How to write documentary narration that sounds like a film, not a Wikipedia page

Rules for writing narration that carries a documentary: sentence length, rhythm, where to pause, how to structure for retention, and how to write for a performed voice.

Documentaries are voice-led. The picture follows the narration, the music follows the narration, and the audience stays or leaves on the narration. Most weak documentaries have a script that reads well and sounds wrong. Here's how to write for the ear.

Write short

Spoken sentences are shorter than written ones. Aim for 8–15 words. One idea per sentence. If a sentence has a comma and an "and", split it.

Written: Despite having no coastline, and facing hostility from all of its neighbours, the country decided in 1957 to build a navy.

Spoken: The country had no coastline. Every neighbour was hostile. In 1957, it built a navy.

Open with the strangest fact

The first ten seconds are the cold open. No context, no "in this video". The single most surprising thing, stated plainly.

Pose one question

By 40 seconds the audience should know what the film will answer. Say it as a question. Every chapter then pays off part of it.

Mark the rhythm on the page

Write pauses in. A line on its own is a pause. Three short lines in a row are a build. A one-word line is a stop.

The fleet sailed at dawn. Nobody came back.

Editors cut to those breaks; narrators breathe at them. If they're not on the page, neither happens.

Give the picture room

Don't describe what's on screen. Narration says what the picture can't: the date, the cause, the consequence. When a map is doing the work, write less.

Use dates and numbers as beats

"In 1683" is a cut point. "Four hundred thousand" is a hold. Place them where you want the picture to change.

Write the turn

Around two-thirds through, the film needs a reversal: the cause wasn't what we thought, the winner lost, the modern consequence. Write it as a short paragraph after a pause, with the shortest sentences in the script.

Land, don't summarise

The last thirty seconds say what it meant, in two or three lines, and stop. Don't recap the chapters.

Writing for a performed voice

Modern narration models perform the script: pauses, emphasis, pace changes. They do it better when the writing tells them to. In Glisse, narration is performed from your script with pauses and emphasis, and the cut is anchored to the voice word by word, so the rhythm you write on the page becomes the rhythm of the film.

FAQ

How many words is a 10-minute documentary script?

About 1,300 to 1,600 words at a documentary pace, less if maps and footage carry long sections.

Should documentary narration be first or third person?

Third person for evidential documentaries; first person for video essays and personal geopolitics. Pick one and keep it.

How do I make AI narration sound natural?

Write short sentences, put pauses on the page, and use a narration model that performs rather than reads. Glisse's narration is performed and word-timed.