Table of Contents
The Five-Star Review That Should Have Triggered a Product Recall
In one brand analysis, a customer wrote a glowing five-star review. She gave the product top marks. She said she loved the brand. And then, almost as an aside, she mentioned the product tasted like expired peanuts.
Every analytics dashboard this brand used classified her as a promoter. NPS calculation: positive. Sentiment score: positive. She shows up in the “happy customer” bucket. The review sits in a sea of five-star ratings and is never read by a human being.
That’s not a data problem. That’s a reading problem. And it’s costing Amazon brands more than they know.
This article explains how to use Claude’s `amazon-review-insights` skill – connected to the Smacient Cloud Connector – to surface exactly these kinds of findings across your entire brand catalogue, automatically.
What Review Analysis Actually Looks Like Today
no em dash. no quotes
10:56 PM
Claude responded: Ask most Amazon brand teams how they use their reviews and you’ll hear the same answer.
Ask most Amazon brand teams how they use their reviews and you’ll hear the same answer. They monitor their star rating, check for sudden drops, and filter for one-star reviews when something goes wrong. Some teams go a step further with keyword counts or sentiment scoring, but that is typically where it stops.
This is the state of the art for most brands, and it is deeply insufficient.
Average star ratings are a lagging indicator of a problem you already have. Keyword counts tell you what words appear – not what they mean in context, not how they cluster across products, not what they reveal about the buyer who wrote them. Sentiment scoring collapses nuanced human language into a binary that loses most of the signal.
The result is that brand teams are flying on incomplete information. They know their rating. They don’t know what their customers are actually saying.
And what customers are actually saying is extraordinary – if you read it.
The Categories of Insight That Live in Your Reviews Right Now
There are at least eight distinct categories of business intelligence sitting in a typical Amazon brand’s review corpus that standard analysis will never surface.
Buried complaints inside positive reviews. When you read five-star reviews as text rather than tallying them as scores, you find a consistent pattern: customers who give five stars and then say “but,” “however,” “unfortunately,” “though.” These are your most honest customers. They like you enough to give you stars, and they’re telling you something is wrong. Every dashboard counts them as wins.
Cross-product patterns are invisible at the product level. If one reviewer leaves a negative review on Product A, it’s one complaint. Easy to ignore. But if the same reviewer leaves negative reviews on Products A, C, and F in the same month, that’s a signal about your formulation, your supplier, or your packaging – not about individual SKUs. This pattern is entirely invisible when you look at products one at a time.
Undiscovered purchase drivers. Customers often buy your product for reasons you never anticipated, and they tell you about it in their reviews. They describe use cases your listing copy doesn’t mention. These are not marginal edge cases – they can represent significant purchase segments that you’re not converting because you’re not speaking to them.
Buyer segments you didn’t know existed. Demographics, health conditions, life situations – reviewers reveal themselves in their writing. Segments emerge from the text that your customer acquisition strategy has never addressed, because you didn’t know they were there.
Regulatory and compliance exposure. Label errors, prohibited claims, ingredient disclosures – customers notice these and write about them. The review is often buried. The risk is not.
Competitive intelligence, written by your own buyers. Customers who have defected to competitors often explain exactly why. They name the competitor. They describe the specific conditions under which they would switch or return. This is a completed competitive brief, written for free, sitting in a three-star review that no one has read.
None of this shows up in your dashboard.
How the Skill Works
The skill is available at https://github.com/smacient/marketing-skills/tree/main/skills/amazon-review-insights. When you trigger the Amazon-review-insights skill, here is what happens step by step:
Step 1 – Review retrieval
The skill uses the Smacient MCP to fetch up to 50 reviews per ASIN across your entire brand catalogue. It batches ASINs in groups of five and runs the fetches in parallel, so a 20-ASIN catalogue is retrieved in a fraction of the time a manual process would take. Every review is retrieved in full – the text, the star rating, the reviewer name, and the date.
Step 2 – Context gathering
Before analysis begins, the skill asks five quick questions: your key business objective, who the insights are for, any issues you already know about, customer segments you care about, and which metric matters most. These answers shape how findings are framed in the final report.
Step 3 – Eight parallel analyses
The skill reads every retrieved review in full – no sampling – and runs eight analysis types simultaneously across the complete corpus:
- Silent complaint audit – scans all reviews, including five-star ones, for qualifying language that signals a buried complaint
- Time-trend analysis – identifies whether complaint patterns are new (last 3-6 months), persistent, or resolving – a new complaint cluster with no historical precedent is flagged as urgent
- Cross-ASIN pattern matching – looks for signals that repeat across products, including the same reviewer appearing on multiple ASINs in the same window
- Segment identification – surfaces buyer segments emerging from review language: gifters, repeat buyers, cross-category users, health-condition-specific buyers
- Competitive intelligence extraction – identifies reviews that name competitors or describe defection conditions
- Hidden use case discovery – finds purchase drivers and use cases not reflected in the current listing copy
- Expectation mismatch detection – identifies gaps between what the listing implies and what buyers received (listing problems, not product problems)
- Operational issue flagging – catches fulfilment, packaging, label accuracy, and delivery complaints that affect repurchase and carry compliance risk
Step 4 – Report generation
The output is a structured markdown file saved to your project’s outputs folder. It includes a stakeholder navigation guide (so founders, marketing, product, and ops each know which insights to read first), a silent complaint audit table, a priority action matrix (P0 to P3), detailed insights with full customer quotes and recommended actions, and a verbatim customer language bank – real phrases from real reviews, ready to use in listing copy or ad creative.
What It Actually Finds: Four Examples from Real Brand Analyses
The silent 13%
In one brand analysis, 13% of five-star reviews contained buried complaints. One of those reviews contained a food safety concern – a product tasting as it had gone off – that the brand had zero visibility into because their analysis pipeline never looked past the star rating. The silent complaint audit reads every five-star review as text and surfaces these for human attention.
The cross-product signal looked like noise
A brand had multiple products in a closely related category. One product had a few recent negative reviews. A second product had a couple of complaints. A third had one or two. Looked at per-product, none of the complaint volumes was alarming. Cross-ASIN analysis revealed the same reviewers appearing across all three products in the same time window, describing the same experience. This was not three isolated product complaints – it was a single batch issue showing up as scattered noise because no tool was reading across the catalogue at once.
The keyword is worth more than any ad campaign
One brand’s single biggest undiscovered purchase driver was “pre-workout snack.” The phrase appeared in 30-plus reviews across seven products. Buyers were discovering on their own that this product worked well in a specific high-intent use context – and writing about it. Not one of the brand’s product listings mentioned pre-workout. Not one ad campaign targeted that keyword. The brand was missing conversions from buyers searching for exactly what their product delivered.
The regulatory issue is buried in a 1-star review
One brand had a nutritional label showing “260 grams of sodium.” Sodium is always measured in milligrams. Never grams. This is a labelling error with potential compliance implications. The error was visible to exactly one customer, who mentioned it in a one-star review buried in a catalogue with hundreds of positive reviews. No dashboard surfaced it. Operational issue flagging reads for this category of risk specifically.
How to Get Started
Here is the complete setup, from zero to running your first analysis.
What you need
1. Claude Code – Anthropic’s AI coding tool. Free to install from claude.ai/code. Available as a CLI, a desktop app, and extensions for VS Code and JetBrains.
2. The Smacient Cloud Connector – the data layer that gives Claude access to Amazon review data. Connect it once, and it works across all sessions. Available at smacient.com/products/marketing-context-claude/. The Smacient MCP connects to Amazon’s review data on your behalf – no API keys or scraping setup required on your end.
3. Your brand’s ASINs – a list of the product ASINs you want to analyse. You can submit any number; the skill handles batching automatically.
Installation
Install the skill from https://github.com/smacient/marketing-skills/tree/main/skills/amazon-review-insights
Mac or Linux: cp -r skills/amazon-review-insights ~/.claude/skills/amazon-review-insights
Windows: xcopy /E /I skills\amazon-review-insights %USERPROFILE%.claude\skills\amazon-review-insights
Or clone the full repo and copy the folder you need: git clone https://github.com/smacient/marketing-skills cp -r marketing-skills/skills/amazon-review-insights ~/.claude/skills/
Running your first analysis
- Open a Claude Code session in any project folder with the Smacient Cloud Connector active
- Type /amazon-review-insights or “analyze Amazon reviews for [your brand]”
- Claude will ask for your brand name, your ASINs, and which Amazon marketplace (India, US, UK, etc.)
- Answer the five quick context questions (or skip them – the skill proceeds with reasonable defaults)
- The skill fetches all reviews and runs the analysis automatically
- Your report is saved to outputs/<brand>-amazon-review-insights.md in your project folder
A typical analysis of 20 ASINs with 50 reviews each takes 5-10 minutes. The output is a markdown report you can read directly, export to PDF, share with your team, or use as a brief for your next listing or campaign update.
What the Smacient Cloud Connector covers
The Smacient MCP is the data connector that makes the skill work. Beyond Amazon reviews, the Smacient Cloud Connector gives Claude access to Google Ads data, Meta Ads data, Google Analytics 4, Google Search Console, Instagram post data, TikTok data, App Store reviews, and more. Once connected, any Claude Code skill that uses the Smacient MCP can access all of these sources in the same session. The Amazon-review-insights skill uses it specifically for Amazon review retrieval – but the connector opens up a much broader set of brand intelligence workflows.
Why This Matters More Than Most Analytics Work
Brand teams on Amazon spend significant money on advertising, on listing optimisation, and on A/B testing their images and titles. The review corpus – which customers are writing for free, continuously, with genuine purchase intent and direct product experience – sits largely unread.
The data is there. The problem is bandwidth. No brand manager has time to read hundreds of reviews per SKU across a catalogue of twenty or thirty products. So the reviews become a star rating, a sentiment score and a pile of text that no one gets to.
What changes when you read all of it is not a marginal improvement in analytics. It is a different picture of your brand – one that includes the silent complaints, the emerging segments, the regulatory exposure, the competitive intelligence, and the purchase drivers that your current go-to-market strategy is missing entirely.
The five-star reviewer who said the product tasted like expired peanuts was trying to help. Most brands will never know she existed.
Get the skill: https://github.com/smacient/marketing-skills/tree/main/skills/amazon-review-insights
Connect the Smacient Cloud Connector: smacient.com/products/marketing-context-claude/
FAQs
It’s a free, open-source skill for Claude Code that fetches your Amazon reviews and runs a deep hidden-patterns analysis – finding complaints buried in five-star reviews, cross-product signals, untapped customer segments, competitive intelligence, and more. The output is a structured report with prioritised insights.
Yes. The skill itself is open source and available at github.com/smacient/marketing-skills. You need a Claude Code session and the Smacient Cloud Connector (MCP) to run it.
It’s the data layer that gives Claude access to Amazon review data. Without it, Claude can’t fetch your reviews. You connect it once to your Claude Code session, and it handles all the data retrieval. More on setup below.
Existing tools count keywords, average star ratings, and run sentiment scoring. They do not read. This skill reads every review in full – including the five-star reviews where your real problems are often hiding – and cross-references across your entire catalogue. The findings that surface are invisible to any tool that works at the level of ratings and keywords.
Claude Code (free from Anthropic), the Smacient Cloud Connector, and a list of your brand’s ASINs. Full setup is in the “How to Get Started” section below.
