Everyone has an opinion about how long a product demo script should be. We can check ours against 1,119 AI-written scenes across 970 demos.
Two numbers do most of the work. The median scene is 64 words, and 91.4% of demos are a single scene.
How many scenes does a product demo need?
One, nearly always. Of 970 completed demos:
| Scenes | Demos | Share |
|---|---|---|
| 1 | 887 | 91.4% |
| 3 | 69 | 7.1% |
| 2 | 14 | 1.4% |
This is downstream of something we found earlier: the median site we film has only two discoverable pages. A product with two pages does not have a five-act structure in it. The script follows the product, and the product is small.
If you are staring at a scene list trying to get to six, the honest question is whether your product has six things worth showing, not whether your script is thin.
How long is one scene of narration?
64 words at the median. Across 1,119 scenes with narration: median 64 words, 90th percentile 93 words, longest 505.
That is roughly 25 to 30 seconds of speech at a natural pace — which lines up with the median finished demo running 33 seconds, once you add the title card and the sign-off.
Nine in ten scenes come in under 93 words. If a scene of yours is running past 150, it is probably two scenes.
Why does 64 words work?
A single paragraph is about as much as a viewer will absorb over one continuous shot of a page. Beyond that the narration outruns the visual: you are still talking about the pricing table while the camera has nothing left to show on it.
The constraint is the shot, not the sentence. One page of a product supports about a paragraph of explanation, and the script length falls out of that.
What should a demo script actually contain?
Working backwards from the numbers above — one scene, about 64 words, roughly 30 seconds — the structure that fits is narrow:
Say what it is, in the first sentence. With one scene there is no second chance to establish context, and a viewer who does not know what they are looking at spends the whole clip working it out instead of watching.
Name the thing on screen. The narration and the visual have to agree. Sixty-four words is not enough to describe something the camera is not showing.
Stop before the shot does. Narration that runs to the last frame leaves no beat to land on. A short silence at the end reads as confidence; a sentence still finishing as the video cuts reads as a mistake.
Methodology and limits
Every completed demo in our production database as of 19 August 2026, not a sample: 970 demos with a scene structure recorded, and 1,119 individual scenes with non-empty narration. Word counts are whitespace-delimited. Percentages are rounded to one decimal place.
Caveats:
- These are AI-written scripts, some of them subsequently edited by the customer. We cannot cleanly separate the two in this data, so treat 64 words as "what a finished scene looks like", not "what the model produces unedited".
- Scene counts are what the generator chose, which is influenced by how many pages the crawler found. A product with more discoverable pages tends to get more scenes, so this is partly a fact about the sites, not only about scripts.
- The 505-word outlier is real and not an error; a small number of demos are built from long single-page sites where one scene carries everything.