---
name: video-ad-creative
description: Plan, brief, and evaluate performance video ads for paid social (Meta, TikTok, Reels, Shorts). Covers the 3-second hook, UGC-style framing, creative testing cadence, and reading engagement metrics as a signal for what to scale, iterate on, or kill. Use when writing ad scripts, briefing video creators, auditing an ad account's creative, or deciding what to do with live creative data.
---

# Video Ad Creative

## Purpose

Turn a product and offer into performance video ads, and grade live creative so winners get scaled and losers get killed fast. The working unit is the creative, not the campaign: on modern ad platforms the creative is the targeting, so this skill treats ads as hypotheses that get tested, measured, and replaced on a schedule.

## Workflow

1. **Intake.** Get the product, the one-line promise, the audience, and the single biggest objection the ad must beat. One angle per video, never a bundle of messages.
2. **Pick angles.** Choose 3+ fundamentally distinct angles, not minor variations. Candidates: pain/frustration, speed or time saved, social proof, product demo, objection killer. See `references/hooks.md` for patterns per angle.
3. **Write hooks first.** For each angle, draft 5+ hook lines built for the first 3 seconds. The hook carries the most weight in any test, so isolate it: test hooks before bodies, bodies before CTAs.
4. **Structure the cut.** Standard 30-60s skeleton: Hook (0-3s) / Pain or setup (3-10s) / Discovery or demo (10-25s) / Proof (25-40s) / CTA (40-60s). Shorter cuts compress the same order; never open with a logo card or a greeting.
5. **Brief the creator.** One angle per video. Shot list with timings, required claims and legal lines, and a clear line between the brand's constraints and the creator's creative freedom. Template in `assets/brief-template.md`.
6. **Grade live creative.** Feed measured metrics into `bin/grade_creative.py` to get a per-creative verdict: Scale, Iterate (with which stage is broken), or Kill, plus the next test to run.
7. **Iterate winners, replace losers.** Same concept with a new hook; same hook with new visuals; number swaps on stat hooks. Budget posture: 70% scaling proven winners, 20% promising new tests, 10% experiments. Watch for fatigue (frequency climbing while CTR falls) and rotate hooks before reshooting.
8. **Preserve social proof when scaling.** Move the winning Post ID into scaling campaigns rather than recreating the ad from scratch, so likes, comments, and shares carry over.

## Output Contract

- For concept work: an angle table (angle, why this angle could win, which objection it beats), hook lines per angle, and one completed creator brief per video.
- For live accounts: per-creative verdict with the metric evidence behind it and one exact next action per creative. Never a bare "looks good" and never a verdict without numbers.
- All numbers come from real measurement. Benchmarks in `references/metrics.md` are practitioner ranges, not platform-published figures; compare against the account's own trailing median first.

## Operating Rules

1. Never scale before testing. Never kill a creative before minimum data: at least 48 hours or 5 conversions on a low-budget test.
2. Every video must be understandable with the sound off: burned-in high-contrast captions, key text inside platform safe zones (see `references/specs.md`).
3. Keep UGC native: be specific on constraints (claims, legal, brand voice) and hands off on execution. Over-briefing kills authenticity.
4. Do not claim to know what a platform algorithm does internally. Describe what the test will reveal instead.
5. If a claim cannot be verified from a primary source or real measurement, cut it or frame it as opinion. This applies to copy inside the skill and to copy you write for the user.