# Ad Creative
Use this skill to produce paid ad copy that can survive platform constraints and creative testing.
The goal is not only "write ads." The goal is to create distinct angles, validate them against platform limits, make the next test clear, and feed results back into `/marketing-learning-loop`.
Context Loading
Before asking questions, check whether one of these files exists and read it:
- `.agents/product-marketing-context.md`
- `.claude/product-marketing-context.md`
- `references/facebook-ads-playbook.md` when the work is for Conversint's own Meta/Facebook demand tests
- `references/paid-ads-audit-engine.md` when diagnosing creative performance, Meta creative diversity, fatigue, clustering risk, or whether new variants are distinct enough to test
- `references/digital-marketing-funnel-best-practices.md` when the creative promise, destination CTA, or give-first offer is part of a paid lead path
For Conversint-branded work, also read `references/marketing-message-governance.md`. When an ad adapts a testable Conversint message, validate and consume the active message catalog; do not invent a replacement message from a prior channel example.
Ask only for missing context that affects creative quality or compliance.
Intake
If not clear, ask for:
- Platform and format: Google RSA, Meta feed, LinkedIn sponsored content, TikTok, X, display, video
- Product or offer: trial, demo, lead magnet, audit, consultation, event, service
- Audience and awareness stage
- Core value proposition and proof
- Existing performance data, if iterating
- Brand voice, banned words, and compliance constraints
If the user needs output quickly, assume the most likely format and clearly label the assumptions.
Modes
Generate From Scratch
- Define 3-5 creative angles.
- Ensure the angles are meaningfully distinct across concept, format, visual direction, hook, and headline structure.
- Generate variations for each angle.
- Validate character limits using `references/platform-specs.md`.
- Deliver in a structured format the user can paste into the platform.
Iterate From Performance Data
- Identify top and bottom performers by the metric the user cares about.
- Extract winning themes, structures, words, and length patterns.
- Score whether the current set is truly diverse or mostly near-duplicates.
- Generate new variants that extend winners and test 1-2 new angles.
- Include an iteration log.
- Record what changed and what result would count as learning.
Visual / Video Creative Direction
For image or video concepts, read `references/generative-tools.md`. If actual imagery is needed, use the image-generation routing in that reference: prefer Higgsfield for specialized workflows, and fall back to the runtime's available image generator for static images when Higgsfield is unavailable or unnecessary.
Creative Angles
Choose angles before writing copy:
- Pain point
- Outcome
- Speed
- Cost of inaction
- Social proof
- Comparison
- Curiosity
- Identity
- Contrarian belief
- Risk reduction
- Human touch
- Operational clarity
Creative Diversity Score
When the platform is Meta, TikTok, YouTube, or any algorithmic feed, score each creative set from 0-10 using `references/paid-ads-audit-engine.md`.
Assess:
- Concept diversity
- Format diversity
- Visual diversity
- Hook diversity
- Headline diversity
If the score is below 8/10, do not solve the problem by making more near-duplicates. Rebuild the weakest dimension first and state which one changed.
Output Format
Standard
```markdown ## Assumptions - Platform: - Offer: - Audience:
## Angles 1. [Angle] - [why it should work]
Creative
### Angle: [Name] #### Headlines ([limit]) 1. "[copy]" ([count])
#### Primary Text / Descriptions ([limit]) 1. "[copy]" ([count])
## Recommended Test - [Which variants to launch and what to measure]
## Creative Diversity - Score: - Weakest dimension: - What changed: ```
Bulk
When the user asks for 10+ variations, offer CSV output.
```csv platform,angle,headline,primary_text,description,cta ```
Iteration
```markdown ## Performance Summary - Analyzed: - Top performers: - Underperformers: - Winning patterns:
## New Variations [copy]
## Iteration Log - New angles: - Angles retired: - Metric to watch: ```
Quality Rules
- Validate every line against platform limits.
- Give trimmed alternatives for anything over limit.
- Keep RSA headlines independent; they may appear in any combination.
- Vary angles, not just synonyms.
- Vary creative concepts, not only sentence wording. Near-duplicate ads do not create a clean test.
- Avoid unsupported superlatives such as "best" or "leading."
- Do not use all caps or fake urgency.
- Match the landing page promise. Do not create clickbait the destination cannot satisfy.
- For regulated industries, avoid rates, guarantees, and advice claims unless the user provides approved language.
- When performance results exist, summarize the winning or losing angle for `/marketing-learning-loop` and flag whether the result is a creative signal, audience signal, or offer signal.
Reference Files
- `references/platform-specs.md` - paid platform character limits and format notes.
- `references/generative-tools.md` - visual/video creative production workflow and tool choices.
- `references/paid-ads-audit-engine.md` - creative diversity scoring, fatigue diagnosis, Meta demand-test scorecard, and paid-ads loop routing.