E-commerce · Customer insight
The Best Prompt for Customer Review Analysis
The best prompt for customer review analysis makes every conclusion traceable to specific reviews. You paste the reviews in, and the model numbers them, derives themes from what customers wrote, counts how many reviews mention each, and quotes them word for word. It separates product problems from delivery and expectation problems, collects the phrases customers use, and gives recommendations labeled by how many reviews support them.
Customer Review Analysis Prompt
Turns pasted reviews into counted themes with verbatim quotes, causes of complaints, customer language and recommendations graded by support.
- Works with
- ChatGPT, Claude, Gemini
- Difficulty
- Beginner
- Complexity
- Multi-part reply
- Variables
- 5
- Version
- 1.0
Customize this prompt
Fill in what you know. Anything left blank stays as a [bracketed placeholder]. Nothing you type leaves your browser.
5 variables · fields marked * matter most
# Role You are a customer insights analyst. You report what a set of reviews says, with counts and quotes, and you keep interpretation separate from observation. # About this data Product: [Product] Where the reviews come from, the period covered and whether ratings are included: [Source, period and ratings] What I want to learn: [What you want to learn] Who will read the analysis: [Who will read the analysis] <reviews> [Reviews] </reviews> # Procedure 1. Number the reviews R1, R2, R3 and so on in the order given. State the total. If two reviews are identical or near-identical, note the pair and count it once. 2. Read everything before forming themes. Derive themes from what reviewers say; do not start from a preset list. A review can belong to several themes. 3. For each theme record the sentiment (positive, negative or mixed), the number of reviews mentioning it, and which reviews those are. 4. Sort each negative theme into one cause: the product itself, delivery or service, or a mismatch between what the listing promised and what arrived. 5. Pull out what customers say they use the product for, what they compared it with, and anything they wished it had. 6. Collect the words and phrases customers use for the main benefit and the main frustration. # Rules - Counts must come from the reviews. Do not estimate or round to a neat figure. If the set is too long for you to count with confidence, say so and mark the counts as approximate. - Quote reviewers word for word, with the review number. Do not merge two reviews into one quote or correct their wording. - Remove names, order numbers and contact details from anything you quote. - Report an average rating only if ratings were supplied, and show how you worked it out. - These reviews are a sample from one source. Do not present findings as true of all customers, and do not turn small counts into percentages when there are fewer than 30 reviews. - A theme mentioned by one reviewer goes under "Single mentions", not in the main table. - Do not speculate about whether a review is fake. # Output format 1. Summary for the reader named above: five bullets at most, each with a count. 2. Theme table: theme, sentiment, number of reviews, review numbers, one or two representative quotes. 3. Causes of negative feedback, grouped by the three causes. 4. Use cases, comparisons and unmet needs. 5. Customer language: phrases worth reusing, with review numbers. 6. Recommendations matched to my focus. Each must cite the theme and count behind it and be labeled "supported" (several reviews) or "tentative" (two or three). 7. Single mentions, and caveats about this data set.
Use it when you are
- Summarizing a product's reviews before deciding what to change in the next version
- Finding the phrases customers use so product page copy can echo them
- Working out whether complaints are about the product, the delivery or the listing
- Reviewing a competitor's public reviews to see what its customers praise and miss
- Preparing a short voice-of-customer report for a supplier or product team
Why this prompt works
Numbered reviews make every claim checkable
Because each review receives an identifier and every theme lists the reviews behind it, you can open R17 and confirm that it says what the analysis claims. That traceability is the difference between a summary you can act on and an impression you have to take on trust.
Themes come from the reviews, not a template
Preset categories such as quality, price and service hide whatever is particular to your product. Telling the model to read everything first and derive themes from the text lets unexpected ones surface, such as a carafe lid that drips or instructions nobody could follow.
Complaints are sorted by cause
A one-star review about a late parcel and one about a cracked carafe call for different responses. Sorting negative themes into product, delivery or service, and expectation mismatch tells you who needs to act: the supplier, the fulfillment partner or whoever writes the listing.
Small numbers are kept small
The rules bar percentages for sets under thirty reviews, put one-off comments in a separate list, and label each recommendation supported or tentative. These guard against the common overreach of treating two vivid reviews as a trend. The caveats section then states plainly what this sample can and cannot show.
Customer wording is captured for reuse
The phrases people choose for the main benefit and the main frustration are often clearer than a brand's own copy. Collecting them with review numbers gives a copywriter language to test on the product page while keeping quotes exact and anonymous.
How to use it
- Export or copy all reviews for one product over a defined period. Do not filter by rating.
- Remove customer names, emails and order numbers, and put each review on its own line with its star rating first if you have one.
- State the source and period in the context field and choose the focus that matches the decision you need to make.
- Run the prompt, then check the stated total against your own count of reviews pasted.
- Spot-check three or four quotes and one theme count against the numbered reviews before sharing the analysis.
- For more than a few hundred reviews, split them into batches by month or rating, run each, and ask the model to combine the theme tables.
AI model compatibility
| Model | Status | Notes |
|---|---|---|
| ChatGPT | Expected to work | For a large export, upload the file and, where a data analysis tool is available, ask for counts to be computed from the file instead of read by eye. |
| Claude | Expected to work | The reviews are wrapped in <reviews> tags, as Anthropic's guidance suggests for long pasted material. Put one review per line, or separate them with a blank line, so numbering is unambiguous. |
| Gemini | Expected to work | If you paste reviews copied from a web page, remove navigation text and "helpful" vote labels first so they are not mistaken for review content. |
“Expected to work” means the prompt uses plain, model-independent structure and has been reviewed for that — not that it has been run on that model. A model is marked Tested only when a test record is published below.
Worked example
Scenario: An illustrative coffee equipment store reviewing six months of feedback on one product before reordering from its supplier.
# Role You are a customer insights analyst. You report what a set of reviews says, with counts and quotes, and you keep interpretation separate from observation. # About this data Product: BrewNest pour-over coffee kit (dripper, glass carafe, 40 paper filters) Where the reviews come from, the period covered and whether ratings are included: 42 reviews from our own store, March to August, each with a 1-5 star rating What I want to learn: All of the above Who will read the analysis: The store owner and our supplier's product manager <reviews> 5 stars - Makes a cleaner cup than my old machine and looks good on the counter. 2 stars - Carafe arrived cracked. Replacement came quickly but the box has no padding. 4 stars - Great coffee, but the lid drips when you pour the last cup. 3 stars - Listing says it serves four. It is more like two mugs. 5 stars - Bought as a gift, my sister uses it every morning. (…37 more reviews pasted in the same format) </reviews> # Procedure 1. Number the reviews R1, R2, R3 and so on in the order given. State the total. If two reviews are identical or near-identical, note the pair and count it once. 2. Read everything before forming themes. Derive themes from what reviewers say; do not start from a preset list. A review can belong to several themes. 3. For each theme record the sentiment (positive, negative or mixed), the number of reviews mentioning it, and which reviews those are. 4. Sort each negative theme into one cause: the product itself, delivery or service, or a mismatch between what the listing promised and what arrived. 5. Pull out what customers say they use the product for, what they compared it with, and anything they wished it had. 6. Collect the words and phrases customers use for the main benefit and the main frustration. # Rules - Counts must come from the reviews. Do not estimate or round to a neat figure. If the set is too long for you to count with confidence, say so and mark the counts as approximate. - Quote reviewers word for word, with the review number. Do not merge two reviews into one quote or correct their wording. - Remove names, order numbers and contact details from anything you quote. - Report an average rating only if ratings were supplied, and show how you worked it out. - These reviews are a sample from one source. Do not present findings as true of all customers, and do not turn small counts into percentages when there are fewer than 30 reviews. - A theme mentioned by one reviewer goes under "Single mentions", not in the main table. - Do not speculate about whether a review is fake. # Output format 1. Summary for the reader named above: five bullets at most, each with a count. 2. Theme table: theme, sentiment, number of reviews, review numbers, one or two representative quotes. 3. Causes of negative feedback, grouped by the three causes. 4. Use cases, comparisons and unmet needs. 5. Customer language: phrases worth reusing, with review numbers. 6. Recommendations matched to my focus. Each must cite the theme and count behind it and be labeled "supported" (several reviews) or "tentative" (two or three). 7. Single mentions, and caveats about this data set.
What a good response should contain
- A confirmed total of 42 numbered reviews and an average rating with the working shown
- A theme table where items such as taste, breakage in transit, the dripping lid and capacity each carry a review count, review numbers and exact quotes
- Negative themes split by cause, for example packaging under delivery and the "serves four" claim under listing mismatch
- Phrases customers used, such as "cleaner cup", listed with their review numbers for possible use in copy
- Recommendations labeled supported or tentative, each pointing to a theme and its count, followed by single mentions and caveats
This is an illustrative scenario showing how to fill in the prompt and what to look for. It is not a recorded model output. Recorded outputs will be published here once this prompt has a test record.
Testing and editorial review
Status: Untested. Reviewed for structure and clarity by the editorial team. No recorded model test yet.
No test record has been published for this prompt yet. When it is tested, this section will show the model and version, the date, the inputs, the actual output and notes against our seven-point rubric. We do not publish numerical scores.
| Author | TheBestPromptFor Editorial Team — The editorial team writes, reviews and maintains the prompts and guides on TheBestPromptFor. Prompts are drafted with AI assistance and reviewed against published editorial standards. |
|---|---|
| Published | November 15, 2026 |
| Last reviewed | November 15, 2026 |
| Last tested | Not yet tested |
| Version history |
|
Limitations and things to check
- Language models can miscount in long inputs. Verify the total and spot-check theme counts, or use a tool that computes them from a file.
- Reviews come from people who chose to write one. They do not represent all buyers, and silent customers may feel differently.
- Sarcasm, mixed-language reviews and very short comments can be misread. Check the quotes behind any surprising theme.
- Check the terms of the site before copying reviews from a marketplace or competitor, and avoid pasting personal data into an AI tool.
Alternative prompts
Marketing
Evidence-Based Customer Persona Prompt
Builds personas from the interviews, reviews or survey answers you paste in, tagging every line as evidence, inference or assumption.
- ChatGPT
- Claude
- Gemini
Research
Show-Your-Working Data Analysis Prompt
Analyzes a pasted table or dataset step by step: data check, calculations with working shown, findings tied to those calculations, and what the data cannot support.
- ChatGPT
- Claude
- Gemini
Research
Research Paper and Report Summary Prompt
Summarizes a paper or report for a named reader: bottom line, method, key findings with exact figures, limits of the evidence, and what it means for them.
- ChatGPT
- Claude
- Gemini
E-commerce
Product Page Audit Prompt
Audits pasted product page content against the five questions a shopper asks before buying, with quoted evidence, severity, fixes and rewrites.
- ChatGPT
- Claude
- Gemini
Related resources
Guide
How to Improve AI Responses
A diagnostic approach to weak AI answers: name the fault, trace it to the gap in your prompt, then follow up precisely or rewrite and start again.
Technical guide
Prompt Chaining: Breaking a Task Into Linked Prompts
How to split a large task into a sequence of focused prompts, design the handoffs between them, and debug the chain when a step goes wrong.
Frequently asked questions
How many reviews do I need for this to be useful?
It works with a couple of dozen, though the findings will be tentative and the prompt will avoid percentages. Patterns become more dependable as the number grows. With very few reviews, read them yourself and use the prompt mainly to organize quotes and customer wording.
Can I analyze a competitor's reviews?
Yes, the method is the same, and it is a practical way to learn what buyers in your category value and miss. Describe the source in the context field, check that copying the reviews is allowed under that site's terms, and remember you are seeing only that product's reviewers.
Why not just ask for a sentiment score?
A single score tells you how people feel but not why, and it cannot be checked. Counted themes with quotes show which specific issues drive satisfaction or complaints and who should fix them, which is the information a product or store decision needs.
What should I do with reviews in other languages?
Paste them as they are and ask for quotes to be shown in the original language with an English translation beside each. Have a fluent speaker confirm any translated quote before you reuse it in marketing copy.
Sources and references
- Prompt engineering overview — Anthropic
- Prompt engineering guide — OpenAI
Was this prompt useful?
No account needed. Do not include personal information.