Verified against ChatGPT · 2026-08-13
Extract the real recurring themes from a batch of product reviews instead of a generic sentiment summary
Analyzes a pasted batch of product reviews to surface specific recurring praise and complaint themes with supporting quotes and rough frequency, instead of a vague 'customers love the quality' summary.
The prompt
Ready to copy — highlighted parts are example details you can swap.
Analyze the batch of product reviews below and extract the actual recurring themes — not a generic sentiment summary like "most customers are satisfied with the quality." PRODUCT Trailblaze hiking backpack, 40L REVIEWS (paste as many as you have) [Pasted block of 30-80 real customer reviews with star ratings] WHAT I ALREADY SUSPECT (to check, not to confirm blindly) I suspect the hip belt padding is the most common complaint, based on a handful of recent one-star reviews I noticed STEP 1 — THEME EXTRACTION Group the reviews into specific recurring themes (not "positive" and "negative" as categories — actual topics: a specific feature working or failing, a specific use case it excels or struggles at, a specific comparison customers keep making to something else). For each theme, note roughly how many reviews mention it and pull 1-2 short direct quotes as evidence — do not summarize a theme you can't actually point to quotes for. STEP 2 — CHECK THE HYPOTHESIS Evaluate what I said I suspected against what the reviews actually show — confirm it, contradict it, or say the evidence is mixed/insufficient. Do not tell me what I want to hear; if the reviews don't support my hypothesis, say so plainly. STEP 3 — SURPRISES Name anything in the reviews that doesn't fit an obvious theme but seems important — an unexpected use case, a recurring minor gripe that isn't a dealbreaker but shows up often enough to matter, a demographic or use pattern I might not have expected. WHAT NOT TO DO Do not round every complaint into a single generic "some customers had issues" bucket — a complaint about slow shipping and a complaint about a product defect are different problems requiring different fixes, and collapsing them loses the actionable signal. OUTPUT FORMAT Themes table (theme, approximate frequency, sample quotes), hypothesis check paragraph, surprises list.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Asked to "summarize customer sentiment" without further constraint, GPT-5.1 tends to produce a high-level, evenly-hedged paragraph ("customers generally appreciate the quality, though a few had concerns") because that's the safest, most broadly-true statement it can make across a mixed batch of reviews — which is exactly the output that gives a merchant nothing actionable, since it could describe almost any product's reviews. Requiring specific themes with a frequency estimate and supporting quotes forces the model to actually cluster the review text by topic rather than by sentiment polarity, which is the structural difference between a summary and an analysis — sentiment polarity (positive/negative) is nearly always available from any review batch, but topic clustering (which specific feature or use case is driving that sentiment) requires the model to do real categorization work it will skip unless explicitly told the output has to be organized that way. The instruction to check a stated hypothesis rather than confirm it addresses a real and common failure mode: a model given a leading suggestion in the prompt ("I suspect X is the problem") will often bias its summary toward confirming X even when the evidence is mixed, simply because the hypothesis primed what the model looks for first — explicitly telling it to potentially contradict the hypothesis counteracts that anchoring. The ban on collapsing complaints into one generic bucket matters operationally because a shipping complaint and a manufacturing defect complaint point a business toward completely different fixes, and losing that distinction in an AI-generated summary directly costs the merchant time diagnosing the wrong problem.
What you get back
Theme: Hip belt padding compresses after ~2 months of regular use — mentioned in roughly 9 of 62 reviews, concentrated in 3-star and below. Quote: 'Loved it for the first month, then the padding went flat on long days.' Hypothesis check: Partially confirmed — hip belt padding is a real recurring theme, but it's the third most common complaint, not the first; strap adjustment difficulty is mentioned more often...
Verified against
ChatGPT GPT-5.1 · 2026-08-13
Changelog
- 2026-08-13 — 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
