How to Mine Customer Reviews for CRO Gold

You are analyzing customer reviews for [PRODUCT/BRAND NAME], a [brief product description, e.g. "premium magnesium supplement for sleep and recovery"].

I will paste raw customer reviews below. Your job is to read every review carefully and create a structured taxonomy of purchase drivers - the specific reasons customers decided to buy this product.

Rules:
- Each category should be a clear, specific driver (e.g. "Sleep quality improvement" not just "Health")
- Aim for 8-15 categories. Fewer means you're being too broad. More means you're splitting hairs.
- Use customer language for category names when possible
- Include a one-sentence description for each category explaining what it covers
- Group related sub-drivers under parent categories only if the parent has 3+ sub-drivers
- If a review mentions multiple drivers, note that - do not force reviews into one category

Output format:
A numbered list of purchase-driver categories, each with:
- Category name
- One-sentence description
- 2-3 example phrases from the reviews that fall under this category

Here are the reviews:

[PASTE REVIEWS HERE]

You are tagging customer reviews for [PRODUCT/BRAND NAME] against a purchase-driver taxonomy.

Here is the taxonomy (from the previous step):
[PASTE YOUR TAXONOMY HERE]

For each review below, assign one or more purchase-driver tags from the taxonomy above. Follow these rules:

- A review can have multiple tags. Most will have 2-4.
- Only tag a driver if the review explicitly mentions or strongly implies it. Do not infer drivers that are not clearly present.
- If a review does not match any category, tag it as "Uncategorized" and flag it for taxonomy review.
- Preserve the exact review text. Do not paraphrase or summarize.
- Include the reviewer's name or identifier if available.

Output format (for each review):
- Review #[number]: "[First 10 words of the review...]"
- Tags: [Driver 1], [Driver 2], [Driver 3]
- Confidence: High / Medium / Low (how clearly the drivers are expressed)

Here are the reviews to tag:

[PASTE REVIEWS HERE]

You are analyzing tagged customer reviews for [PRODUCT/BRAND NAME].

Here are all the tagged reviews from the previous step:
[PASTE TAGGED REVIEWS HERE]

Create a summary table that quantifies every purchase driver. For each driver, calculate:

1. Total mentions (how many reviews tagged with this driver)
2. Percentage of all reviews (mentions / total review count)
3. Average sentiment (Positive / Mixed / Negative - based on context of the mentions)
4. Trend signal (if noticeable patterns exist, e.g. "frequently paired with [other driver]")

Output format:
A markdown table sorted by mention count (highest first) with columns:
| Rank | Purchase Driver | Mentions | % of Reviews | Avg Sentiment | Notable Patterns |

After the table, add a brief "Key Takeaways" section (3-5 bullet points) highlighting:
- The dominant purchase drivers
- Any surprising findings
- Drivers that are frequently paired together
- Gaps (expected drivers that barely appear)

Total reviews analyzed: [NUMBER]

You are selecting the best customer quotes for each purchase driver for [PRODUCT/BRAND NAME].

Here is the purchase-driver summary table:
[PASTE SUMMARY TABLE HERE]

Here are the tagged reviews:
[PASTE TAGGED REVIEWS HERE]

For each purchase driver, select 2-3 of the most compelling, authentic customer quotes. A great quote is:

- Specific (mentions concrete details, numbers, timeframes, or before/after comparisons)
- Emotional (conveys genuine feeling - relief, surprise, excitement)
- Credible (sounds like a real person, not a marketing script)
- Self-contained (makes sense without needing the full review for context)
- Concise (under 2 sentences ideally - trim if needed, but never change the customer's words)

Output format (for each driver):
## [Purchase Driver Name]
1. "[Quote]" - [Reviewer name/identifier]
   Why selected: [1 sentence explaining what makes this quote compelling]
2. "[Quote]" - [Reviewer name/identifier]
   Why selected: [1 sentence]

Also flag any drivers where no strong quote exists - these are content gaps you will need to address with other evidence.

You are a conversion copywriter extracting voice-of-customer (VoC) language from reviews for [PRODUCT/BRAND NAME].

Here are the tagged reviews with their purchase driver tags:
[PASTE TAGGED REVIEWS HERE]

Extract customer phrases, sentences, and language patterns for direct use in marketing copy. For each category below, pull 5-10 examples:

1. HEADLINE HOOKS - Short, punchy phrases (under 10 words) that could work as ad headlines or hero text.
2. OBJECTION BUSTERS - Quotes where customers describe overcoming skepticism or switching from competitors.
3. OUTCOME DESCRIPTIONS - Phrases where customers describe specific results or transformations in their own words.
4. EMOTIONAL TRIGGERS - Language that reveals the deeper motivation behind the purchase.
5. PRODUCT DIFFERENTIATORS - Phrases where customers compare to alternatives or call out what makes this product different.

Output format per category:
- The exact phrase (in quotes)
- Which purchase driver it maps to
- Suggested use: "Headline" / "Subheadline" / "Ad copy" / "PDP bullet" / "Email subject" / "Social post"

Prioritize phrases that sound natural. Never polish or rewrite the VoC.

You are mining negative reviews (1-3 stars) for [PRODUCT/BRAND NAME] to find hidden purchase drivers, objections, and friction points.

Here are the negative and mixed reviews:
[PASTE 1-3 STAR REVIEWS HERE]

For each negative review, extract:

1. THE UNMET EXPECTATION - What did the customer expect based on the marketing?
2. THE PURCHASE DRIVER THAT FAILED - Which driver brought them in, and why did it fall short?
3. THE HIDDEN OBJECTION - What concern did other prospective customers probably share but did not voice?
4. THE FIX SIGNAL - What would the customer have needed to be satisfied?

Output format per review:
- Review snippet: "[First 15 words...]"
- Star rating: [1-3]
- Unmet expectation: [1 sentence]
- Failed driver: [Driver name from taxonomy]
- Hidden objection: [1 sentence]
- Fix signal: [1 sentence]

After all reviews, provide a "Priority Fixes" summary:
- Top 3 messaging gaps
- Top 3 objections to address proactively on the product page
- Top 3 product/experience improvements

You are performing a quality audit on AI-generated review analysis for [PRODUCT/BRAND NAME].

Here is the original set of reviews:
[PASTE ORIGINAL REVIEWS HERE]

Here is the AI-generated analysis to verify:
[PASTE THE TAXONOMY, TAGS, QUOTES, AND VoC EXTRACTS HERE]

Check every claim against the source reviews. Verify:

1. QUOTE ACCURACY - Is every quoted phrase present in the original reviews? Flag any paraphrased or fabricated quotes.
2. TAG ACCURACY - For a random 20% sample of tagged reviews, verify tags are justified.
3. NUMBER ACCURACY - Verify mention counts and percentages by spot-checking.
4. ATTRIBUTION ACCURACY - Make sure quotes are attributed to the correct reviewer.
5. SENTIMENT ACCURACY - Check that sentiment labels match actual review tone.

Output format:
## Accuracy Report
### Quotes Verified: [X/Y passed]
- [List inaccurate quotes with corrections]
### Tags Verified: [X/Y passed]
- [List incorrect tags with explanation]
### Numbers Verified: [Pass/Fail]
- [List counting errors]
### Overall Confidence: [High / Medium / Low]
- [Summary and whether analysis is reliable enough to use]

You are formatting a complete review mining analysis for [PRODUCT/BRAND NAME] into a Notion-ready document.

Here is the full analysis:
- Taxonomy: [PASTE]
- Tagged reviews: [PASTE]
- Summary table: [PASTE]
- Representative quotes: [PASTE]
- VoC extracts: [PASTE]
- Negative review insights: [PASTE]
- Quality check results: [PASTE]

Format into a clean Notion document:
- H1 for report title: "[BRAND NAME] Review Mining Report"
- H2 for major sections
- Toggle blocks for detailed data (collapsed by default)
- Markdown tables for summary data
- Callout blocks for key takeaways
- Dividers between sections

Structure:
1. Executive Summary (3-5 key findings)
2. Purchase Driver Taxonomy (with toggle for details)
3. Quantified Summary Table
4. Top Quotes by Driver
5. Voice-of-Customer Language Bank (by copy use case)
6. Objection & Friction Map
7. Recommended Actions (prioritized)
8. Appendix: Full Tagged Review Data (collapsed toggle)

Keep formatting clean and scannable. Every section needs a clear "so what" takeaway.

You are exporting the review mining analysis for [PRODUCT/BRAND NAME] into CSV format.

Here are the tagged reviews and analysis:
[PASTE TAGGED REVIEWS, SUMMARY TABLE, AND QUOTES HERE]

Create a CSV with columns:
review_id | reviewer_name | star_rating | review_text | driver_1 | driver_2 | driver_3 | driver_4 | sentiment | is_quote_worthy | selected_quote | voc_category | notes

Rules:
- review_id: Sequential number
- Leave driver columns blank if fewer than 4 apply
- is_quote_worthy: "Yes" / "No"
- voc_category: "Headline Hook" / "Objection Buster" / etc., or blank
- Escape commas within review text

Also provide a second summary CSV:
driver_name | mention_count | percentage | avg_sentiment | top_quote

Ready for Google Sheets, Notion databases, or any analytics tool.