Use when you have a few hundred reviews sitting unread and want to know what customers keep raising.
review-mining-ours-and-theirs.md
You are mining review text for themes. You are not writing replies and you are not summarising the business.
Our reviews, pasted with star rating and date: {{OUR_REVIEWS}}
Reviews of nearby competitors, pasted the same way: {{COMPETITOR_REVIEWS}}
Period the reviews cover: {{DATE_RANGE}}
Services we sell: {{SERVICE_LIST}}
Pages these findings could change: {{TARGET_PAGES}}
Produce four sections.
1. A theme table for {{OUR_REVIEWS}}: Theme, Praise or complaint, Reviews mentioning it, Rating spread, Two quotes copied exactly, Whether it appears more in the later half of {{DATE_RANGE}}.
2. The same table for {{COMPETITOR_REVIEWS}}, then a list of complaints that recur about them and not about us.
3. Customer language: the words reviewers use for the things in {{SERVICE_LIST}} where they differ from our own wording, with a count for each.
4. Two ranked lists. Copy changes, each mapped to a page in {{TARGET_PAGES}}, and operational changes that no wording will fix.
Rules:
- A theme needs three or more reviews. Anything below that goes under a heading called single mentions and is not counted as a theme.
- Quote verbatim. Do not tidy spelling or grammar, and do not use a quote that names a member of staff or identifies a customer.
- Counts only. No percentages, no sentiment scores, no inferred causes.
- If part of {{DATE_RANGE}} is thinly covered, say so rather than reading a trend into it.
- No em dashes.
Replace each placeholder with your own detail. The more specific you are, the less the model invents.
When should I mine reviews rather than reply to them?
When a few hundred have accumulated and nobody has read them as a body of text. This is analysis, not reply writing, so the positive and negative response prompts handle the replies. Reach for this when you want to know what keeps coming up and what competitors get complained about repeatedly.
What do I need to paste?
At least a hundred of your reviews with star rating and date, competitor reviews in the same format, the {{DATE_RANGE}} they cover, your {{SERVICE_LIST}} and the {{TARGET_PAGES}} findings could change. Without dates the later half column is noise, and with fewer reviews the three mention threshold leaves almost nothing standing.
What comes back?
Ranked theme tables for both sides with verbatim quotes, the words reviewers use where they differ from your own wording, and two lists splitting copy changes from operational ones. The competitor complaints that do not recur about you are the most directly usable output, since each is a line for your own page.
What is the mistake?
Trusting the counts without checking. Search the pasted text for three or four themes before acting on any of them. Do not read a trend into a thinly covered stretch of {{DATE_RANGE}}, and do not move an operational complaint into the copy list because rewording is cheaper than fixing the thing itself.