The first three seconds: engineering a hook that survives the scroll.

Everyone in short-form repeats the same line: you have three seconds. It is true and it is useless. A stopwatch is not a brief. The question worth answering is not how long you have - it is what the person on the other side is deciding inside that window, because once you know that, the opening stops being a guess and starts being a specification.

The viewer is not deciding whether to watch. They are deciding whether to leave.

Attention on a feed does not start at zero and get earned upward. It starts full and drains, and the thumb is already moving. Nobody arrives curious. They arrive mid-scroll with a default action queued, and the only thing your first second can do is fail to give them a reason to complete it.

Which makes grab attention the wrong instruction. It produces loud openings, hard cuts and shouted first lines - things that announce a video is trying, and trying is itself a reason to leave. The useful instruction is quieter and much harder to execute: remove the exit. Give the eye something it has not finished reading yet.

Four things that hold a first second.

Start mid-action. The frame opens on something already happening - the cut is made, the box is open, the hand is moving. Setup is what you show people who have already chosen to stay. No greeting, no logo, no establishing shot, because each of those spends a second telling the viewer the video has not started yet.

Open a loop the eye can see. The unanswered question has to be visual, not spoken. A verbal tease asks the viewer to trust that an answer is coming; a visible anomaly - something in frame that does not add up yet - makes them stay to resolve it themselves. Nobody scrolls away from a question they have already started answering.

Be legible at thumbnail scale. Most of the feed is watched small, in bad light, at an angle, one-handed. If the subject is not readable when the frame is the size of a stamp, the composition has failed regardless of how it looks on the edit monitor.

Land the audio on a beat. Sound arriving on a breath, on room tone or on half a second of silence reads as amateur before a single word is understood. Sound arriving on a hit reads as deliberate. Nobody analyses this. They feel it, and they act on it.

A hook does not grab attention. It removes the reason to leave.

Why the hook cannot be written last.

The default production order is backwards: write the film, cut the film, then find something usable in the first ten seconds and move it to the front. That produces openings which are technically first and functionally arbitrary.

The opening has to be the constraint everything else is built around, because everything downstream is only ever seen by the people it kept. A brilliant middle watched by four percent of impressions is a brilliant middle nobody bought. That is why on our line the hook set is agreed before anything is shot - not as a creative preference, but because it changes what gets filmed.

How we find out which one works.

Not by discussing it. The same body gets several openings, they go out together, and the feed sorts them inside days. Watch-through tells you which first second held, and the rest is arithmetic. It is the cheapest test available anywhere in marketing and it costs nothing to run - which is exactly why we run it before a media budget exists rather than after.

What breaks a hook that tested well.

Hooks decay, and this is the part teams are least prepared for. The opening that carried a format for six weeks stops working, and usually nothing about it got worse - it stopped being new to an audience that had already seen it four times, and novelty was part of what it was doing. Find one winner, run it into the ground, and you learn this the expensive way.

Which is why the asset is not the hook. The asset is the library: a stock of openings tested against each other, so that when one fatigues there is a tested replacement rather than a meeting. Any single hook has a shelf life. A hook library does not.

Build the constraint in.

If there is one thing to take from this, it is to stop treating the opening as packaging applied at the end and start treating it as the specification everything else answers to. The first second is not the trailer for the video. On this platform it is most of the video - because for most of the audience it is the only part that ever plays.