Published 27 October 2026 · Figures measured 15 September 2026 · Every figure carries the number of Shorts behind it
The question nobody answers
If you have searched good retention but no views, or high average view percentage but low views, you already know the shape of the advice you get back. Work on your hook. Tighten the first three seconds. Make them stay.
You already did. That is why you are searching.
Here are creators describing it in their own words, collected from public forums in October 2026:
"It's currently around 36.7K views and has 82% stayed to watch, 18% swiped away, 105.3% average percentage viewed, and a 36-second average view duration on a 34-second video ... then completely flattened out."
"I've had videos in Roblox shorts with 82 vvsa and 100 retention at like 50-60 seconds and they didn't blow up."
"I don't get into the same jails consistently even though my shorts are the same quality."
That last one is the whole problem in one sentence. Same quality, different outcome. So we checked whether that is actually true, on real catalogues, with the numbers in front of us.
First, the thing that will make you feel better
A Short with worse metrics beating a Short with better metrics is not a glitch. It is the normal case.
We took every pair of Shorts on the same channel and asked which one had the better average percentage viewed, and then which one actually got more views. Across 16,930 pairs, the Short with the better retention lost 44.6% of the time.
Not rare. Not a weird edge case. Close to half.
So when your better Short loses, nothing has gone wrong that you need to explain. You have landed on the ordinary outcome.
We tested four different retention numbers. None of them picked the winner.
Someone will reasonably say: average percentage viewed is a blunt instrument, look at the shape of the curve instead. Fair. So we tested four separate things.
To make it a fair fight we compared Shorts on the same channel and in the same length band. No comparing a 9 second Short to a 50 second one, no comparing a big channel to a small one. Then we asked how often the Short with the better number was the Short that actually won.
| What we measured | Got it backwards |
|---|---|
| How many stayed through the hook | 44.7% |
| How many were still there in the middle | 49.9% |
| How many made it to the end | 50.4% |
| Average percentage viewed | 49.0% |
Three of the four are a coin flip. Hook retention is the only one that leans at all, and it leans about as far as a slightly bent coin.
If retention ranked your Shorts, these numbers would be nowhere near 50%. They are.
It does not even point the same way on different channels
Here is the part that convinced us. We ran the same test inside individual channels. If finishing strong were a real advantage, it should at least be an advantage consistently.
On one channel, the Short people finished was the loser 64.6% of the time. On another, it was the loser 37.7% of the time. Same test, opposite answers.
A real effect does not flip direction depending on whose channel you run it on.
We believed the opposite. We published it. Then we withdrew it.
This is worth telling you because it is the reason we trust the result above.
Earlier in 2026 we measured something exciting: Shorts that held viewers to the end did 6.5 times better than Shorts that did not. We nearly built product advice on it. "Make sure your ending resolves" was about to become something ShortLens told people.
Then we checked it properly, and it fell apart.
The original measurement had compared Shorts across a 100x range of view counts. And here is the trap: retention rises when a Short gets distributed further. More people see it, the people who see it are better matched to it, and your measured retention goes up as a result. On that same channel, Shorts with 300 to 1,000 views averaged 59% viewed. Shorts above 20,000 views averaged 98%.
So we had not discovered that good endings cause views. We had discovered that views cause good endings to show up in your analytics. Backwards.
Hold the view range fixed, as the table above does, and the 6.5x becomes 50.4%. A coin flip.
Why we are telling you about our own mistake. Because the correct version of this test is unglamorous and the incorrect version is thrilling, and the incorrect version is the one that circulates. If you see a dramatic retention claim anywhere, including from us, the first question is whether the Shorts being compared had similar view counts. Usually nobody says.
So what does retention actually do?
Our best read of our own data is this. Retention is a gate, not a ranking.
There is a bar your Short has to clear. Below it, almost nothing happens, and that failure is real and it is fixable and it is worth your attention. Above the bar, retention stops sorting your Shorts, and something else takes over.
That is why both of these are true at the same time, and why they feel contradictory when nobody separates them:
- Retention correlates strongly with views when you look across very different Shorts and channels. Low retention really does go with going nowhere.
- Retention tells you almost nothing about which of your own comparable Shorts will win.
This is not us saying retention does not matter. Clear the bar. It is us saying that once you have cleared it, polishing further is not the lever you were told it was.
Two Shorts, one channel, fourteen times apart
One creator on our data posted a Short that held 75% of its audience to the end, with 97.4% average percentage viewed. Genuinely strong numbers. It took 1,902 views.
Six weeks later, on the same channel, a Short that held 78% took 26,833 views.
Three points of retention. Fourteen times the reach.
Nothing in the retention data explains that gap, and we are not going to invent something that does.
What we think is actually going on, and what we cannot see
Once a Short clears the bar, the things that appear to decide the outcome are the audience it got seeded to, whether the topic happened to be in demand that week, and timing.
We cannot observe any of those three. Neither can any other tool reading your analytics, whatever it says in its marketing.
We would rather end on an honest unexplained remainder than sell you a fourth factor we made up.
The limits, before you use any of this
One limit works against us, and you should know it. Only 252 of the 675 Shorts in this base carry a stored second-by-second curve, because we cap how many detailed curves we fetch per sync. That subsample leans recent. We cannot fully rule out that the Shorts carrying curves differ from those that do not, and that is the weakest point on this page.
Two limits work in our favour, which is why we are comfortable with the result. The reverse-causality problem above means retention's real predictive power is weaker than 50.4% suggests, not stronger. And the 2,513 comparisons come from 252 Shorts, not 2,513 independent ones, so we report the honest number: this is 252 Shorts across 17 channels. Anyone quoting a pair count as a sample size, us included, is inflating it.
These are small and mid-sized channels. Roughly 1.1% of Shorts in our whole catalogue have passed 10,000 views. Nothing here describes a channel that routinely reaches six figures.
The one thing we genuinely cannot see
Every serious account of Shorts distribution points at the same metric: viewed versus swiped away, the share of people who were shown your Short and stayed instead of flicking past.
We do not have it. It is visible to you in YouTube Studio and it is not in the data a creator can authorise us to read, and there is no honest substitute for it. Average percentage viewed is not one: it describes the people who started watching and says nothing about how many never did.
So we will not estimate it, and we will not tell you whether you clear anybody's threshold for it. We explain what happened to the people who started. Studio tells you how many never did.
Questions creators ask about this
Taken from public creator forums in October 2026, close to the wording they were asked in.
My Short has good retention but no views. What did I do wrong?
Quite possibly nothing. On 16,930 same-channel comparisons, the Short with the better average percentage viewed got fewer views 44.6% of the time. That is the base rate, not a malfunction.
The useful question is not "what did I do wrong on this Short" but "is this Short below the bar or above it". Below the bar there is usually a specific moment where people left, and that is worth fixing. Above it, the honest answer is that the difference was decided by something neither of us can see.
Why did my worse Short do better than my good one?
Because within one channel and one length band, retention does not rank Shorts. We tested four retention statistics on 252 Shorts across 17 channels and three of them were a coin flip at picking the winner, with the fourth close to it.
It also will not be consistent. On one channel in our data the Short people finished was the loser 64.6% of the time, and on another 37.7%.
Should I stop worrying about retention then?
No, and this is the part that is easy to get backwards. Retention looks like a gate: below a certain level Shorts reliably go nowhere, and that is a real and fixable problem. What our data says is that above that level, improving further does not sort your Shorts.
So it is worth finding out whether a given Short failed for a reason you can fix, and it is not worth chasing a higher number on a Short that already cleared the bar.
Does my topic or niche get suppressed?
We looked and found nothing. We ran four pre-registered topic splits on our own catalogue and every one came back null, below the smallest effect our sample could have detected.
Pages that attribute low views to topic category are usually reasoning from a single channel without controlling for view band. We would be reporting it if we had found it.
Is my Short shadowbanned?
We cannot tell you, and neither can anyone else looking at your analytics from the outside. There is no field in the data a creator can share with us that says a Short was suppressed, so any tool reporting a shadowban is inferring it from low views and calling the inference a finding.
What we can say is that low views on their own are not evidence of anything unusual. Roughly 1.1% of the Shorts in our whole catalogue have ever passed 10,000 views. A Short that went nowhere is the ordinary outcome, not a sign that something was done to it.
What is a good retention rate for a YouTube Short?
There is no single number, and a number quoted without a length band is close to meaningless. Short Shorts naturally read far higher than long ones, because the same person watching the same way produces a bigger percentage on a shorter video.
We put the actual distribution by length on its own page, including what happened when we checked published "viral" thresholds against our catalogue: 51% of our Shorts cleared their own band's bar, and 96% of those never reached 4,000 views.
My Short got almost no views at all. Is this the same problem?
Probably not, and the distinction matters. This page is about Shorts that got distributed and then stalled, where the retention numbers are real and simply did not predict the outcome.
A Short sitting at single or double digit views never got far enough for any of this to apply. Its retention figure is measured on so few people that it is mostly noise, which is why we withhold detailed drop analysis below 50 views rather than present a shape we do not trust.
If retention does not predict it, what should I actually look at?
Whether a specific Short has a specific problem, rather than whether its summary number is high.
A retention curve with a visible cliff at 0:07 tells you something you can act on: people left at a moment, and you can go and look at what was on screen. A curve that never drops below one complete viewing has no drop-off in it to study, however impressive the headline percentage looks. That is a different question from "is my number good", and it is the one with an answer.
Three questions this page deliberately does not answer. What average percentage viewed is normal for a Short of a given length has its own table behind it. Whether a figure above 100% is good is a question about replays, not about the gate. And whether posting more would help is about how many chances you take rather than what any one Short did. And whether the number it stopped at is a real ceiling is the question most creators ask first.
How this was measured
Why we looked at this at all
ShortLens reads the measured retention curve of each of your Shorts and tells you where viewers left, to the second. This analysis is the reason we are careful about what we claim that means. We can tell you that 29% of your audience left at 0:07 and what was happening on screen when they did. We cannot tell you that fixing it will get you views, and we are not going to pretend otherwise.