Comment Insights Engine
Mine ad and organic comments for objections, intent language, and reply templates.
Comment Insights Engine
Mine ad and organic comments for objections, intent language, and reply templates.
Your best ad copy is hiding in your comments section. This skill pulls comments from your Meta pages via Glued MCP, clusters them into actionable themes (objections, desires, social proof, competitor mentions), and generates copy angles, FAQ entries, and reply templates — turning voice-of-customer data into creative strategy.
What it does
- Resolve workspace — Identifies the target workspace and pages via Glued MCP.
- Pull comments — Fetches up to 500 comments using
query_page_commentswith platform and source filters (ad vs organic, Facebook vs Instagram).
- Clean and deduplicate — Removes spam, duplicates, and noise from the comment set.
- Cluster themes — Classifies comments into actionable categories: objections (price, trust, delivery, compatibility, policy), desires (outcome statements, intent language), proof (customer endorsements, success indicators), and risk (negative themes requiring escalation).
- Generate outputs — Produces copy angles mapped to recurring themes, FAQ entries, and ready-to-use reply templates.
- Publish (optional) — Syncs clusters to Notion and sends a digest to Slack.
Output
output/
├── comment_insights.md # Themed clusters with representative quotes and copy angles
├── comment_insights.json # Structured data for programmatic consumption
└── reply_templates.md # Ready-to-use response templates by theme
What the analysis actually looks like
Each insight cluster includes:
- Theme classification — Objection, desire, proof, or risk
- Representative quotes — Short, anonymized examples from real comments
- Frequency signal — How often this theme appears
- Copy angles — Suggested ad hooks and messaging inspired by the theme
- FAQ entries — Pre-written answers addressing common questions
- Reply templates — Brand-safe responses for community management
What this replaces
| Before | After |
|---|---|
| Manually reading hundreds of comments | Automated clustering of up to 500 comments |
| Gut-feel copy decisions | Data-backed copy angles from actual customer language |
| No systematic objection tracking | Structured objection taxonomy (price, trust, delivery, etc.) |
| Ad hoc comment replies | Templated, brand-safe reply library |
| Missing the signal in organic comments | Separate analysis of ad vs organic comment themes |
| Customer insights trapped in comments | Structured JSON and markdown for team-wide use |
Who this is for
- Copywriters who want real customer language to fuel ad hooks and landing page copy
- Community managers who need reply templates organized by objection type
- Creative strategists looking for underexploited messaging angles hidden in comments
- Brand managers who need to track recurring objections and sentiment shifts
- Growth teams building FAQ pages and knowledge bases from actual customer questions
Requirements
| Requirement | Details |
|---|---|
| Glued MCP | Connected Meta pages with comment access |
| Claude Code | Runtime environment (also compatible with Codex) |
| Notion (optional) | For syncing insight clusters |
| Slack (optional) | For posting digests |
No API keys or environment variables required.
Skill structure
comment-insights-engine/
├── SKILL.md # Skill definition with classification rubric and procedure
└── README.md # This file
How it works under the hood
Single-agent skill using three Glued MCP tools:
| MCP Tool | Purpose |
|---|---|
list_workspaces | Resolve workspace ID if not provided |
list_pages | Discover connected pages and select scope |
query_page_comments | Pull comments with platform/source filters |
Filters: Platform (facebook, instagram, all), Source (ad, organic, all), Limit (default 200, max 500)
Classification rubric: Objection (price, trust, delivery, compatibility, policy), Desire (outcome statements, intent language), Proof (endorsements, success indicators), Risk (negative themes requiring escalation)
Guardrails: Redacts sensitive personal data. Maps hooks to recurring themes only (not one-off comments). Recommends validating messaging changes with performance data before rollout.
For marketers who know the best copy comes from customers, not brainstorms.