Verified against ChatGPT · 2026-08-14
Mine reviews for product roadmap signal, separating what's fixable from what's just a mismatch
Analyzes a batch of reviews specifically to identify actionable product or listing changes, sorting findings into fixable design issues, listing/expectation-setting fixes, and non-issues, so the output feeds a roadmap instead of a vague complaints list.
The prompt
Ready to copy — highlighted parts are example details you can swap.
Mine the review batch below for actual product development and listing signal — I want output I can hand to a product team and a merchandising team separately, not a general complaints summary. PRODUCT Ferro cast iron griddle pan REVIEWS [Pasted block of 40+ reviews] KNOWN PRODUCT CONSTRAINTS (things that can't realistically change) Cast iron is inherently heavy (weight complaints can't be engineered away without changing the material entirely, which we won't do) and requires seasoning maintenance by design CURRENT LISTING COPY / IMAGES Listing photos show the pan on a large 6-burner range and don't show scale next to common objects; description mentions 'lightweight enough to handle daily' without a weight figure stated anywhere Sort every recurring complaint or confusion in the reviews into exactly one of these three buckets, and justify the sorting: BUCKET 1 — ACTUAL PRODUCT ISSUE: A real design, quality, or manufacturing problem that a product team could plausibly fix in a future revision. Do not put something here if it conflicts with a stated fixed constraint — flag those separately instead as a known tradeoff customers should be warned about rather than a fixable issue. BUCKET 2 — EXPECTATION MISMATCH (listing fix): The product is working as designed, but customers are arriving with a wrong expectation the current listing copy or images could have prevented — quote what in the current listing likely caused the mismatch, or note if nothing in the listing explains the gap and it may be a marketplace-wide assumption instead. BUCKET 3 — NOT ACTIONABLE: A complaint that's a one-off, unrelated to the product (shipping carrier issue, unrelated account problem), or too vague to act on — do not force something into bucket 1 or 2 just to seem thorough. WHAT NOT TO DO Do not inflate a single mentioned complaint into a "trend" — only call something a pattern if it appears across multiple independent reviews, and state the actual count. OUTPUT FORMAT Three labeled sections (one per bucket), each as a list of issue, supporting quote count, and one-line recommended next step per item.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
The single most common failure in AI-assisted review mining is producing one flat list of complaints with no distinction between a genuine design flaw and a customer expectation the listing itself set up wrongly, because both surface as similarly-worded negative sentiment in the raw text — a complaint about weight and a complaint about seasoning maintenance can read almost identically in tone even though one is an unfixable material property and the other might be solved entirely by changing a product photo. Forcing a three-way sort with an explicit rule that a complaint conflicting with a stated fixed constraint gets routed away from the product-fix bucket is what prevents a roadmap document from filling up with "fix" items the product team can't actually act on, which is exactly the kind of low-credibility deliverable that gets a review-mining exercise ignored by a busy product team the next time it's handed to them. Requiring the expectation-mismatch bucket to trace back to a specific line in the current listing (or explicitly note when it can't) turns a vague "customers were confused" complaint into an assignable task for whoever owns the listing copy, which is the actual mechanism by which this kind of analysis produces a fix rather than just documenting a problem. The explicit ban on inflating a single mention into a trend addresses GPT-5.1's tendency, when summarizing qualitative text, to generalize from a vivid single example because a strongly-worded individual complaint is more salient in the text than a duller pattern repeated across many milder mentions — requiring an actual count per item forces the model to check its own generalization before presenting it as a pattern.
What you get back
BUCKET 1 — Actual product issue: Handle heat retention — 11 reviews mention the handle staying hot long after cooking, several noting they were surprised since 'cast iron handles are supposed to cool faster.' Recommended: evaluate a handle material or design change in next revision. BUCKET 2 — Expectation mismatch: Perceived weight — 14 reviews call it 'way heavier than expected'; listing photos show it on a large range with no scale reference and the description says 'lightweight enough to handle daily' with no actual weight stated. Recommended: add exact weight to listing and a photo showing it held one-handed for scale. BUCKET 3 — Not actionable: 3 reviews mention a damaged box on arrival, tied to one specific shipping carrier route, not the product itself...
Verified against
ChatGPT GPT-5.1 · 2026-08-14
Changelog
- 2026-08-14 — Initial publish, verified against ChatGPT GPT-5.1.
Need this built into your business?
If a prompt isn't enough — Google Ads management, built and maintained for you — that's Scult's day job.
EXPLORE GOOGLE ADS MANAGEMENT
