The method · 3 October 2026 · 2 min read

The first frame decides: 43.6% stayed versus 12.9%

In our own Shorts, the picture on screen at second zero predicted who kept watching better than anything else we changed. Openers you understand in a split second, like a giant sneaker or a man asleep in a dog bed, kept 28–44% of viewers. Ordinary scenes and screens kept 11–24%.

These are early numbers from one video per opener, so treat them as a strong hint, not a proven rule. Here’s what we saw.

The number we watched

YouTube Studio shows what share of people who were shown a Short in the feed watched it rather than swiping away. We call it “stayed to watch”. It’s the best single number for judging the first second of a video, because that’s when people decide.

Batch 02: odd objects at the top

Opener (first frame) Stayed to watch
A man asleep in a giant dog bed beside a Great Dane 39.7%
A can flying out of a drink holder 30.7%
Teens on a truck piled high with sofas 28.1%
A car covered in stickers 24.9%
Two guys cooking 19.8%
A crowded party 11.2%

The top three open on something odd you can read instantly. The bottom two open on scenes you’ve seen a thousand times.

Batch 03: objects beat screens

Opener (first frame) Stayed to watch
A giant sneaker dripping red, held up to the camera 43.6%
Founders toasting round a laptop 23.9%
A man mid-pushup behind a padlocked phone 18.6%
A man smirking at a laptop with a glowing AI window 12.9%

The winner is a physical object. Three of the losers open on screens: a laptop, a profile page, a phone. A screen can’t be read at swiping speed, so people swipe.

The sneaker video was also the only one in its batch that YouTube kept pushing to new viewers. In both batches, the video with the best “stayed to watch” got the most reach.

What we got wrong

The hooks were different too. The dog bed’s first words named the odd thing straight away (“A dog bed for humans”). The party video’s first words, “How did a sixteen-year-old…”, named nothing odd at all. So some of this gap could be the words, not the picture.

That’s exactly why the next test changes only the first frame and keeps the script fixed.

How you’d test this

Pick one script. Make four versions that open on a strange, physical object and four that open on an ordinary scene or a screen. Change nothing else, post them alternately, and compare the middle “stayed to watch” figure on each side. If objects win on your channel too, make it part of your style and move on to testing the hook.