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

1. 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).
2. 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)
3. Aggregate patterns — Count themes, calculate sentiment distribution, identify the top recurring issues.
4. Surface insights — Find the non-obvious patterns: themes that correlate, sentiment shifts, signals that suggest churn risk or expansion opportunity.
5. 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.