gooseworks-ai/goose-skills / skills/video/write-video-ad-script
write-video-ad-script
VerifiedWrite or revise a short-form ad from a sourced angle and reference. Plan connected words, product proof and sound; expose production gaps before script review.
references/dialogue-writing.md
View on GitHub# Podcast and street-interview dialogue
**Summary.** A conversation needs two people responding to each other. Product facts
authorize claims; the brand message determines the idea; observed conversations teach
turn structure. Establish that message before selecting examples. Write the exchange
before fitting timing and visuals. A feature list split between speakers is a failed
conversation even when every fact is correct.
## Establish what the ad needs to communicate
Start with the brand's positioning, exact product, audience and objective. State what the
viewer should understand, why it matters, and the supported role of this offering. For a
purchase-oriented ad, explain a relevant reason to choose it. A fact about preparation can
serve a requested FAQ, but is not automatically a persuasive lead message.
Choose a premise, then find examples that help express it. Do not choose a tasting clip
because both products are drinks and reshape the brief around it. Check the source's
buyer question, objection, proof and payoff as well as its conversation mechanics. A
reference may teach delivery while being unsuitable for the brand's central argument.
Reject that transfer rather than adding more natural wording to an irrelevant premise.
## Gather the right reference
Reuse one or two inspected same-format sources from the angle research or project.
Prefer the brand's existing work, then relevant category creative. Read or transcribe
a complete exchange, not just its hook or caption. Record who initiates, what the
other person actually answers, what prompts the follow-up and how the exchange ends.
Use short quotations only when needed; preserve source links and paraphrase mechanics.
Do not invent an observation. A publisher transcript supports words and turn order;
it does not establish vocal delivery or camera reactions. Sampled frames establish
visible context at those times, not continuous motion or what the observer heard.
For street ads, use a **complete commercial interaction**. Inspect the full visual
timeline and spoken exchange, from the published opening through the ending. Record
the visible setup, why the person engages, hook, relevant brand explanation or action,
and payoff. Keep timestamped words and actions, label paraphrases, and link full transcript
evidence. If recruitment or the tasting invitation is outside the edit, say it is unknown;
do not invent an observed approach. Match product/service, interaction mechanics and
persuasive purpose against the brand message already established.
A sampling ad is not automatically a service reference. An editorial radio excerpt,
hook-only snippet, or isolated funny answer cannot satisfy the street-ad reference gate.
Fetch and inspect a suitable ad when the library has this gap.
For podcasts, a relevant editorial exchange may provisionally supply conversation
mechanics. Its authority, claims, reputation and performance do not transfer. Record
missing commercial/delivery evidence. A real ad is not verified human-written simply
because real people appear in it.
## Write an exchange, then an ad
Load [human speech and behavior](human-behavior.md). Use matched parent/reply evidence
and explicit human feedback to establish motives, bounded knowledge and local responses.
An openly promotional exchange can sound human; hiding the advertisement is not the goal.
First load [scene and hook](scene-and-hook.md). Plan the encounter and the opening the
viewer receives separately, then map development and payoff. A reply to an omitted
question must still have an identifiable subject. Review the first 0–3 seconds without
the brief before reviewing the whole conversation. Private scene notes cannot supply
context that is absent from the actual words, action or rendered text.
- Define what each person wants from this moment. Give the second person a role beyond
prompting the sales pitch.
- Start with a concrete question, observed action or disagreement. Let the answer change
the follow-up. Use the reply's detail instead of moving down a prepared feature list.
- Vary turn length where it serves the exchange. A clarification can be two words;
an explanation may need more. Do not add filler, fake laughter or interruptions as
decoration.
- Keep the product explanation selective. Demonstrate one useful detail. Put supporting
specifications, website and required copy in an insert or ending where appropriate.
- Preserve a real speaker's testimony only with attribution and authorization. Staged
dialogue can discuss a problem or present an observable action, not invent purchase
history, a professional credential or a product result.
## Distinguish the formats
**Podcast:** the hosts may know the topic but differ in what they assume or care about.
A skeptical follow-up or correction should change the explanation. A clip can start
mid-discussion if the opening is still intelligible. Do not simulate a named host,
expert endorsement or genuine unsponsored episode.
**Street interview:** ask something an unprepared person could answer. A response may
be mundane, practical or unexpected. Follow that answer. When an actor demonstrates a
product, keep the reaction about what is observable, not a made-up review.
Disclose a staged ad; do not pretend it is consumer research.
Write a situation brief before dialogue: participant role, interviewer purpose, invitation
or task, each person's intention, visible opening, and relevant brand help/payoff. A
plausible independent concern must survive an ordinary person's next obvious action.
If the problem disappears through a trivial action, do not manufacture an elaborate
emotional response to it. Keep a natural edited reaction opening when the viewer can
understand the object or task. A greeting is optional in the edit; causal context is not.
For a service, the help can be explained verbally after a concrete task is established.
For a sample, leave real time to taste before speaking and do not infer an immediate
health effect. Generic problem talk plus a detached ending is insufficient brand work.
## Review the words and the situation
Read only the words as an alternating exchange. For each turn ask:
1. Why would this person say it now?
2. What in the previous turn caused it?
3. Would someone use this wording out loud?
4. Is it explaining a useful detail or volunteering a marketing slogan?
Reject convenient setup questions, unprompted spec recitals, identical sentence rhythm,
rehearsed admiration and slogan-shaped answers. For example, “What would you change
first?” followed by a catalogue sentence gives no evidence of actual listening.
Contractions alone do not make it human.
Then read the words **with** the visible setup and actions. Can this person reasonably
participate and answer now? Does the question follow the visible task or previous reply?
Does the brand address that specific situation, with a useful explanation or action?
Is there time to do it? Reject a structurally valid exchange when these fail. Scene notes
must not rescue incomprehensible words, and a words-only score must not ignore the scene.
Treat 4/5 as usable with minor delivery edits and 5/5 as ready to perform as written.
Natural language and conversational logic must each clear the bar; their average must
not hide a weak exchange. A model score still needs calibration against operator
rejections and accepted examples. Never relabel an earlier metric to manufacture a pass.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| receiving-code-reviewobra/superpowersUse when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementation | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| requesting-code-reviewobra/superpowersUse when completing tasks, implementing major features, or before merging to verify work meets requirements | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| using-git-worktreesobra/superpowersUse when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallback | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| verification-before-completionobra/superpowersUse when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |