Research

We analyse Shorts for the creators who connect a channel to ShortLens. That gives us something most advice about Shorts does not have: retention curves and view histories for real Shorts on real channels, measured over time rather than recalled.

These pages are what came out of looking at it. A few of them disagree with numbers that circulate widely. Where that happens we show the working, state how many Shorts it rests on, and say what would change our mind.

Three rules we hold ourselves to here. Every figure carries its sample size and the date it was measured, because a number without those is not a finding. We say what we cannot see, and there is a lot: our earliest reading of any Short is around day three, so the window that decides distribution is one we never observe. And we do not guess at what the platform is doing. We describe what we measured on the catalogue in front of us, and stop there.

• Your Short says 127% viewed. Is that good?

  What a number above 100% actually is, how common it is, and what it predicted. The advice everywhere says a looping Short is a strong signal. On our data, outcomes peaked between 100% and 150% and fell away above 200%.

  575 Shorts across 18 channels, measured 15 September 2026.

• The "viral" retention bar, and what length does to it

  Published threshold tables say you need a certain average view percentage to break out. 51% of the Shorts we measured cleared their own band's bar. 96% of those never reached 4,000 views. A threshold quoted without a length band is not a threshold.

  575 Shorts across 18 channels, measured 15 September 2026.

• How long does a small Short keep getting views?

  One widely cited study says views keep arriving into week four. Our small-channel Shorts stop at roughly nothing after day 14. Both can be true, and the reason why is the most useful thing on the page.

  310 Shorts with at least two readings, measured 15 September 2026.

More to come. If you want to run any of these measurements on your own channel, every page ends with the steps, and none of them require our product.