Route customer feedback into consistent, useful themes
Classify incoming feedback into transparent categories your team can review and act on.
Best for
- Repeated feedback from forms or support queues
- Teams with an agreed taxonomy and human review
What you need
- A sample of real feedback with personal data removed
- A short list of approved themes
- A destination spreadsheet or tracker
Limitations
Automated labels can miss sarcasm, context and emerging themes; do not use them for high-stakes customer decisions.
Recommended tool
Zapier + OpenAI
Start with the free manual route and confirm your categories hold. Zapier plus the OpenAI API only earns its cost once the rubric is stable, and both services bill separately.
AI4E may earn a commission from partner links. This never affects which tool we recommend. Read the affiliate disclosure →
Pick your stack
Do it for nothing, or move faster with a paid tool.
Google Sheets + ChatGPT
$0 · Completely free
Paste batches of feedback and classify manually — slower, but completely free.
AI4E may earn a commission from partner links. This never affects which tool we recommend. Read the affiliate disclosure →
Your copy-paste prompt
Fill in the blanks on the left and the prompt updates as you type.
Fill in your details
0 of 2 fields filled — every matching placeholder updates as you type.
Act as a customer-feedback analyst. Classify only the feedback in [FEEDBACK]. Use these approved themes: [THEME LIST]. Return valid JSON with: sentiment (positive, neutral, negative, mixed), theme, product_area, urgency (1-5), evidence_quote, one_sentence_summary, confidence (low, medium, high), and needs_human_review. Do not infer identity, intent or facts not stated. Use "other" when no theme fits and explain why. Ask for the theme definitions if they are missing.
Step-by-step instructions
- 1
Remove personal data and test the rubric on 20 varied examples manually.
- 2
Compare labels with a human reviewer and refine ambiguous theme definitions.
- 3
Only then connect the feedback source and destination in your automation tool.
- 4
Audit a sample every week and pause routing if accuracy drops.
Check the result
- Every label is supported by a quoted phrase
- Unknown cases use other rather than a forced category
- Low-confidence items reach a person
- No personal data is copied unnecessarily
If the result is poor
- If labels drift, add one positive and one negative example for each theme.
- If JSON breaks, request only the schema with no commentary or markdown.
Privacy and accuracy
- Remove names, email addresses, account numbers and sensitive customer details.
Sample input
Ten anonymised support tickets; themes: billing, onboarding, reliability, feature request.
Expected output
One JSON record per ticket with an evidence quote, confidence level and human-review flag.
What people do next
Three recipes chosen for the task that usually follows this one, not just for sharing a category.
Business Automations
Analyse customer reviews and find recurring complaints
30 mins · Claude
Start this workflow →
Business Automations
Turn spreadsheet data into a plain-English weekly report
25 mins · ChatGPT
Start this workflow →
Business Automations
Qualify incoming leads with a transparent scoring rubric
35 mins · ChatGPT
Start this workflow →