How to Write a Video Essay Script
The footage is not the essay. The edit is not the essay. The essay is the argument you wrote before you opened the timeline, and almost every video that drags is a video whose script never decided what it was claiming.
An essay is an argument, not a summary
The most common failure in the form is a script that recounts. It walks through a film, a scandal, a game, a career, and at the end it says something like "and that's why it's so interesting." That is a recap wearing an essay's coat. It has no engine, so the only thing pulling a viewer forward is curiosity about what happens next, which runs out around minute four.
The test is slightly uncomfortable: could a reasonable person disagree with your thesis? "The sequel changed the tone" is a description. "The sequel's tone change is why the trilogy has no ending it can honestly reach" is a claim, and a claim is something a viewer can lean into or argue with. Both keep them watching.
Write the thesis as one sentence before you write anything else. Put it at the top of the document. If you cannot get it into one sentence, you have two essays or none.
The shape most good essays share
There is no mandatory structure, but the essays that hold up tend to move through five moves in roughly this order:
- The hook. Not a trailer for the video and not a greeting. The strongest openings state a specific, strange, verifiable fact, or pose the question the essay will answer. Front-load the interesting thing. The first twenty seconds carry more weight than the next five minutes, and "hey everyone, welcome back to the channel" spends them on nothing.
- The thesis. Say the claim out loud, early, in plain words. Withholding it for a big reveal at minute eleven is a structure that works in mystery fiction and almost nowhere else. A viewer who knows what you are arguing can follow the argument.
- The argument. Two to four sections of evidence, each a self-contained beat that advances the claim rather than restating it. If a section could be cut without weakening the thesis, it is a tangent, and tangents are where retention dies.
- The turn. The best counter-argument, stated fairly and in its strongest form, then answered. This is the most underused move in the form. Steelmanning the opposition buys more credibility than any amount of confident narration, and it disarms the comment that was going to be written anyway.
- The landing. Return to the hook, changed. The image or fact you opened with should mean something different now that the argument has happened. Do not summarize; a viewer who watched the video does not need it recapped in the last ninety seconds.
Hold that shape loosely once the draft starts arguing back. Our guide to outlining covers that tension in depth.
Write for the ear, not the page
Narration is heard once, in order, with no ability to scroll back. That single constraint changes almost every sentence-level decision.
- Short sentences carry, long ones fall over. A subordinate clause that reads elegantly on a page becomes a listener holding their breath waiting for a verb. When a sentence runs long, split it.
- Front-load the subject. "In 1974, after two failed attempts and a change of studio, Coppola" makes the ear wait. "Coppola tried twice and failed twice" arrives.
- Signpost the transitions. A reader sees a section break. A listener gets nothing unless you give it to them: "That is the version everyone knows. Here is the part the documentary left out."
- Numbers need shape. "Roughly one in three" beats "34.2 percent" when the ear is the only receiver. Save the precise figure for the on-screen text, which is exactly what the visual lane is for.
- Read it aloud, standing up. Everything you cannot say without stumbling is a sentence to rewrite, not a delivery problem to power through. This is the most reliable revision pass in the form and it costs nothing.
Timing follows from this. Most narrators land around 140 to 160 words a minute, so a twelve-minute essay is roughly 1,700 to 1,900 words of voiceover. Use that to check proportion, not to hit a target. A script padded to clear a runtime is a script viewers leave.
The second lane: writing what the viewer sees
A video script has two columns for a reason. Narration alone is a podcast, and the fastest way to make a fourteen-minute essay feel like forty is to let the visuals turn into wallpaper while the voice does all the work.
Write the visual lane as you write the voice lane, not afterward. Beside each block of narration, note what is on screen: the clip, the still, the on-screen text, the graph. Two habits pay for themselves:
- Let the picture make the point the voice does not. If the narration says the building was enormous and the screen shows the building being enormous, one of them is redundant. Cut the line, keep the shot.
- Mark your silences. Deliberate gaps in narration are pacing, and they only survive if they are in the script. Unwritten silence gets filled in the edit.
Settle the boring practical question early: whether you can legally use the footage a section is built around. That is a rights question rather than a craft one, and far cheaper to answer before the script depends on forty seconds of somebody else's film.
For a channel, accuracy is the whole game
Prose gets a private reader. A video gets a public one, with a comment box attached and an algorithm that rewards whoever corrects you. One wrong date, one misattributed quote, one number repeated from another video that got it wrong first, and the top comment under your work is a correction. It will outrank your argument for as long as the video exists.
You also cannot quietly fix it. A typo in a manuscript is a patch. A wrong claim in narration is a re-record, a re-edit, and a re-upload that resets the video's life. So the checking has to happen before you sit down at the microphone.
Check five things specifically: names and titles, dates and chronology, numbers (especially any figure you took from another video), quotations word for word against the primary source, and attributions, meaning who actually said or did the thing. Keep the source beside the claim in the script as you go, so when a commenter challenges you in six months the receipt is already filed. It is the same method working novelists use, laid out in our guide to fact-checking a draft.
One warning specific to this form: a general-purpose chatbot is a bad checker, because it draws on the same fuzzy memory that let the error in and will happily confirm a plausible wrong date. If a tool cannot show you a source, treat what it says as a rumor.
How Skarvia fits
Skarvia treats a channel as a project, and video scripts open in a two-lane editor: voiceover on one side, visuals on the other, with a running-time estimate updating as you write. So the script is timed while you are still shaping the argument, instead of at 2am in the edit.
The Review tab is where the staff lives. Check Facts reads the script, pulls out every checkable real-world claim, verifies each against retrieved evidence, and then a second reviewer whose only job is to refute those findings tries to tear them apart before you see any of them. What comes back is a punch list with sources, or one quiet line: "No issues found." There is no praise channel. Research runs the other direction: highlight a passage, ask the question you actually have, and the cited answer files itself into the project's research library for next time. The Codex keeps recurring names, dates, and facts straight across a whole run of videos.
Skarvia never writes a word for you. It reads, checks, and remembers. The argument stays yours, which is the only part anybody subscribed for.