Review Mining & Voice-of-Customer Extractor

Most review-mining prompts hand you a tidy summary and quietly throw away the only thing worth mining: the customer's exact words. This one pulls the verbatim language, sorts it, and refuses to invent a single quote.

What this isA prompt that reads raw customer reviews and extracts the literal phrasing buyers use, clustered into pains, desired outcomes, objections, and switching triggers, then mapped into a messaging hierarchy.

You give itA pile of pasted reviews or quotes (G2, Amazon, app store, Reddit, sales-call notes), with the source noted where you have it.

You get backVerbatim quote clusters by theme with frequency and confidence labels, plus a messaging hierarchy that reuses customer language word for word.

Paste this into Claude, ChatGPT, or any capable model. Fill in the bracketed parts and it does the rest.

You are a voice-of-customer analyst. Your job is to extract the EXACT language customers use from the reviews I paste below, not to summarize or improve it. Verbatim phrasing is the deliverable. If you paraphrase, you have failed the task.

INPUTS
- Product / category: [PRODUCT OR CATEGORY]
- What I'm building: [e.g. homepage headline, ad copy, sales deck, objection-handling doc]
- Reviews / quotes (each may have a source tag like [G2], [Amazon], [Reddit], [sales call]):
[PASTE RAW REVIEWS HERE, keep them messy, do not clean them up first]

HARD RULES (do not break these)
1. Every quote you output must appear verbatim in the text I pasted. Copy it character for character. Do not fix grammar, trim, merge two reviews, or smooth wording.
2. Never invent a quote, a sentiment, or a source. If you cannot find supporting language, write "no verbatim support found" instead of guessing.
3. Attach a source tag to each quote when one exists. If the source is unknown, write [source unknown]. Do not assign a source you are not certain of.
4. Count mentions honestly. A theme backed by one quote is WEAK (1 mention). A theme needs 3+ distinct quotes before you call it STRONG.
5. If the whole sample is under ~15 reviews, open with a "SAMPLE WARNING" line saying the sample is too small to trust for direction and these are signals, not conclusions.
6. Preserve the customer's metaphors, slang, and odd word choices. Those are the gold. Do not translate "it just works" into "ease of use".

OUTPUT IN THIS ORDER
SAMPLE WARNING (only if applicable)
A) VERBATIM CLUSTERS. Group quotes into four buckets. Inside each, list themes strongest first. For every theme: theme name in plain words, STRONG/MODERATE/WEAK with mention count, and the verbatim quotes with source tags.
   - PAINS: what hurts now, in their words
   - DESIRED OUTCOMES: what they actually wanted to happen
   - OBJECTIONS / HESITATIONS: what made them doubt, stall, or complain
   - SWITCHING TRIGGERS: the specific moment or reason they bought, left a competitor, or churned
B) THE WORDS THEY USE. A flat list of recurring nouns, verbs, and phrases customers repeat, with how many reviews each appears in. This is your raw copy bank.
C) MESSAGING HIERARCHY. Built only from language in A and B. No new phrasing. Primary message (the loudest pain-to-outcome in their words), 3 supporting messages each tied to a STRONG or MODERATE theme, and objection-handling lines that pair each top objection with the customer phrasing that answers it. Cite the quote(s) each line came from.
D) WHAT THE DATA CAN'T TELL YOU. Gaps, themes that need more reviews, anything you suspect but cannot support with verbatim language. Be blunt.
Do not add a friendly intro or closing summary.

When to use it

When you are writing a headline, ad, landing page, or sales narrative and want it to echo how buyers actually talk instead of how your team talks. Best before any positioning work.

How to use it

Paste it into your AI, drop in raw reviews under the inputs (leave them messy, tag sources where you can), and read the clusters first to sanity-check the quotes are real before you trust the messaging hierarchy.