Ad Research

Step 1: Define Research Scope

Gather from user or context:

RESEARCH SCOPE:
  niche: <industry/product category>
  competitors: [<competitor 1>, <competitor 2>, ...]
  platforms: [meta, tiktok, google]  # which ad libraries to search
  keywords: [<keyword 1>, <keyword 2>, ...]  # search terms
  date_range: last_90_days  # how far back
  max_ads_per_competitor: 50

Step 2: Discover Competitor Ads

Using brave-search + tavily MCP:

1. Search for "[competitor name]" site:facebook.com/ads/library for each competitor
2. Search for "[competitor name]" ad creative to find coverage and breakdowns
3. Search for "[niche keyword]" best performing ads 2026 for angle discovery
4. Search for competitor landing pages to understand positioning

Using playwright MCP (if available):

1. Navigate to Meta Ad Library (facebook.com/ads/library)
2. Search by advertiser name or keyword
3. Filter by active ads, sort by longest running (oldest = most profitable)
4. Capture ad creative details: copy, headline, CTA, media type

Key principle: Sort by longevity. Ads running for 60+ days are likely profitable. Ads running for 7 days are still being tested.

Step 3: Analyze Each Ad

For each discovered ad, extract:

## Ad Analysis: [Competitor] — Ad #[N]

**Duration Running:** [X days/weeks/months]
**Platform:** [Meta/TikTok/Google]
**Format:** [Video/Image/Carousel]
**Estimated Profitability:** [High/Medium/Low] (based on run duration)

### Hook (First 3 seconds)
- Type: [Question/Statement/Pattern-interrupt/Story/Statistic]
- Text: "[exact hook text]"
- Why it works: [analysis]

### Angle
- Primary angle: [Pain point/Aspiration/Social proof/Authority/Urgency]
- Target emotion: [Fear/Desire/Frustration/Curiosity/FOMO]
- ICP signal: [who this ad targets based on language/imagery]

### Structure
1. Hook: [what grabs attention]
2. Problem: [what pain point is highlighted]
3. Solution: [how the product is positioned]
4. Proof: [testimonials/stats/demonstrations]
5. CTA: [what action is requested]

### Body Copy
- Headline: "[exact headline]"
- Primary text: "[exact body copy]"
- CTA button: "[exact CTA text]"

### Transcript (if video)
"[word-for-word transcript]"

### Key Takeaways
- What makes this ad work: [specific reasons]
- Angle reusability: [High/Medium/Low]
- Adaptation notes: [how to adapt for our product]

Step 4: Pattern Synthesis

After analyzing all ads, produce:

# Competitive Ad Intelligence Report

## Research Summary
- Competitors analyzed: [N]
- Total ads reviewed: [N]
- Date range: [start] to [end]
- Platforms: [list]

## Winning Patterns (sorted by frequency across competitors)

### Top Hooks
| Rank | Hook Type | Example | Competitors Using | Avg Duration |
|------|-----------|---------|-------------------|-------------|
| 1 | [type] | "[example]" | [N] | [X days] |

### Top Angles
| Rank | Angle | Description | Success Signals |
|------|-------|-------------|----------------|
| 1 | [angle] | [what it is] | [why we think it works] |

### Top Ad Structures
| Rank | Structure | Description | Best For |
|------|-----------|-------------|----------|
| 1 | [structure] | [steps] | [which ICPs] |

## Competitor Breakdown
[per-competitor summary: what they're running, their main angles, estimated spend level]

## Recommended Angles for Our Product
| Priority | Angle | Based On | Adaptation Notes |
|----------|-------|----------|-----------------|
| 1 | [angle] | [which competitor ads] | [how to adapt] |

## Anti-Patterns (What to Avoid)
- [pattern that multiple competitors tried and abandoned]

Output Format

The skill produces a Competitive Ad Intel Dump — a structured document that feeds directly into ad-script-generator as input.

Tools Used

Agents

UAOP Integration

• Stage 1 output feeds into Stage 3 (ad-script-generator)
• Stage 5 (ad-performance-loop) feeds winners back into Stage 1 scope
• Research results stored in memory MCP for longitudinal tracking
• Weekly cadence: run every Monday to catch new competitor launches

Scheduling

Cadence: Weekly (Monday)
Trigger: Manual or scheduled via GAOS pipeline
Input: Competitor list + keywords from growth/domain.md
Output: Competitive Ad Intel Dump → stored in memory + fed to ad-script-generator