You have 2.3 seconds. Here is what to do with them.

The read. Across 55 Meta ad accounts we manage, 80% of video plays end before the three-second mark. The decision to keep watching is made in about 2.3 seconds, before the viewer has read a word. This note explains what has to happen in that window, how to keep someone watching once you have them, and what we changed in our own creative standard this year.

This is the first founder note on our new site, so it starts with the idea our work is built on: people decide fast, mostly without thinking, and an ad has to earn its place in that decision before it can say anything.

Where 2.3 seconds comes from

I spent my academic life studying how people read each other, and my Google years watching what they do with a screen. The two agree on one thing. Attention is not given, it is caught, and it is caught by a small number of signals the brain processes before conscious thought: a face, movement, contrast, and something unexpected in an expected place.

Published research puts the first read of a face and its expression at well under half a second. Our own numbers say the feed gives you roughly 2.3 seconds before the thumb moves. That is not an attention span, and the "eight-second attention span" claim you will have read elsewhere does not survive contact with evidence. It is a decision window. People are not distracted; they are deciding.

What happens before three seconds in a real feed

This year we measured video retention across 55 client accounts in hospitality, fitness, beauty and e-commerce. Eight plays in ten end before three seconds. Most of those plays happen with the sound off, on a phone, between two other pieces of content the person actually chose to see.

Three patterns showed up in the ads that lost people early. A logo or title card in the first frame. A slow establishing shot (the empty dining room, the studio before class). A person talking to camera with no on-screen text, which means nothing at all reaches a sound-off viewer.

The first frame: four things that work

1. A face looking at the camera, feeling something real. Not a pose. The brain reads a real expression faster than a performed one, and it trusts it more.
2. Movement inside the frame in the first second. A pour, a lift, a brush stroke, a door opening. Motion is read before meaning.
3. Contrast between the subject and the background. If the frame is uniformly busy or uniformly calm, nothing is selected by the eye.
4. The promise in five words of on-screen text. Sound is off. Say what this is for, in the language the viewer would use, in the safe zone.

Notice what is missing: the logo, the offer and the brand name. They belong later, once the viewer has decided to stay.

After the hook: keep giving reasons to watch

Winning the first 2.3 seconds buys you the next few. It does not buy the rest of the video. Retention curves in every account we looked at show the same shape: a cliff, then a slope, and the slope steepens wherever the ad stops giving the viewer a new reason to stay.

So we rehook. A rehook is a new reason to keep watching: a change of shot, a new face, a question, a reveal, the result. We place them at fixed beats in the first 13 seconds (5, 9 and 13) so a viewer never goes more than a few seconds without one. The ads that hold viewers past 15 seconds are almost never the ones with the best offer. They are the ones with the best rhythm.

A real smile beats a professional one

One principle earns its own paragraph because it is the one our clients find hardest to accept. A real smile involves the eyes as well as the mouth (the Duchenne smile in the research literature). People detect the difference in a fraction of a second, and they read the polished version as advertising. In our accounts, the versions with a real expression in the first frame consistently outperform the polished ones from the same shoot. So we now brief for it: the laugh after the take, the reaction to the first bite, the instructor mid-class rather than the instructor introducing the class.

Three first frames that work, by sector

• Restaurant. A guest reacting to a plate, mid-service, room full behind them. Text: "Sunday lunch, booked out by Friday."
• Fitness studio. The last ten seconds of a class, faces lit, movement everywhere. Text: "First class £19."
• Beauty brand. The product doing its job on a real hand, in one motion. Text: "Dry in 60 seconds."

What we do not publish

We score every ad against a set of principles before it goes live, with a weighted breakdown that tells the creative team exactly what to change. That scoring standard is ours and stays in-house. The principles above are the parts that are safe to share, because they only work if you do them well.

What to do on Monday

1. Pull video retention for your top ten ads and read the three-second figure. If more than 80% of plays end before three seconds, you have a first-frame problem, not a targeting problem.
2. Re-cut your three best-performing videos with the logo and title card moved to the end.
3. Put a real face, a movement and a five-word promise in frame one.
4. Add a change of shot or a new reason to watch every few seconds for the first 15.
5. Test the re-cut against the original for a week and keep whichever holds more viewers past three seconds.

With more than 8 million advertisers now using Meta's AI creative tools, the feed is filling with competent, similar ads. The human first frame is the difference.

Frequently asked questions

Is 2.3 seconds a Meta metric? No. It is our working figure for the decision window, drawn from retention data across the accounts we manage. Meta reports three-second plays and ThruPlays; we treat the three-second figure as the first test any ad must pass.

Does this apply to static ads? The same signals apply to the first look at a static image: face, contrast, movement implied by the composition, and one promise. The difference is that a static ad has no second chance, so the first frame is the whole ad.

Where does the 80% figure come from? Our own analysis of Meta video retention across 55 client ad accounts in 2026 (Invincible Digital creative research, Evidence Note 001). It is a baseline, not a target.

Sources

• Invincible Digital creative research, Evidence Note 001, 2026: Meta video retention baseline across 55 client ad accounts (internal).
• PPC Land, 30 April 2026: Meta Q1 2026, more than 8 million advertisers using at least one AI ad creative tool. https://ppc.land/meta-q1-2026-56-3b-revenue-as-ai-tools-double-advertiser-adoption/