How to write a short-form video script

There are two camps on scripting short video. One says scripting kills spontaneity and you should just talk. The other says every second should be planned. Both are describing real failures: unscripted video meanders, and over-scripted video sounds like a person reading.

The way out is to script the structure tightly and the wording loosely. Know exactly what happens at each beat and roughly how you will say it, then let the actual sentences come out slightly differently every take. That is how professional presenters work and it is the only version of this that survives contact with a camera.

Know your word count

Comfortable conversational speech is around 150 words a minute. On camera, with energy, most people land between 150 and 170. That gives you a hard budget, and knowing it in advance prevents the single most common script failure, which is writing four times too much and then rushing.

The four beats

Nearly every short video that works has the same four parts, in the same order, at wildly different proportions.

Hook (1 to 4 seconds)

One sentence, written first, before anything else exists. It opens a loop rather than closing one. This is a big enough subject to have its own page, and if the hook is not working, stop and fix it rather than writing the rest.

The turn (4 to 8 seconds)

The most under-written part of a short script and the place most videos die. Here you pay back a little of what the hook borrowed: the first item, the first piece of evidence, a partial answer. If your script has context, setup or a caveat here, move it or cut it.

The body (the bulk)

Where the actual content lives. The only structural rule is that something new must arrive every three to five seconds, and “something new” includes a change in what is being said. A body that is one continuous explanation with no beats in it will sag no matter how good the explanation is.

If the body is a list, say the number of items at the top. It gives the viewer a map and it measurably helps them stay.

The close (2 to 4 seconds)

Short, and not a summary. A summary tells the viewer the video is over, which is fine at the very end and catastrophic anywhere earlier. The strongest closes do one of three things: land the payoff you promised, ask a question that is worth answering in a comment, or loop back into the first frame so the video plays twice.

What a close should almost never be is “follow for more”. It is the least effective sentence in short video, it costs you two seconds of retention, and the follow decision is made by the video rather than by the request.

Write for the ear

Text written to be read and text written to be spoken are different crafts, and the tells are consistent.

• Short sentences. If you run out of breath reading it aloud, it is too long. Full stops are free.
• Contractions. Nobody says “do not” out loud. Write “don’t” if that is what you would say.
• One idea per sentence. Subordinate clauses are invisible in speech; the listener has no commas.
• Numbers first. “Three things” before the three things, always.
• No signposting. “In this video I am going to...” is written-word furniture. Cut every instance.

Keeping it yours when a model wrote it

AI-written scripts have a recognisable sound, and audiences have become extremely good at spotting it. The tells are consistent: tricolons everywhere, “in today’s fast-paced world”, em dashes in speech, an even, unvarying rhythm, and a closing sentence that summarises what was just said.

A generated script is best treated as a structural draft: the right beats in the right order, with placeholder wording. Then do three passes.

1. Replace every generic example with one from your own life. This is where nearly all the personality lives.
2. Break the rhythm. Delete a sentence entirely, and make another one much shorter than the ones around it.
3. Read it aloud and rewrite anything you would not say. That includes any word you would have to think about before using.

Viral Pop writes the structural draft, including the shot list and the caption, and the three passes above are still yours to do. That is the correct division of labour: a model is good at the shape and bad at sounding like a specific person.

A worked 45-second script

Hook (0:00)

“Two of the three lights in this shot are doing nothing.” That is 11 words.

Turn (0:03)

“This one.” (Switch it off. Nothing changes.) “And this one.” (Off. Nothing.) Eight words, and the payoff is visual.

Body (0:09)

The one that matters, why it is placed where it is, what happens when you move it a metre, and the one situation where you do want a second light. About 75 words.

Close (0:40)

“So if you own three lights and your video still looks flat, it is not the lights.” Eighteen words, and it lands back on the hook.

Roughly 112 words, four beats, something new every few seconds, and the close loops. That is the whole format. Once the script exists, the next problem is delivering it without sounding like you are reading, which is what the teleprompter page is about.