Ad Creative Variants

Ad Creative Variants (A/B Batch)

Generate 3–5 distinct Meta static creatives from one store listing for structured creative testing. Each variant isolates a different angle (message) and/or visual format (style + preset).

One creative teaches you nothing. A disciplined batch tells you which hook and layout wins before you scale spend.

Initial Assessment

1. Read app-ads-context.md — target CPA, geo, documented winning/losing formats
2. Check competitor-ad-teardown output if available — borrow hook patterns, not copy
3. Confirm monthly budget tier (determines batch size)

Batch Size by Budget

Rule: Each ad needs enough spend to reach decision threshold (~2× target CPA before kill). Smaller budgets → fewer variants.

Variant Matrix (default 5)

Customize angles from app-ads-context.md value proposition and competitor teardown gaps.

3-Variant Minimal Batch

When credits or budget are tight:

Angle Library (fill per app)

Workflow

Step 1 — Confirm Inputs

Step 2 — Analyze Once (1 API credit)

generate_app_ad_creative
  app_url: "https://apps.apple.com/us/app/id123456789"
  mode: "analyze"
  generate_image: false

Present result.keyPoints. User approves or edits. Reuse confirmed keyPoints as MCP key_points on all variant jobs for consistent positioning.

Step 3 — Launch Parallel Jobs

One job per variant. Run in parallel when MCP supports concurrent calls.

# Variant A — Problem
generate_app_ad_creative
  app_url: "https://apps.apple.com/us/app/id123456789"
  mode: "generate"
  style: "problem_solution"
  image_preset: "problem_solution_split"
  angle: "For busy professionals drowning in tasks — one app to plan your day"
  key_points: { ...confirmed }
  quality: "medium"

# Variant B — UGC
generate_app_ad_creative
  app_url: "https://apps.apple.com/us/app/id123456789"
  mode: "generate"
  style: "ugc"
  image_preset: "person_holding_phone"
  angle: "POV: you finally found an app that actually sticks"
  key_points: { ...confirmed }
  quality: "medium"

# ... repeat for C, D, E

REST equivalent: POST /v1/app-ad-creatives/generate per variant → collect jobIds.

Step 4 — Poll All Jobs

get_app_ad_creative_job
  job_id: "<variant-a-job-id>"

Poll every 3–5 seconds per job. Track status in a table while waiting.

If one job fails, retry that variant only — do not restart the full batch.

Step 5 — Present Comparison

Use output template. Highlight differentiated hooks, not just images.

Meta Test Structure

Run all variants in one ad set for fair auction comparison:

Campaign: App Promotion (PAUSED)
└── Ad Set: US - iOS - Creative Test
    ├── Ad: Variant A - Problem
    ├── Ad: Variant B - UGC
    ├── Ad: Variant C - Premium
    ├── Ad: Variant D - Transform
    └── Ad: Variant E - Social proof

Do not split variants across ad sets in the first test — you lose comparative signal.

Decision Rules (during test)

Full audit framework: meta-campaign-audit.

Credit Estimate

Inform user before running:

BYOK (X-OpenAI-Key): 1 API credit per job, 0 creative credits.

Output Template

# Ad Creative A/B Batch — [App Name]

**Batch date:** [date]  |  **Country:** [us]  |  **Quality:** medium

## Positioning (shared keyPoints)
- Pain: ...
- Promise: ...
- Audience: ...

## Variant Comparison

| Label | Angle | Style / Preset | Image | Headline | Primary (excerpt) | Chars |
|-------|-------|----------------|-------|----------|-------------------|-------|
| A | Problem | problem_solution / split | [url] | | | |
| B | UGC | ugc / phone | [url] | | | |
| C | Premium | professional / mockup | [url] | | | |
| D | Transform | before_after / split | [url] | | | |
| E | Social proof | lifestyle / testimonial | [url] | | | |

## Differentiation Check
- [ ] No two variants share the same hook line
- [ ] At least 2 different presets used
- [ ] Primary text lengths within Meta limits

## Meta Test Plan
- **Ad set:** 1 ad set, all variants, $[X]/day
- **Kill rule:** Pause at 2× CPA, 0 conversions
- **Scale rule:** +20%/day on winner when CPA < target
- **Review:** 48h → meta-campaign-audit

## Next Steps
1. Publish → ad-creative-to-meta (all variants)
2. Winner iteration → ad-creative-edit (hook tweaks on winner preset)
3. Loser post-mortem → note angles to retire

Variant Naming Convention

Use consistent ad names in Meta for clean audits:

[App] - [Geo] - Var[A-E] - [AngleSlug] - [YYYYMMDD]

Example: FocusApp - US - VarB - UGC - 20250629

Benchmarks (batch-level)

If entire batch fails, diagnose with failure matrix in meta-campaign-audit — may be store/onboarding, not creative.

Common Mistakes

Pre-Batch Checklist

Planning:
- [ ] app-ads-context.md has target CPA
- [ ] Competitor teardown reviewed (optional)
- [ ] Batch size matches budget tier
- [ ] User approved credit estimate

Generation:
- [ ] Analyze completed, keyPoints confirmed (pass as `key_points`)
- [ ] Each variant has unique angle string
- [ ] All jobs polled to completed
- [ ] Failed variants retried or documented

Meta setup:
- [ ] mmp-setup complete
- [ ] Single ad set test structure
- [ ] Ad names follow convention
- [ ] Campaign starts PAUSED

Related Skills

• meta-ad-creative — single creative generation
• ad-creative-edit — winner iterations (A′, B′)
• ad-creative-to-meta — publish batch to Meta
• competitor-ad-teardown — angle research input
• meta-campaign-setup — ad set settings
• meta-campaign-audit — pick winner after 48h
• meta-budget-optimizer — scale winner
• tiktok-creative-strategy — parallel video batch for TikTok