engagement-miner

Engagement Miner

Finds engaged audiences on competitor posts, industry discussions, and thought-leader content. Extracts commenters and engagers from LinkedIn posts, Twitter threads, and blog comments. These are warm prospects who already care about the topic and are pre-qualified by their own engagement behavior.

Prerequisites

• agency.config.json at repo root with services, icp, and scoring sections
• WebSearch tool available
• Optional: person-researcher skill for enriching extracted prospects
• Optional: crm-writer skill for logging prospects to CRM

Phase 0: Intake

1. Read agency.config.json from the project root.
2. Extract:
   • services[].keywords -- topic relevance for filtering
   • icp.segments[].titles, icp.segments[].seniority -- prospect qualification criteria
   • icp.segments[].industries -- industry filters
   • icp.segments[].company_size -- company size filters
   • icp.primary_keywords, icp.secondary_keywords -- topic matching
   • icp.negative_keywords -- filter out competitors and service providers
3. Accept parameters:
   • source_posts -- list of specific post URLs to mine (default: none, discover automatically)
   • competitors -- competitor names/profiles whose posts to mine (default: none, user must supply)
   • topics -- topic keywords to find relevant posts (default: use config keywords)
   • platforms -- platforms to mine: linkedin | twitter | blogs (default: all)
   • max_prospects -- max prospects to return (default: 50)
   • min_engagement -- minimum engagement threshold for source posts (default: 10 comments)

Phase 1: Source Post Discovery

If source_posts not provided, discover high-engagement posts to mine.

LinkedIn post discovery:

• site:linkedin.com/posts "{competitor_name}" "{service_keyword}"
• site:linkedin.com/posts "{service_keyword}" "comments" OR "agree" OR "great point"
• site:linkedin.com/posts "{icp_industry}" "{intent_keyword}"
• Prioritize posts with visible comment counts > min_engagement.

Twitter thread discovery:

• site:twitter.com OR site:x.com "{competitor_name}" "{service_keyword}" "replies"
• site:twitter.com OR site:x.com "{service_keyword}" "thread" OR "unpopular opinion" "{icp_industry}"
• Look for threads with high reply counts.

Blog comment discovery:

• "{competitor_name}" blog "{service_keyword}" "comments"
• "{industry_keyword}" blog "leave a comment" OR "responses" "{service_keyword}"
• Target popular industry blogs with active comment sections.

For each source post, capture:

{
  "post_url": "URL",
  "platform": "LinkedIn | Twitter | Blog",
  "author": "Post author name",
  "author_company": "Author's company if visible",
  "topic": "Main topic of the post",
  "engagement_count": "Number of comments/replies visible",
  "posted_date": "Date if available",
  "relevance": "HIGH | MEDIUM"
}

Target 10-20 high-engagement source posts across platforms.

Phase 2: Engager Extraction

For each source post, extract people who engaged (commented, replied, shared):

LinkedIn engager extraction:

• Search: site:linkedin.com "{post_title}" "{commenter_snippet}" to find comment previews
• Search: site:linkedin.com/in "{keyword_from_post}" "{icp_title}" to find people who likely engaged
• Extract from search snippets: commenter names, their titles, companies
• Note: LinkedIn comments are not always fully indexed. Accept partial results.

Twitter engager extraction:

• Search: site:twitter.com OR site:x.com "replying to @{author_handle}" "{topic_keyword}"
• Search for quoted retweets: site:twitter.com "{post_url_fragment}"
• Extract: replier handles, names, bios if visible in snippets

Blog comment extraction:

• Use WebFetch on the blog URL (if available) to read comment sections
• Extract: commenter names, linked websites, comment content
• Commenter websites are valuable for company identification

For each extracted engager:

{
  "name": "Full name",
  "handle": "Social handle or username",
  "platform": "LinkedIn | Twitter | Blog",
  "title": "Job title if visible",
  "company": "Company name if visible",
  "profile_url": "Profile URL if available",
  "comment_snippet": "What they said (first 200 chars)",
  "source_post_url": "The post they engaged with",
  "source_post_topic": "Topic of the source post"
}

Phase 3: Prospect Qualification

Filter and score extracted engagers against ICP:

Title matching:

• Compare title against icp.segments[].titles and icp.segments[].seniority.
• Exact title match: +30 points
• Seniority level match: +20 points
• No title available: +5 points (benefit of the doubt)

Company relevance:

• If company matches ICP industries: +20 points
• If company size is estimable and within ICP range: +10 points
• If company is a known competitor (from competitors param): -100 points (disqualify)

Engagement quality:

• Comment shows pain point or need: +25 points
• Comment asks a question: +20 points
• Comment shares experience: +15 points
• Generic agreement ("Great post!"): +5 points
• Self-promotional comment: -50 points (disqualify)

Negative keyword filter:

• Check name, title, company, and comment against icp.negative_keywords.
• Disqualify any match.

Scoring thresholds:

• HOT PROSPECT (60+): Strong title match + relevant engagement
• WARM PROSPECT (35-59): Partial match, worth researching
• COOL (<35): Weak match, skip unless volume is low

Phase 4: Deduplication

1. Deduplicate by name + company combination (same person across multiple posts).
2. For duplicates, keep the entry with the richest data (most fields populated).
3. Merge comment snippets from multiple engagements into a single prospect record.
4. Deduplicate by profile URL if available (exact match).

Phase 5: Output

Return structured prospect list:

{
  "mining_summary": {
    "source_posts_analyzed": 15,
    "total_engagers_extracted": 120,
    "after_qualification": 50,
    "after_dedup": 42,
    "breakdown": {
      "hot": 8,
      "warm": 22,
      "cool": 12
    }
  },
  "prospects": [
    {
      "name": "Jane Doe",
      "title": "Head of Ecommerce",
      "company": "BrandCo",
      "platform": "LinkedIn",
      "profile_url": "https://linkedin.com/in/janedoe",
      "score": 75,
      "tier": "HOT",
      "engagement_context": "Commented on CRO post asking about checkout optimization",
      "comment_snippets": ["We've been struggling with cart abandonment..."],
      "source_posts": ["https://linkedin.com/posts/..."],
      "recommended_approach": "Reference their checkout concern, offer CRO audit insight",
      "next_step": "RESEARCH | CONNECT | EMAIL"
    }
  ]
}

Present formatted summary:

ENGAGEMENT MINING REPORT
Source posts analyzed: {N} across {platforms}
Total engagers found: {N}
Qualified prospects: {N} ({hot} HOT, {warm} WARM)

HOT PROSPECTS ({count}):
1. {name} -- {title} at {company} -- Score: {N}
   Context: "{comment_snippet}"
   Approach: {recommended_approach}
   Next: {next_step}

WARM PROSPECTS ({count}):
1. {name} -- {title} at {company} -- Score: {N}
   Context: "{comment_snippet}"
   Next: {next_step}

DISQUALIFIED: {count} (competitors: {a}, self-promoters: {b}, irrelevant: {c})

TOP ENGAGEMENT TOPICS:
1. {topic} -- {count} qualified engagers found

Phase 6: CRM Logging

If crm-writer is available and user approves:

• Log HOT and WARM prospects to CRM pipeline tab
• Columns: Date, Name, Title, Company, Platform, Source Post, Score, Tier, Context, Status
• Set initial status: "MINED"

Example Usage

Trigger phrases:

• "Mine engagement on competitor posts"
• "Find prospects from LinkedIn comments"
• "Extract engagers from this post: [URL]"
• "Who's engaging with Shopify CRO content on LinkedIn?"
• "Mine warm leads from competitor content"
• "Find people commenting on ecommerce topics"

User: Mine engagement from WebSavvy and Starter Labs LinkedIn posts about Shopify
Assistant: [reads config, discovers high-engagement posts by those competitors, extracts commenters, qualifies against ICP, scores and deduplicates, presents ranked prospect list with approach recommendations]

User: Extract prospects from these 3 LinkedIn posts: [URL1, URL2, URL3]
Assistant: [same flow but uses provided URLs directly as source posts, skips discovery phase]