inventory nobody reaches.
- Your correct length is not looked up, it is read off your own retention curves: find where viewers reliably leave, and stop making video past that point until the curve moves.
Ask how long a YouTube video should be and you will get a number within seconds. Seven to fifteen minutes. Ten. Eight, because that is when mid-rolls unlock. Under five, because attention spans are collapsing. Each answer is delivered with the confidence of a spec, and none of them is one.
Length is not a setting you choose and then fill. It is the output of two things you decide earlier: how big the idea actually is, and how long your audience will stay with you while you explain it. Get those right and the runtime falls out of them. Get them wrong and no duration rescues the video — a padded twelve minutes and a truncated four are the same mistake pointing in opposite directions.
This piece works through the whole question properly: what the platform genuinely enforces, what changed in 2026 about how long people stay, the arithmetic that decides whether an extra minute earns its place, what the 8-minute mid-roll line really costs, honest retention benchmarks by length, and a twenty-minute exercise that gives you your channel's own answer instead of somebody else's.
The three lengths that are real
Start with the parts that are not opinion. YouTube enforces exactly three duration thresholds, and only one of them is about money.
| Threshold | What it does | Applies to |
|---|---|---|
| 3 minutes | A vertical or square video under this length, uploaded after 15 October 2024, is classified as a Short and lives in the Shorts feed. Cross it and the same file is long-form. | Anything 9:16 or square |
| 8 minutes | Minimum total runtime for mid-roll ads. Shorter videos still carry pre-roll, post-roll and overlays; only the breaks in the middle need the eight minutes. | Monetised long-form |
| 15 minutes | The default upload cap on an unverified account. A phone number lifts it. | New channels |
| 12 hours / 256 GB | The hard upload ceiling, whichever limit you hit first. Not affected by channel size or Partner Programme status. | Verified accounts |
That is the complete list. There is no length band that gets promoted, no penalty for going long, no bonus for landing on ten minutes. YouTube has said consistently that there is no ideal video length, and the reason is structural rather than diplomatic: the recommendation system is not looking at duration at all. It is looking at what happened during and after the watch.
That last part is worth being precise about. YouTube's documentation on how recommendations work describes a system fed by several signals at once: what people watch and for how long, but also likes, shares, "not interested" feedback and satisfaction surveys where viewers are asked directly to rate a video. A long video that people finish and rate well and a short video that people finish and rate well are both good outcomes. A long video that people quit at 90 seconds is a bad one, and the runtime is not what made it bad.
What changed in 2026: more plays, less staying
The reason this question feels different now is that the underlying behaviour genuinely moved. Metricool's 2026 YouTube study analysed 799,718 videos from 71,177 accounts, comparing February 2025 with February 2026. The headline is a split: long-form views rose nearly 76% and total estimated minutes watched rose about 11%, but average view duration fell roughly 37%, from 3.98 minutes to 2.51.
Shorts moved the same direction, harder. Views climbed more than 127% while average view duration fell about 67%, to roughly 16 seconds per Short — around 32 seconds shorter than a year earlier.
Read together, those numbers describe an audience that is sampling more and committing less. More videos get opened; a smaller fraction of each one gets consumed. The study's monetisation figures follow from that directly: ad impressions on long-form fell 51%, monetised playbacks dropped 44% and estimated ad revenue declined 55%, even as views went up. If you have watched your view count hold steady while your revenue sagged, that is the shape of it, and it is not personal — see the diagnostic for a falling view count for how to tell a platform-wide drift from something you caused.
What this does and does not imply
The obvious inference — make everything shorter — is the wrong one. Falling average view duration is partly a mix effect: more casual sampling from feeds drags the average down without any individual video getting worse. What it does imply is that the opening minutes now carry more of the load than they used to, and that an extra five minutes of runtime is a weaker bet than it was two years ago. It raises the bar an added minute has to clear. It does not set a ceiling.
Watch time is a product, not a length
Here is the arithmetic that settles most length arguments. Watch time is views multiplied by average view duration. Average view duration is length multiplied by the percentage viewed. So runtime enters the equation only through a term that usually falls as runtime rises.
Work it through with numbers you can check yourself:
| Runtime | Average percentage viewed | Average view duration |
|---|---|---|
| 6 minutes | 55% | 3 min 18 s |
| 8 minutes | 40% | 3 min 12 s |
| 12 minutes | 25% | 3 min 00 s |
| 20 minutes | 40% | 8 min 00 s |
The first three rows are the same video, told at three lengths, and they deliver near-identical watch time. The stretch bought nothing. The fourth row is the case people quote as proof that long wins, and it is real — a 20-minute video that genuinely holds 40% is an enormous watch-time machine — but notice what it requires. It is not a longer version of the six-minute video. It is a different video, with enough structure to survive twenty minutes, made for an audience that already trusts the channel.
So the honest test for any added minute is narrow: does this minute hold as well as the minutes around it? If it does, length is free. If it does not, you have traded average percentage viewed for runtime at roughly one for one, and you get a flat watch time with a worse-looking retention graph — which is the version the recommendation system sees.
The 8-minute rule, and what stretching to it actually costs
The eight-minute line is the single most distorting number in this whole topic, so it deserves care. It is real: mid-roll ads require a total runtime of at least eight minutes, a threshold YouTube lowered from ten minutes in 2020. A video at 7:59 gets no mid-rolls. Below the line you still run pre-roll, post-roll and overlays; YouTube's own help page on mid-roll breaks is the place to confirm the current mechanics before you plan around them.
What creators then do with that fact is where it goes wrong. Padding a six-minute idea to 8:05 adds inventory in the middle of a video whose audience is now leaving earlier, and mid-rolls do not pay for existing. They pay when someone reaches them. Ads are counted against monetised playbacks, not runtime — the chain from an advertiser's bid to your bank account is worked through in how much YouTube pays per 1,000 views — so a break at 4:30 in a video whose viewers now median out at 3:00 is an ad slot that mostly goes unserved, bought with retention you gave away to create it.
The placement side has also changed in a way that rewards not fighting it. Since 12 May 2025, YouTube has weighted mid-roll insertion towards natural break points — pauses, transitions — and away from moments that feel interruptive, like the middle of a sentence or an action beat. Studio now flags a manually placed break as interruptive when it thinks you have put one in a bad spot. In YouTube's own write-up of the change, channels that combined automatic placement with their manual breaks saw an average 5% increase in ad revenue in early testing against manual placement alone.
The practical reading: if your idea genuinely runs past eight minutes, take the mid-rolls and let the automatic placement help. If it does not, publishing at six minutes with strong retention is very often worth more than eight minutes with weak retention, because the shorter video keeps the monetised playbacks that the longer one spends. And the deciding factor is your retention curve, not the rule.
What counts as good retention at each length
Benchmarks here are softer than they look, and it is worth saying why before quoting any. Every published figure mixes niches, traffic sources and audience temperatures, and a channel whose views come mostly from search behaves differently from one fed by Browse. Treat these as rough bands, not targets.
vidIQ's published benchmarks put a healthy average percentage viewed for videos under five minutes at roughly 50–70%, with the band sliding down as runtime rises — into the 40s and low 50s for videos in the ten-to-twenty-minute range. More useful than the bands is the first-party figure vidIQ publishes for its own channel: about 30.3% average percentage viewed across 16.5 million long-form views in the year to June 2026, with average view duration around 3 minutes 11 seconds. That is a large, well-run, established channel sitting at 30%, which should recalibrate anyone who has been told that 50% is the pass mark.
Two structural facts explain the whole benchmark table without any data at all. Shorter videos post higher percentages because the fixed cost of the opening — the seconds everyone loses to people who clicked and immediately bounced — is a bigger share of a small denominator. And the same fixed cost is why percentage viewed is nearly useless for comparing videos of different lengths on the same channel. Compare average view duration for that, and compare percentage viewed only against videos of similar runtime.
Evidence from outside YouTube
Wistia's 2026 State of Video report, drawn from more than 13 million videos on its business video platform, finds engagement declining with length in the way you would expect: videos under a minute average around a 52% engagement rate, three-to-five-minute videos around 43%, with a further drop once videos pass thirty minutes. The interesting exceptions are the ones that survive length by being requested — instructional videos in the three-to-five-minute band score far above the average for their length, and webinar replays in the 31–45 minute range outperform shorter replays.
That is business video, not YouTube, and the metrics are not identical — worth stating plainly rather than borrowing the numbers as if they transfer. But the pattern it shows is the one that matters here: length is punished when it is unrequested and tolerated when the viewer arrived wanting the whole thing. Intent, not duration, decides how much runtime you are allowed.
The 3-minute line and the Shorts cliff
The other threshold catches people by surprise, because it is a classification rule rather than a feature. A vertical or square video under three minutes, uploaded after 15 October 2024, is a Short. At three minutes and one second, the identical file is a long-form video. The change was not retroactive, so older uploads keep whatever they were.
Everything downstream of that line differs: which feed the video enters, how it is monetised, how it is discovered, and how much of the packaging work your thumbnail does. In the Shorts feed the thumbnail barely arbitrates the click at all, which is why the Shorts thumbnail rules are their own subject rather than a subset of the long-form ones. If you are cutting vertical clips near the boundary, decide which side you want deliberately. Drifting over three minutes because an edit ran long moves your video into a completely different distribution system without you choosing it.
The Metricool figure for Shorts — average view duration around 16 seconds — is also a useful sanity check on ambition. A three-minute Short is technically allowed. The behaviour in that feed says almost nobody is finishing one.
Find your own length in about twenty minutes
This is the part that replaces the guesswork, and it uses data only you have. Open Studio and work through it in order.
- List your last 15–20 long-form videos with runtime, average view duration and average percentage viewed. Not views. Views tell you about packaging; this exercise is about what happens after the click.
- Sort by average view duration, not percentage. The top of that list is your channel telling you which formats hold people. Note the runtimes there — this is the closest thing to a natural length your channel has.
- Open the retention curve on your three best and three worst. Ignore the first 30 seconds for a moment; that cliff is a packaging problem, covered in what the first 30 seconds have to do. You are looking for the point where the curve stops sloping gently and drops.
- Find your median drop-off point across those six videos. If your curves consistently break at around the 4-minute mark, that is where your content stops being worth someone's time — not at the 8-minute mark where the ads live.
- Audit the minutes past that point. Watch your own video from the drop-off. Nine times out of ten you will find the same three things: a recap of what you already said, a section that answers a question nobody asked, and an ending that took ninety seconds to arrive.
- Cut to just past the drop-off on the next three uploads and compare average view duration, not percentage viewed. If it holds or rises, the cut minutes were dead weight. If it falls, you cut something people were staying for, and you know where to look.
Two cautions on running this. Retention curves are noisy on videos with few views, so weight the ones with real traffic. And change one thing at a time — if you shorten the video and also change format and thumbnail style in the same week, you have learnt nothing about length.
Five ways length actually goes wrong
In practice almost every length problem is one of these, and none is fixed by picking a different number.
Padding to a threshold
Stretching a six-minute idea to clear eight minutes. Discussed above: it usually converts retention into unserved ad inventory. The exception is a video whose extra material is genuinely wanted — an extended example, a second case study — rather than the same material said slower.
Front-loading everything
The opposite failure, and more common since everyone learnt to hook hard. Answer the title's question completely in the first ninety seconds and you have told the viewer, accurately, that they can leave. Strong openings should establish that the answer is coming and that it is worth the wait, then deliver something in the middle that the opening promised.
Chapters as an excuse for length
Chapters make a long video navigable; they do not make it watched. A viewer who skips to chapter four and leaves produces a worse retention curve than one who watched a tight video end to end. Chapters are a service to people who came for one section, not a licence to add sections.
One length for every idea
Channels settle into a house runtime and then bend every idea to fit it. Some of your ideas are four-minute ideas. Some are twenty-five-minute ideas. Forcing both into a twelve-minute template makes the first bloated and the second shallow, and the retention graphs show both.
Cutting a good video short for a rule
The mirror-image mistake, and the one this article risks encouraging. If a video is holding people at fourteen minutes, the correct length is fourteen minutes or more. Nothing above applies to a curve that is still flat when the video ends.
Length does not get the click
One last thing, and it is the reason a length article on a thumbnail site is not a detour. Runtime decides what happens after the click. It has almost no influence on whether the click happens at all — that is the thumbnail, the title and the topic, competing in a feed where your video is one rectangle among twenty. A perfectly judged nine minutes on a video nobody opens is nine minutes nobody watched.
The two halves are linked more tightly than they look, though, and Test & Compare is where you can see it. YouTube's experiment tool picks its winner by watch time per impression rather than raw click-through rate, which means a package that wins the click and loses the viewer can lose the test to a duller one that keeps people. Title A/B testing joined thumbnails in that tool in a wide rollout at the end of 2025, so the whole package can now be tested against the same measure. The mechanics are in the guide to running a three-variant test, and what a healthy click-through rate looks like before you start is in what counts as a good CTR.
That measure is also the cleanest possible summary of the length question. YouTube is not asking how long your video is. It is asking how much watching an impression produced. Length is one of the levers on that number, and not the strongest one.
So how long should it be?
As long as the idea holds, and not one minute longer. In practice that means: make it as long as it needs to be, publish, read the retention curve, and let the next video's length be an argument the last video's data won. If your curves break at four minutes, make four-minute videos until they stop breaking there. If they run flat to twelve, you have permission to go to twenty.
The three numbers worth remembering are the three the platform enforces: three minutes decides whether a vertical video is a Short, eight minutes decides whether mid-rolls exist, twelve hours is the ceiling. Every other number you have been given is somebody's average, and averages are made of channels that are not yours.
And once the length is right, the constraint moves back to the front of the funnel — to the rectangle that decides whether any of those minutes get watched. That is the part Thumblore exists for: generating thumbnails that survive a crowded feed at the size people actually see them, so the retention work you just did has an audience to work on.