Meta Video Ad Deconstructor

**Judgement** A focused, well-scoped skill that is genuinely useful for marketers and growth teams who already have (or can obtain) transcripts, overlays, and scene descriptions for their video ads and who have access to Google Vertex AI / Gemini. It excels at turning messy ad content into structured, multi-dimensional marketing insight suitable for pipelines and dashboards. **What it does well** - Breaks down video ads into many marketing dimensions (hooks, social proof, urgency, emotion, target audience, CTAs, unique mechanism, etc.). - Produces clean, structured JSON that is easy to feed into downstream tools or analytics. - Offers a clear Python integration pattern and progress callbacks for longer analyses. - Allows prompt customization for teams that want to standardize how ads are scored and described. **Key limitations / risks** - **Upstream dependency**: It assumes you already have high-quality `ExtractedVideoContent` (transcript, text overlays, scene descriptions). If the upstream video-to-text step is noisy or missing elements, the analysis will be incomplete or misleading. - **Model dependence & hallucination risk**: Because it uses a general-purpose LLM (Gemini), there is a non-trivial risk of it inferring or fabricating marketing elements not clearly present in the source content, especially vague claims or social proof. Prompting can mitigate but not eliminate this. - **Platform lock-in**: Requires Google Cloud setup (service account, `vertexai`, Gemini). Teams not already on GCP or without Gemini access will face friction. - **Limited to text-level understanding**: It does not itself process raw video; effectiveness is bounded by the fidelity of the transcript and scene descriptions you feed it. **Recommended scenarios** Use this skill when: - You are systematically analyzing many competitor ads and need consistent, structured breakdowns. - You want to enrich ad datasets with labeled marketing attributes (hooks, urgency tactics, emotional angles). - You’re generating or reviewing creative briefs and want to quickly extract the marketing logic from reference ads. - You are building internal tools (dashboards, creative analyzers, knowledge bases) where structured JSON output is valuable. Use with caution or additional guardrails when: - Operating in regulated verticals where fabricated claims in analyses could mislead stakeholders. - Feeding it low-quality or partial transcripts, as this can produce overconfident but incomplete deconstructions. - Handling sensitive or proprietary ad content without verifying your data governance and GCP privacy posture.