Feedback Analyzer
Take raw, messy user feedback and turn it into structured, actionable insights. Works with pasted text, CSV files, or any text-based feedback dump.
Process
- Ingest the feedback — If a file path is provided, read it. If CSV, parse it using Bash (csvtool, awk, or python). If pasted text, split by logical entries (one per line, or by paragraph).
- Classify each piece — For every feedback entry, assign:
- Theme (e.g., "Onboarding", "Pricing", "Performance", "Missing Feature")
- Sentiment (Positive / Neutral / Negative)
- Urgency (High / Medium / Low — based on language intensity and frequency)
- Type (Bug report / Feature request / Complaint / Praise / Question)
- Aggregate patterns — Count themes, calculate sentiment distribution, identify the top recurring issues.
- Surface insights — Find the non-obvious patterns: themes that correlate, sentiment shifts, signals that suggest churn risk or expansion opportunity.
- Write recommendations — Translate patterns into concrete next steps for a PM.
Output Format
Summary Stats
- Total feedback entries: X
- Sentiment breakdown: X% positive, X% neutral, X% negative
- Top theme: [theme] (X mentions)
- Date range (if available): [range]
Themes Table
| Theme |
Count |
Sentiment (avg) |
Urgency |
Example Quote |
| ... |
... |
... |
... |
"..." |
Top 3 Patterns
Numbered list. Each pattern includes: what it is, why it matters, and how confident we are (based on volume).
Recommended Actions
| Priority |
Action |
Based On |
Expected Impact |
| P0 |
... |
... |
... |
| P1 |
... |
... |
... |
| P2 |
... |
... |
... |
Raw Classified Data
If fewer than 50 entries, include a full table with each entry classified. If more than 50, save to a file and note the path.
Rules
- Do not editorialize feedback. Classify what users said, not what you think they meant.
- If a feedback entry touches multiple themes, assign the dominant one but note the secondary theme.
- Urgency is based on language intensity ("broken", "can't use", "blocking" = High) and frequency, not on your opinion of importance.
- Minimum 3 themes required. If all feedback collapses into one theme, split it into sub-themes.
- Always include at least one verbatim quote per theme. Real words beat summaries.
- Write in English.
- Keep analysis output under 2 pages. The raw classified data table can be longer if needed.
- Be opinionated in recommendations: tell the PM what to do first and why, don't just list options.
- If the feedback volume is too small to draw conclusions (fewer than 5 entries), say so explicitly and flag which patterns are weak signals.
1---2name: feedback-analyzer3description: Classify and analyze raw user feedback into themes, sentiment, and actionable insights. Takes pasted feedback text or a path to a CSV file and returns structured patterns and recommendations.4---56# Feedback Analyzer78Take raw, messy user feedback and turn it into structured, actionable insights. Works with pasted text, CSV files, or any text-based feedback dump.910## Process11121. **Ingest the feedback** — If a file path is provided, read it. If CSV, parse it using Bash (csvtool, awk, or python). If pasted text, split by logical entries (one per line, or by paragraph).132. **Classify each piece** — For every feedback entry, assign:14 - **Theme** (e.g., "Onboarding", "Pricing", "Performance", "Missing Feature")15 - **Sentiment** (Positive / Neutral / Negative)16 - **Urgency** (High / Medium / Low — based on language intensity and frequency)17 - **Type** (Bug report / Feature request / Complaint / Praise / Question)183. **Aggregate patterns** — Count themes, calculate sentiment distribution, identify the top recurring issues.194. **Surface insights** — Find the non-obvious patterns: themes that correlate, sentiment shifts, signals that suggest churn risk or expansion opportunity.205. **Write recommendations** — Translate patterns into concrete next steps for a PM.2122## Output Format2324### Summary Stats25- Total feedback entries: X26- Sentiment breakdown: X% positive, X% neutral, X% negative27- Top theme: [theme] (X mentions)28- Date range (if available): [range]2930### Themes Table3132| Theme | Count | Sentiment (avg) | Urgency | Example Quote |33|-------|-------|-----------------|---------|---------------|34| ... | ... | ... | ... | "..." |3536### Top 3 Patterns37Numbered list. Each pattern includes: what it is, why it matters, and how confident we are (based on volume).3839### Recommended Actions4041| Priority | Action | Based On | Expected Impact |42|----------|--------|----------|-----------------|43| P0 | ... | ... | ... |44| P1 | ... | ... | ... |45| P2 | ... | ... | ... |4647### Raw Classified Data48If fewer than 50 entries, include a full table with each entry classified. If more than 50, save to a file and note the path.4950## Rules5152- Do not editorialize feedback. Classify what users said, not what you think they meant.53- If a feedback entry touches multiple themes, assign the dominant one but note the secondary theme.54- Urgency is based on language intensity ("broken", "can't use", "blocking" = High) and frequency, not on your opinion of importance.55- Minimum 3 themes required. If all feedback collapses into one theme, split it into sub-themes.56- Always include at least one verbatim quote per theme. Real words beat summaries.57- Write in English.58- Keep analysis output under 2 pages. The raw classified data table can be longer if needed.59- Be opinionated in recommendations: tell the PM what to do first and why, don't just list options.60- If the feedback volume is too small to draw conclusions (fewer than 5 entries), say so explicitly and flag which patterns are weak signals.
Mehdibargach