ad-angle-miner（广告角度矿工）

Ad Angle Miner

Dig through customer voice data — reviews, Reddit, support tickets, competitor ads — to extract the specific language, pain points, and outcome desires that make ads convert. The output is an angle bank your team can pull from for any campaign.

Core principle: The best ad angles aren't invented in a brainstorm. They're extracted from what real people are already saying. This skill finds those angles and ranks them by strength of evidence.

When to Use

• "What angles should we run in our ads?"
• "Find pain points we can use in ad copy"
• "What are people complaining about with [competitors]?"
• "Mine reviews for ad messaging"
• "I need fresh ad angles — not the same tired stuff"

Prerequisites

• Environment variable: APIFY_API_TOKEN — required for review scraping and Reddit scraping
• Web search access — your AI agent must support web_search or equivalent for Twitter/X and competitor ad lookups

Phase 0: Intake

1. Your product — Name + what it does in one sentence
2. Competitors — 2-5 competitor names (for review mining)
3. ICP — Who are you targeting? (role, company stage, pain)
4. Data sources to mine (pick all that apply):

- G2/Capterra/Trustpilot reviews (yours + competitors) - Reddit threads in relevant subreddits - Twitter/X complaints or praise - Support tickets or NPS comments (paste or file) - Competitor ads (Meta + Google)

1. Any angles you've already tested? — So we can skip those

Phase 1: Source Collection

1A: Review Mining (Apify)

Use the Apify Amazon Reviews Scraper (or web_search for G2/Capterra/TrustRadius reviews).

Option 1: Amazon product reviews via Apify

Start a run of the web_wanderer/amazon-reviews-extractor actor:

POST https://api.apify.com/v2/acts/web_wanderer~amazon-reviews-extractor/runs?token=$APIFY_API_TOKEN
Content-Type: application/json

{
  "products": [
    "https://www.amazon.com/dp/PRODUCT_ASIN"
  ],
  "maxReviews": 100
}

Poll until the run finishes:

GET https://api.apify.com/v2/acts/web_wanderer~amazon-reviews-extractor/runs/{RUN_ID}?token=$APIFY_API_TOKEN

When status is SUCCEEDED, fetch results:

GET https://api.apify.com/v2/datasets/{DATASET_ID}/items?token=$APIFY_API_TOKEN

Output fields: Each review has rating (1-5), reviewTitle, reviewText, reviewDate, verifiedPurchase (bool), productAsin, productTitle, helpfulVoteCount.

Option 2: G2/Capterra/TrustRadius reviews via web_search

For B2B products, run web searches to find review content:

web_search: "<product_name> reviews site:g2.com"
web_search: "<product_name> reviews site:capterra.com"
web_search: "<product_name> reviews site:trustradius.com"
web_search: "<competitor_name> reviews site:g2.com"

Focus on:

• 1-2 star reviews of competitors — Pain they're failing to solve
• 4-5 star reviews of you — Outcomes that delight buyers
• 4-5 star reviews of competitors — Strengths you need to counter or match
• Review language patterns — Exact phrases buyers use

1B: Reddit/Community Mining (Apify)

Use the trudax/reddit-scraper-lite actor to search Reddit for relevant threads:

Search by keyword:

POST https://api.apify.com/v2/acts/trudax~reddit-scraper-lite/runs?token=$APIFY_API_TOKEN
Content-Type: application/json

{
  "searches": [
    "<product category> OR <competitor> OR <pain keyword>"
  ],
  "maxItems": 50
}

Browse a specific subreddit:

POST https://api.apify.com/v2/acts/trudax~reddit-scraper-lite/runs?token=$APIFY_API_TOKEN
Content-Type: application/json

{
  "startUrls": [
    {"url": "https://www.reddit.com/r/SUBREDDIT_NAME/hot/"}
  ],
  "maxItems": 50
}

Poll until complete:

GET https://api.apify.com/v2/acts/trudax~reddit-scraper-lite/runs/{RUN_ID}?token=$APIFY_API_TOKEN

Fetch results when status is SUCCEEDED:

GET https://api.apify.com/v2/datasets/{DATASET_ID}/items?token=$APIFY_API_TOKEN

Output fields: Each item has dataType ("post" or "comment"), title (posts only), body, communityName, upVotes, numberOfComments (posts), url, createdAt.

Extract:

• Questions people ask before buying
• Complaints about current solutions
• "I wish [product] would..." statements
• Comparison threads (vs discussions)

1C: Twitter/X Mining (web_search)

Use web_search to find relevant Twitter/X posts — no scraper or credentials needed:

web_search: "<competitor> (frustrating OR broken OR hate) site:x.com"
web_search: "<competitor> (love OR switched to OR replaced) site:x.com"
web_search: "<product category> (recommendation OR alternative OR looking for) site:twitter.com"
web_search: "<competitor> site:x.com" (for general sentiment)

Run 3-5 queries covering:

• Competitor complaints and frustrations
• Product category praise / switching stories
• "What do you use for X?" buying-intent threads

1D: Competitor Ad Mining (web_search)

Use web_search to check the Meta Ad Library for competitor ad creatives — no separate tool needed:

web_search: "<competitor_name> site:facebook.com/ads/library"
web_search: "<competitor_name> facebook ads library"
web_search: "<competitor_name> ad creative examples"

This reveals:

• Angles they've validated (long-running ads = working)
• Angles they're testing (new ads)
• Angles nobody is running (white space)

1E: Internal Data (Optional)

If the user provides support tickets, NPS comments, or sales call transcripts — ingest and tag with the same framework below.

Phase 2: Angle Extraction

Process all collected data through this extraction framework:

Angle Categories

For Each Angle, Extract:

1. The angle — One-sentence framing
2. Proof quotes — 2-5 verbatim quotes from sources
3. Source count — How many independent sources mention this?
4. Competitor weakness? — Does this exploit a specific competitor's gap?
5. Emotional register — Frustration / Aspiration / Fear / Relief / Pride
6. Recommended format — Search ad / Meta static / Meta video / LinkedIn / Twitter

Phase 3: Scoring & Ranking

Score each angle on:

Total score out of 100. Rank all angles.

Phase 4: Output Format

# Ad Angle Bank — [Product Name] — [DATE]

Sources mined: [list]
Total angles extracted: [N]
Top-tier angles (score 70+): [N]

---

## Tier 1: Highest-Conviction Angles (Score 70+)

### Angle 1: [One-sentence angle]
- **Category:** [Pain / Outcome / Identity / Fear / Displacement / Proof / Contrast]
- **Score:** [X/100]
- **Emotional register:** [Frustration / Aspiration / etc.]
- **Proof quotes:**
  > "[Verbatim quote 1]" — [Source: G2 review / Reddit / etc.]
  > "[Verbatim quote 2]" — [Source]
  > "[Verbatim quote 3]" — [Source]
- **Source count:** [N] independent mentions
- **Competitor weakness exploited:** [Competitor name + specific gap, or "N/A"]
- **Recommended formats:** [Search ad headline / Meta static / Video hook / etc.]
- **Sample headline:** "[Draft headline using this angle]"
- **Sample body copy:** "[Draft 1-2 sentence body]"

### Angle 2: ...

---

## Tier 2: Worth Testing (Score 50-69)

[Same format, briefer]

---

## Tier 3: Emerging / Low-Evidence (Score < 50)

[Brief list — angles with potential but insufficient evidence]

---

## Competitive Angle Map

| Angle | Your Product | [Comp A] | [Comp B] | [Comp C] |
|-------|-------------|----------|----------|----------|
| [Angle 1] | Can claim ✓ | Weak here ✗ | Also claims | Not relevant |
| [Angle 2] | Strong ✓ | Strong | Weak ✗ | Not relevant |
...

---

## Recommended Test Plan

### Week 1-2: Test Tier 1 Angles
- [Angle] → [Format] → [Platform]
- [Angle] → [Format] → [Platform]

### Week 3-4: Test Tier 2 Angles
- [Angle] → [Format] → [Platform]

Save to angle-bank-[YYYY-MM-DD].md in the current working directory (or user-specified path).

Cost

Tools Required

• Environment variable: APIFY_API_TOKEN — for Apify actors (review scraper, Reddit scraper)
• Web search — built into your AI agent (for Twitter/X, competitor ads, G2/Capterra reviews)
• No third-party libraries needed. All data collection uses HTTP APIs (requests or equivalent) and web_search.

Trigger Phrases

• "Mine ad angles from reviews"
• "What angles should we run?"
• "Find pain language for our ads"
• "Build an ad angle bank for [client]"
• "What are people complaining about with [competitor]?"