Analyze a batch of customer reviews for a product or service and return a structured summary: overall sentiment, top pros and cons, common use cases mentioned, quality concerns, notable quotes, and actionable recommendations for the product team.
Inputs
| Field | Type | Details |
|---|---|---|
| Reviews | List of text values | Paste individual reviews, one per entry. Include as many as you have (the agent handles anywhere from 3 to 500+). |
| Product or Service Name | Text | What the reviews are about (e.g., "Acme CRM," "CloudSync Backup Pro") |
| Analysis Focus | Dropdown (optional) | General / Feature Requests / Quality Issues / Competitor Mentions |
If an Analysis Focus is selected, the agent still produces the full summary but gives extra depth in that area.
Outputs
Sentiment overview
| Metric | Value |
|---|---|
| Overall Score | 1.0 to 5.0 (average across reviews) |
| Sentiment Label | Very Negative / Negative / Mixed / Positive / Very Positive |
| Distribution | Count and percentage at each star level (1-5), or count by positive/neutral/negative if star ratings aren't available |
Top 3 pros
The three most frequently praised aspects across all reviews, with the approximate number of reviews that mention each one. Phrase these the way customers actually say them, not in marketing language.
Top 3 cons
Same format as pros, but for the three most common complaints.
Common use cases mentioned
A list of the specific ways customers describe using the product. Include roughly how many reviews mention each use case. This is useful for understanding whether customers are using the product the way you expect.
Quality and reliability concerns
Any patterns around bugs, downtime, data loss, performance issues, or other reliability problems. If multiple reviews mention the same issue, group them and note the count. If no quality concerns are found, state that explicitly.
Notable quotes
Five to eight verbatim quotes from the reviews that are especially useful. For each quote, include:
-
The quote itself (exact text)
-
Whether the sentiment is positive, negative, or mixed
-
Why this quote is notable (e.g., "captures a common frustration with onboarding," "specific feature request that multiple users echo")
Pick quotes that are specific and concrete rather than generic ("love this product" is not useful; "the API integration took 10 minutes and saved us from building a custom webhook handler" is useful).
Recommendations for the product team
Three to five actionable recommendations based on the review data. Each recommendation should cite the evidence from the reviews that supports it. For example: "Consider adding bulk export to CSV. At least 12 reviews mention needing to export data manually, and 3 reviews cite this as a reason they're considering alternatives."
How analysis works
-
Read every review and extract the core sentiment, topics discussed, and any specific features, issues, or use cases mentioned.
-
Cluster by topic. Group mentions of similar themes together (e.g., "slow loading" and "takes forever to load" and "performance is terrible" all go in the same cluster). Use the customers' language to label clusters, not generic categories.
-
Count frequencies. Rank pros, cons, use cases, and concerns by how often they appear. A theme mentioned in 40% of reviews matters more than one mentioned in 2%.
-
Select quotes. Pick quotes that are specific, representative, and would be useful in a product planning meeting. Avoid quotes that are purely emotional without substance.
-
Generate recommendations. Each recommendation must connect directly to review evidence. Don't recommend things that the reviews don't support.
Edge cases
-
Fewer than 5 reviews: still produce the analysis, but note in the output that the sample size is small and patterns may not be reliable.
-
Reviews in multiple languages: analyze all reviews regardless of language. Note the language distribution in the sentiment overview.
-
Suspected fake reviews: if any reviews look obviously fake (identical text, generic praise with no specifics, review bombing patterns), flag them in a separate section and exclude them from the main analysis.
-
Mixed product reviews: if some reviews appear to be about a different product (common on Amazon-style platforms), flag and exclude them.