A short-video talking-head script prompt solves one very specific snag: you have a topic, but you do not know what to say first, and whatever draft you write either reads like a school essay or runs twice as long as planned once you say it out loud. The prompt below hands over the topic, audience, duration and persona in one go, so the AI returns a script you can actually read to the camera: a hook up front, body sections timed in seconds, and a total word count matched to your target length.
What separates it from casually asking an AI to "write me a short-video script" is the constraints. Without a duration-to-words conversion, models happily produce a 500-word "60-second script". Without a sentence-length limit, you get long written sentences that leave you gasping mid-take. Put the constraints in the prompt and the first draft lands close to filmable; only fine-tuning remains.
The prompt (copy and use)
You are a director for short talking-head videos. Write me a script I can read straight to the camera on my topic, and follow these five rules strictly:
1. Duration and length: target duration [duration, default 60 seconds]. Convert at roughly 140 words per minute, stay within 10 percent either way, and state the total word count at the end.
2. Opening hook: open the first 3 seconds with a question, a counter-intuitive claim or a concrete number. It must match what the body actually concludes; never promise something the script cannot deliver.
3. Body structure: output the script in timed sections, each labelled with approximate seconds. One idea per section, no sentence longer than 15 words, short spoken sentences only, no bookish phrasing or rare words.
4. Ending: finish with exactly one clear call to action, chosen from: comment below, save this for later, or follow for the next part. Never stack all three.
5. Facts: for data, years and names, use only what I provide in the materials below. Anything not covered there must be marked [to verify]; do not invent it.
My topic is [one sentence stating the topic], aimed at [target audience]. My on-camera persona is [persona and tone, e.g. eight years in operations, blunt, not selling courses]. The one line I want viewers to remember is [core takeaway].
Source material:
[paste research, personal experience or product facts here; delete this line if you have none]How to swap the variables
| Variable | What to put in | Example |
|---|---|---|
| [duration] | How long the video should run; it sets the total word count | 60 seconds is about 140 words; a 3-minute deep dive is about 420, and note whether your pace is slow or fast |
| [topic] | One sentence, one subject only | "Three common reasons landlords give for keeping a deposit, and why they do not hold up" beats "let us talk about renting" |
| [target audience] | Who would stop scrolling, described by situation, not just demographics | First-time renters fresh out of college, not a vague "everyone" |
| [persona and tone] | How you actually talk on camera | The same topic sounds like a legal explainer from a lawyer and like peer advice from a parent; the scripts diverge completely |
| [core takeaway] | The single line the whole video exists to deliver | Every section serves this line; if you cannot write it yet, do not generate the script yet |
Three follow-ups once you have a draft
First, to fix the length: "Merge sections 2 and 3 and cut the script to under 140 words. Do not change the takeaway line at all." Overruns usually come from the middle circling the same point twice; merging beats trimming word by word.
Second, to redo the hook: "Rewrite only the opening. Give me three different first 3 seconds: one question, one counter-intuitive claim, one concrete number. Leave the body untouched." The opening decides whether people stay, so asking for spares is cheaper than regenerating everything.
Third, to remove the written tone: "Split any sentence that is awkward to say out loud into shorter ones, swap bookish words for words people actually say, and list every change you made." Listing the changes shows you exactly what moved, so your persona does not get edited away in the process.
Three mistakes that ruin the result
First, swapping the topic but never the duration. A 60-second template stretched to 3 minutes gets padded with filler; a long template squeezed into 30 seconds crams so much in that you sound like you are catching a train. Whenever the duration changes, make the model redo the word-count conversion before it writes.
Second, a hook stronger than the body. Opening with "99 percent of people get this wrong" and then failing to back it up does not just lose viewers; it invites arguments in the comments. Test it this way: lift the opening line out and ask which section answers it. If none does, tone the claim down.
Third, leaving the persona blank. Blank means the model defaults to the safest generic explainer voice, identical for every creator, and your video ends up with no identity. Even half a sentence, like "chatty, talks like to a friend, occasionally self-deprecating", pulls the script noticeably toward how you really sound.
