The script

Paste this into a chatbot with your product and it can write the script. The constraints below are the whole spec.

The writing prompt

Write a {duration}-second spoken script for a {format} clip about {product}.

STRUCTURE
- Open on an arguable stance, landed inside the first two sentences. Not a
  result, not a promise. A position somebody could disagree with.
- Pick ONE move and escalate it every beat: escalating specificity, deepening
  sincerity, chronological receipts, or push-pull. No topic changes.
- Put the strongest beat immediately before the end.
- Include exactly one silent physical beat: a sip, an eye-roll, a pause.
- Mention the product 2–3 times max, as receipts inside the story. Never its
  own section.
- End on a completed beat, then stop. No moral, no summary, no callback.

SENTENCE RULES
- 15–20 words per sentence, one idea each.
- Never more than two commas in a sentence.
- Exactly one named third party. After that use he/she.
- Kitchen-table verbs only: "he Venmo'd me", never "he requests".
- No invented hyphen compounds like "scary-good".
- Quoted text-speak ("wyd", "k") at most once, never last.
- No fragment stacks. Not "Thursday. Wine bar." Specificity rides inside
  flowing clauses with connectors.

LENGTH: {18-24 short / 28-36 med / 45-55 long / 62-70 xlong} words.
Budget one beat per 5 seconds of clip.

Output the spoken words only. No stage directions, no scene description.

Length by tier

The engine speaks at about 2.7 words per second regardless of what you ask, so extra clip seconds become dead air at the end rather than slower delivery. When something feels rushed it's claim density, not speed.

When a script needs to be shorter, cut story beats. Never compress the telling, because compression is what produces the fragment stacks that read as written rather than spoken.

Winners run 215 to 500 words at 60 to 180 seconds.

The opening

An arguable stance, not a result and not a promise.

The clips that hold open on something you could argue with. A result-first opener gives a viewer nothing to react to, so they leave while you're still setting up.

If nobody could disagree with your first two sentences, nobody's staying.

Worth watching for this specifically: street opens on a question with an obvious wrong answer, and podcast opens mid-disagreement.

The ending

The final beat completes, then it cuts.

Both failure directions are common. Concluding it with an aphorism or a stance restate or a peak image is still landing the plane, and it reads as an ad. Cutting mid-sentence to avoid that reads as an editor who cut too early, which is worse.

This passes: *"Fine. Agree to disagree. You're still wrong though."* Settled without being resolved, so there's still something to argue with.

Check it before it goes to voice

Read it out loud at normal speed without stopping. Anywhere you restart a sentence or need a second breath, split it there. Count commas, count named people, find the move and check every beat is that same move. Then read the last line and ask whether a person would say it.

Half the problems only show up when your mouth is doing the work, because a listener can't go back and re-read. Miss it once and it's gone.