Turning returns and review data into a product-improvement summary
Return reasons and one-to-three-star reviews are the honest roadmap for your next production run and your listing fixes, but reading hundreds of them by hand is slow. Returns cost U.S. retailers roughly $850 billion in 2025, and the data behind them is where the fixes hide. AI is strong at clustering the themes — provided it works only from your pasted, de-identified data and quotes real language back rather than inventing causes.
You are a product analyst for {{product_name}}. Below is return-reason and review data. Analyze ONLY this data — do not add causes or numbers from your own knowledge. Return-reason export (de-identified): {{returns_export}} Negative reviews (de-identified): {{reviews}} Produce: 1. The top themes across returns and complaints, each with a rough count and two verbatim quotes from the data. 2. A split of each theme into: listing/expectation fixes (the product is fine, the listing oversold or under-described it) vs. product/quality fixes vs. sizing/fit issues. 3. A prioritized shortlist of what to change before the next run or listing update. Rules: - Use only the data I pasted. Do not invent counts, percentages, or root causes. Quote customer language verbatim; do not paraphrase a complaint into something they didn't say. - Mark any theme that shows up only once or twice as [LOW EVIDENCE] — a hypothesis to watch, not a trend to act on. - If any item describes a safety hazard, injury, or defect risk, flag it [SAFETY — HUMAN REVIEW] and do not fold it into a routine listing tweak.
Fill in your details and the prompt updates live — then copy.
Theme 1 — Lid leaks (~38 returns). "Leaked all over my bag the first day." / "Lid doesn't seal." → Product/quality fix: review the lid gasket with the supplier. [SAFETY — HUMAN REVIEW] only if hot liquids are involved. Theme 2 — Smaller than expected (~27 returns). "Thought it was bigger." → Listing/expectation fix: add a size-comparison photo and the exact dimensions to the bullets. Theme 3 — "Metallic taste" — appears twice. [LOW EVIDENCE] watch, don't act yet.
The full workflow
- Export return reasons and reviews, then strip names, emails, addresses, and order numbers before pasting
- Run the prompt, then spot-check that each quote actually appears in your source data
- Separate listing fixes you can make today from product fixes that need the supplier
- Route every [SAFETY — HUMAN REVIEW] flag to a person, and treat [LOW EVIDENCE] themes as hypotheses
- Turn the top listing fixes into concrete edits and the top product fixes into supplier questions
Watch out for
Return comments and reviews contain names, order numbers, and addresses. De-identify before pasting into a consumer AI tool — buyer personal data pushed into a free tool may be retained, and mishandling it can breach marketplace data policies and privacy law.
AI will smooth thin data into confident-sounding causes. Treat any theme from one or two mentions as a hypothesis, and if returns reveal a genuine safety defect, that's a recall and compliance matter for a human — potentially a CPSC report — not a listing tweak.
Where this comes from
Every use case on this site is grounded in real reports from working e-commerce sellers — not invented by us.