Good explainer voice over sounds like a colleague explaining what's on screen: complete sentences, plain words, one short paragraph per scene, read at about 1.9 words per second. It gets written last — after the visuals are locked — because a line written to a finished picture can land its key word on the key visual moment, and words written first never sync.
Across our production record, every rejected narration draft failed in one of exactly two ways. This guide names both, shows the rejected lines next to the rewrites that shipped, and covers the mechanics of getting a clean read from a synthesized voice.
The voice is subordinate to the picture
The authority rule from our handoff sheets, verbatim: "Visuals are authoritative: re-time VO to the picture, not the reverse." The picture shows state; the narration says what it means. That division of labor decides what the voice is for — it carries the why, names the concepts, and tells the viewer what the picture means. It says only what the picture can't show.
The anti-test for every line: if a line could play over any video, it's not doing its job. Narration is about this screen. "Selecting a block shows everything it actually did" only makes sense over the frame where a block gets selected — that's what makes it narration and not copywriting.
This is also why you never bolt new narration onto an animation that was timed to different words. It won't sync, and it reads as floating. If the picture changes, the words get rewritten to the new picture. The full authoring order is in how to write an explainer video script.
The register: condensed, never compressed
The target, in the words of the director who graded our scripts: full sentences that are "compact and dense in meaning — they fit in the allotted time and don't run on — but easy to understand." The house rules that got drafts accepted:
- Complete sentences with connective tissue. Condensed means zero fluff. It does not mean fragments. "Every time you add a new document, the platform keeps it in sync by re-indexing the chunks" — never "Documents in. Chunks synced."
- Explain the screen. Narrate what's visible and what it means, including the interaction the viewer is watching ("selecting a block shows…").
- Name concepts deliberately, in a fixed cadence: term, then its role, then the derived term. "Inside a knowledge base, the key unit of information is the document. A document gets processed into smaller pieces called chunks."
- Make the viewer the subject where you can. "You probably already have…", "your documents", "let's look inside."
- Walk the naive path first when it helps. "The naive solution is to try stuffing everything into one place, but you'd quickly hit the limit and performance degrades" — the why before the what.
- Calibrate every claim. If you can't support "more informed," write "informed." One of our gold scripts makes exactly that downgrade.
- End on behavior or payoff, never a slogan. "When a workflow does something unexpected, the log is the first place to look."
The final gate is the read-aloud test: any line you'd be embarrassed to say to a colleague across a desk gets rewritten. If a person wouldn't say it aloud, it isn't narration.
Failure mode 1: trailer voice
The first draft of one of our videos about run logs came back with a one-line rejection: we are not doing marketing, we are doing explainers. The draft had performed the topic instead of explaining it — scene-setting, suspense beats, punchlines. Three of the pairs, rejected line first:
- ❌ "A ticket comes in, and a run crosses the chain — triage, build, log. From out here, it's already over."
- ✅ "Here's a workflow processing a support ticket — four blocks, start to finish. Every time a workflow runs, the platform records exactly what happened. That record is called a log."
The bad line is fiction with a suspense beat, and it never states the lesson. The rewrite describes the screen, states the fact, and names the concept — the word "log" lands in the first beat.
- ❌ "But the run wrote itself down. Every block, in order, timed — there's where the twelve seconds went."
- ✅ "The log lists every block in the order it ran, along with its timing — so you can see exactly where the twelve seconds of this run were spent."
"The run wrote itself down" personifies a mechanism. The rewrite is one complete sentence: subject, what it contains, what that lets you do.
- ❌ "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 viewer a behavior: where to look, and when.
The same discipline applies to your own manifesto lines. We tried building a film around founder-coined aphorisms once; spoken aloud, they read as grandiose even though the founder wrote them. The rule since: coined phrases appear only when the founder supplies them for that specific video, verbatim. The default register is always plain prose about what's on screen.
Failure mode 2: amputated fragments
The opposite drift: "brief and condensed" over-applied until the prose stops being speech. The tells are labels with colons, stub chains, and counting things the picture already counts. From a video about loops:
- ❌ "For each item of this collection. Three items — three passes. One. Two. Three."
- ✅ "The loop is configured with a collection to iterate over — three items here, so everything inside will run three times."
"For each item of this collection." is a fragment pretending the viewer heard its first half. And "One. Two. Three." narrates a count the picture already shows. Counting beats belong to the visuals; the voice says what the count means.
- ❌ "The loop exits once. Loop dot results: every pass's output, one array."
- ✅ "After the last pass, the loop exits once. Downstream, loop dot results is an array holding every pass's output, in order."
A label with a colon is a caption. The rewrite is a sentence with a subject and a verb — and it keeps "in order," which is half the point of a loop.
- ❌ "One shape, two schedules. Loop when order matters. Parallel when it doesn't."
- ✅ "One shape, two schedules: use a Loop when the order matters, and Parallel when it doesn't."
Nearly identical content. The difference is purely that the second version is a sentence a person would say. Closers may be compact; they still get verbs.
The one-line test that separates the modes: condensing removes fluff and keeps grammar; compressing removes grammar and keeps keywords. When a scene runs long, trim fluff before you amputate a sentence.
The mechanics of a clean synthesized read
Recorded or synthesized, the same numbers hold — but synthesis changes the workflow enough that it's worth its own checklist. This is how we run ours:
- Pace at ~1.9 words per second. That's what a production read actually measures. A 9-second scene holds one short sentence of 15–18 words; a 12-second scene holds two. Write to the scene's duration, then read it against a stopwatch.
- Leave a breath. About 0.7 seconds of silence after each scene's line. The pad is what keeps the read from feeling wall-to-wall.
- The voice may stretch a scene, never squeeze it. Each scene's final length is the larger of its visual minimum and the audio plus the breath pad. Extra time becomes a hold on a frame the scene is already showing. Extension is allowed; fragments are not.
- One paragraph per scene, headed by the scene's exact name. A typo'd scene name silently gets no audio — keep the narration file keyed to the scene list.
- Commas, not em-dashes, in the final speakable copy. Drafts can use dashes; the text that gets read converts them, because a comma produces the pause you actually want.
- No code tokens read aloud. "Loop dot results," never backticks or punctuation the voice would mangle.
- Synthesize per scene, and keep the takes. Generate each scene's audio separately so a one-line fix in scene 4 never touches the approved reads of scenes 1–3. And synthesis isn't deterministic — regenerating "the same" sentence produces a different take. The audio files you approved are the asset; save them, don't plan to reproduce them.
- Silence is legal. A scene with no narration keeps its authored duration. Not every scene needs words.
Done this way, a narration fix is a sixty-second loop: edit the prose, re-synthesize the one changed scene, re-time, done. Words stay the cheapest thing to change in the whole production — which is exactly why they're written last, on top of the expensive layers instead of underneath them.
FAQ
Synthesized voice or human recording? Both ship. The question is taste, not technology — a flat human read loses to a well-directed synthetic one, and vice versa. Recording adds cost and a booking dependency; synthesis adds the take-management discipline above. If a founder records their own, hand them a per-scene sheet: timing, what's on screen at the key moments, what the scene must communicate, and where the picture holds still for the voice to breathe.
How many words fit in a 60-second video? Around 100–110 spoken words: 60 seconds at 1.9 words per second, minus a breath pad per scene. If your draft is 200 words, the video is either going to run long or the read will feel breathless. See explainer video length for how runtime is chosen.
What if the narration runs longer than the animation? Let the scene extend — the extra time holds the scene's settled frame. What you never do is compress the picture below what it needs, or amputate the sentence into fragments to force a fit. Trim fluff first; verbs last.
Can I write the voice over first and animate to it? It's the most common order and the most common mistake. Visuals animated to fixed words float — the key visual never lands on the key word. Lock the picture, then write to it.
When you want to hear a register instead of reading about one, send us your product's URL and twenty voiced candidate directions come back within a day.