An explainer video script is written backwards: you decide the one sentence the viewer must believe by the end, design the picture for each scene, and write the spoken words last. Most weak explainers were made words-first, with pictures bolted on afterward, and that order produces narrated slideshows.
Video has two channels where prose has one: the picture shows state, and the voice says what the state means. A writing process that produces only words has decided, by default, that the picture will illustrate the words, and illustration is the weakest job a picture can do.
The examples below come from a 90-second explainer for a CRM: one inbound lead moves through the pipeline from a cold row to a closed deal.
Write the visuals before the words
Words-first fails for two reasons. Sync only works in one direction: a narration line written to a locked picture can land its key word on the key visual moment, so "the deal closes itself" plays exactly as the pipeline card flips to won, while a picture animated to fit fixed words chases sentence rhythm instead of staging events. And when the words come first, everything worth saying has already been said, so the picture restates the voice and one channel goes redundant.
The work runs in six steps:
- Decide the learning outcome. What must the viewer be able to think after watching? For the CRM video: "I can see where every deal stands without asking anyone."
- Write the one idea. One declarative sentence the whole video argues.
- Build the causal chain. Orient from what the viewer already knows, make the macro claim, show the micro mechanism, then work one example through.
- Cut the chain into beats. One idea per beat; the beats become the locked scene list.
- Describe the visual for every beat, in words, before any production. A beat whose picture you can't describe isn't understood yet.
- Write the narration last, each line to a finished picture.
Start with the one idea
Every script we produce opens with a mandatory field:
The one idea: one sentence. Everything in the video exists to make it feel obvious.
The field holds a claim, never a topic, and the viewer should finish the video feeling the sentence is obvious rather than merely stated. One from our produced scripts: "an agent is just a workflow that can reason." For the CRM video: "the pipeline updates itself, so a rep never types what the system already knows."
Test your sentence three ways:
- Is it falsifiable? "Our CRM is powerful" fails. "The pipeline updates itself, so a rep never types what the system already knows" passes, because a viewer could watch the video and check it.
- Does everything on screen serve it? Any scene that doesn't push the sentence toward obvious gets cut. A beautiful reporting scene dies if the video's sentence is about the pipeline updating itself.
- Does this video own it? In a series, each video owns exactly one idea and defers to its neighbors. Two videos that half-teach the same concept both fail.
One idea per scene
The one-idea rule repeats at scene level, with a built-in test: the scene's
name. Every scene gets a slug that states its idea. The CRM list might read
lead-lands-itself, every-touch-logged, the-stage-moves-alone, and a
scene you can't name with its idea carries two ideas and needs a split.
Each scene also carries a beat intent: one sentence stating what the viewer must take from the scene, written for whoever writes the narration later, never shown on screen. The picture makes the claim, and the intent tells the voice what to add. The full rule has its own guide.
The scene list is the script
A finished script is a contract. Once the stakeholder blesses it, production builds exactly those scenes, and changes arrive as new versions rather than silent edits. Beyond the one idea, the document names its arc, justifies how many times the demo runs on screen, grounds every on-screen value in a real product artifact (real-shaped company names and deal amounts from a seeded workspace, never lorem ipsum), and locks each scene's name, duration, visual, and intent. The row-by-row anatomy, with a worked contract-grade entry, lives in the storyboard guide.
A full video is short: six to eight scenes of 8–12 seconds each, 60–90 seconds total (the runtime guide).
Bad lines vs. good lines
Every rejected draft in our record failed in one of two ways: it performed instead of explaining, or it compressed instead of condensing. Both pairs below are real, rejected line first.
Performing instead of explaining, from the closing scene of a video about run logs:
- ❌ "Every run writes one of these. Nothing about a run is a mystery."
- ✅ "Every run produces a log automatically. When a workflow does something unexpected, the log is the first place to look."
"Nothing about a run is a mystery" asserts a feeling. The rewrite gives the behavior instead: where to look, and when.
Compressing instead of condensing, from the opening scene of a video about loops:
- ❌ "Some jobs are a list — same steps, every item. The Loop is a block that holds blocks: the inside runs once per item."
- ✅ "Some jobs are a list — the same steps need to run for every item. For that, the platform has the Loop: a block that holds other blocks. Whatever you place inside it runs once per item."
The rejected version is clipped: three stubs doing the work of sentences. Condensing removes fluff and keeps grammar, while compressing removes grammar and keeps keywords. More pairs from our record are in the voice over guide.
Write the closing line first
One narration line may exist before the visuals: the closer. The whole video aims at it, and every other line waits for the visuals to lock. A director-approved closer from a real video: "That's the whole model. A trigger starts it, each block runs as soon as its inputs arrive, data moves through connection tags, and the run ends at the Result." A viewer could repeat that line, and repeating it means they hold the whole video. For the CRM video: "From first touch to closed, the pipeline moved the deal, and nobody typed a status."
FAQ
How many words is a 60–90 second script? Around 150–170 spoken words for a 90-second video. Narration reads far slower than silent reading, and a 9-second scene holds one sentence of 15–18 words. The picture carries the rest.
Do I write the script or the storyboard first? They're the same artifact at different resolutions: the scene list is the script, the storyboard renders its descriptions into frames, and the spoken words come after both.
Does every scene need narration? No. A silent beat that holds its authored duration is a legal and useful choice. Every scene needs an idea; words are optional.
Can I reuse my website copy as the script? Read it aloud first. Web copy is written to be scanned, and narration is written to be said. A line you wouldn't say to a colleague across a desk isn't narration yet.
If you'd rather have this done for your product, start with the URL: we come back with twenty short candidates, and you pick one.