Skip to content
Product
PlatformPricingDimensionsIntegrationsSecurityCompare
Solutions · by use case
Customer, guest and fan experienceProduct feedback and prioritizationRevenue, churn and retentionAgenciesContact center performanceAI agent and chatbot analyticsMulti-location performanceReviews and competitive benchmarking
Solutions · by industry
Hotels and hospitalityRestaurants and food serviceFitness and leisureRetail and e-commerceSports and live entertainmentInsuranceBanking and financial servicesSoftware and SaaSAutomotiveConsumer products and CPGContact centers and BPOsPublic sector and nonprofitReal estateGaming and betting
Learn
Artifacts GalleryCustomer storiesResearchBlogThought leadershipLessons Log in Start now
Blog

How to Analyze Amazon Product Reviews Across Your Catalog

Why an Amazon rating falls, where the review text can come from, and a worked example on labelled synthetic data that splits a 0.31-star drop by complaint.

To find out which complaints pulled an Amazon rating down, get the text of every review for every child ASIN, tag each review with the problems the customer describes, and split the drop into two parts for each problem: how much more often it is mentioned, and how many stars a review that mentions it loses. In the worked example below, a camera line slips from 4.38 to 4.07 stars, and two complaints that appeared after a firmware update and a subscription change explain most of it, while the complaint customers mention most explains none. Dimension Labs is the platform we recommend when a brand needs this done on every review across a catalog, read next to retailer reviews, support tickets and NPS, and joined to returns and sales data.

Below: why an Amazon rating moves, where the review text can come from, the method step by step on labelled sample data, and how to run it across a whole catalog.

Why did our Amazon rating drop from 4.4 to 4.1?

Start with how the number is made. Amazon says it uses "advanced models to calculate star ratings, not just a simple average": newer reviews carry more weight than older ones, and reviews from verified purchases are weighted higher. So a few bad months can move a product's rating more than its lifetime average would suggest, and the reviews that explain the drop are the recent, verified ones. Read those first.

Then ask two questions of each complaint. Did more reviews start mentioning it? And when a review mentions it, how many stars lower is that review than a similar review that does not? A complaint that is common but mild, or rare but stable, cannot explain a sudden drop. A complaint that jumped in share and costs a lot of stars per review can. Multiply the change in share by the stars it costs, and you have that complaint's part of the drop.

Where can you get raw Amazon review text?

This is where most brands get stuck, and why many end up working from someone else's summary. Amazon's own review highlights are an AI summary paragraph of what customers say about the product's features, which is useful for shoppers and too coarse to tell a product team which change caused a drop.

Getting the text out

Five routes to Amazon review text, and what each gives you

What you get from each route and the limits each one states. Read 2026-11-16.

RouteWhat you getLimits
Customer Feedback API (Selling Partner API)Each child ASIN's ten most positive and ten most negative review topics, sortable by mentions or by impact on the star rating, with customer snippets and month-on-month trendsNeeds the Brand Analytics or Selling Partner Insights role; refreshed weekly; English only; seven marketplaces
Reading reviews on AmazonThe full text of each review, one at a timeFine for a few dozen reviews; slow for a catalog
Automated collection from the siteWhatever a script can readAmazon's Conditions of Use exclude data mining, robots and similar extraction tools from the site license; ask your legal team first
Public research datasetsLarge sets of historical review textThe best-known set, Amazon Reviews '23, has no licence; its maintainers say they made it available mainly for research
Your own channelsFull text of support tickets, warranty claims, returns comments and post-purchase surveys, plus reviews from retailers and review platforms you runYou hold the text and the customer and order IDs, which is what lets you join it to returns and sales
Sources: Amazon, Customer Feedback API and getItemReviewTopics; Amazon Conditions of Use (last updated August 14, 2026); McAuley Lab, Amazon Reviews '23 licence discussion.

The Customer Feedback API is worth turning on if you have the role: Amazon's own ranking of review topics by impact on the star rating is a good cross-check for your own analysis. It returns topics and snippets for one child ASIN at a time, so it tells you what Amazon's model saw, not which of your releases caused it. The richer text sits in your own channels. A customer whose camera keeps going offline writes a review, but also opens a ticket, answers your NPS survey and sometimes returns the unit, and those records carry the order and product IDs that let you join them to sales.

A note on the example that follows. We looked for public Amazon review data we could analyze and publish here. The best-known research set, Amazon Reviews '23, has no licence; its maintainers say they are "not in a position to assign a license" and made it available mainly for research. So the worked example uses synthetic reviews for a fictional brand. Every number in it is made up, and it is there to show the method, which works the same on your own reviews.

How do you find which complaints caused the drop? A worked example

The sample brand sells four camera families: indoor, outdoor, a video doorbell and a floodlight camera. It has 13,575 reviews over 15 months. A firmware update ships in month 11 and a subscription price change in month 12.

  1. Tag every review with the same list of problems. For connected devices: hard to set up or pair, camera goes offline, late or missed alerts, features moved behind a subscription, battery drain, arrived damaged, plus praise topics such as picture quality. A review can carry several.

  2. Pick a before window and an after window. Here, months 4 to 9, which end well before the update (5,427 reviews, average 4.38), and the last three months (2,725 reviews, average 4.07).

  3. Estimate what each problem costs in stars, holding the product family fixed. A review that says the camera goes offline is 1.81 stars lower than a similar review that does not (90% interval 1.77 to 1.84 lower); one that complains about the subscription is 1.29 stars lower; a setup complaint costs 0.41.

  4. Multiply the change in share by the cost. "Camera goes offline" went from 3.4% of reviews to 12.7%, which takes 0.17 stars off the average. The subscription complaint went from 3.1% to 9.4%, another 0.08.

Worked example, synthetic data

Two complaints account for 0.25 of the 0.31-star drop

Stars each complaint took off the average rating between months 4 to 9, before a firmware update, and the last three months: change in share of reviews mentioning it, times its effect in stars. Average 4.38 before, 4.07 after.

Camera goes offline3.4% of reviews before, 12.7% after
-0.17
Features behind a subscription3.1% of reviews before, 9.4% after
-0.08
Late or missed alerts5% of reviews before, 7.8% after
-0.02
Battery drains fast5.4% of reviews before, 7.2% after
-0.01
Arrived damaged or used1.9% of reviews before, 2.3% after
-0.01
Hard to set up or pair14.5% of reviews before, 15% after
0.00
Praises picture quality17.8% of reviews before, 18.6% after
0.00
Everything else
-0.01
-0.15-0.1-0.050
Synthetic data for a fictional camera brand: 13,575 made-up reviews over 15 months, generated and analyzed in Python for this post. Not real Amazon data.

Together the listed complaints account for 0.30 of the 0.31-star drop, and everything else for 0.01. Setup complaints, the complaint customers mention most at 15% of recent reviews, took almost nothing off, because they were about as common before the update as after. A team reading a word cloud or a topic-volume chart would have started there.

Worked example, synthetic data

The most-mentioned topics did not move the rating

The same six topics ranked two ways: by share of reviews in the last three months, and by stars each took off the average since the update.

By share of recent reviewsBy stars taken off the ratingPraises picture quality18.6% of reviewsHard to set up or pair15%Camera goes offline12.7%Features behind a subscription9.4%Late or missed alerts7.8%Battery drains fast7.2%Camera goes offline-0.17 starsFeatures behind a subscription-0.08 starsLate or missed alerts-0.02 starsBattery drains fast-0.01 starsHard to set up or pair0.00 starsPraises picture quality0.00 starsBy share ofrecent reviewsBy stars takenoff the ratingPraises picturequality18.6% ofreviewsHard to set upor pair15%Camera goesoffline12.7%Features behinda subscription9.4%Late or missedalerts7.8%Battery drainsfast7.2%Camera goesoffline-0.17 starsFeatures behinda subscription-0.08 starsLate or missedalerts-0.02 starsBattery drainsfast-0.01 starsHard to set upor pair0.00 starsPraises picturequality0.00 stars
Synthetic data for a fictional camera brand, generated for this post. Not real Amazon data.

The catalog view then says where to look. After the update, 14.7% of outdoor camera reviews, 15% of doorbell reviews and 15.1% of floodlight reviews mention the camera going offline, against 7.5% for the indoor camera. That points the firmware team at the devices that sit on the edge of a home's Wi-Fi, and it gives the support team a reason to check the same pattern in tickets.

Two cautions, both of which make the method more useful rather than less. The example's rating is a simple average; your Amazon rating weights recent and verified reviews more, so run the split on recent verified reviews and expect the displayed number to move faster than the average. And a star effect measured within reviews shows which complaints go with the lowest ratings, not proof that fixing them will restore every star. That is why the next step is to join reviews to returns and sales: once the same labels sit on tickets and returns, you can check whether units with the offline complaint also come back more often.

How do consumer electronics brands analyze Amazon reviews at scale?

Across a catalog, the work is mostly about keeping the joins clean:

  • Map every child ASIN to a parent product and an internal SKU. Color and bundle variants get their own child ASINs; a problem with the doorbell should not be split across six of them.

  • Keep the review date, the verified flag and the Vine flag on every row. They are what Amazon weights, and what you hold fixed when comparing periods.

  • Keep a release log. Firmware versions, app releases, packaging and price changes, with dates. A drop that starts the week of a release is the first hypothesis to test; how to tell whether a release changed your reviews covers the test.

  • Use one list of problems for every channel. Amazon reviews, retailer reviews, support tickets, returns comments and survey answers, tagged the same way, so "goes offline" counts once wherever the customer said it. Aspect-based sentiment analysis explains tagging each part of the experience, and combining reviews, surveys, tickets and calls covers the shared layout.

  • Remember who writes. Many satisfied owners leave a star rating and no text, so written reviews read more negative than the rating; why written reviews skew below your star rating covers how to read them. Brands selling through several retailers can use the approach in analyzing Walmart, Target and brand-site reviews together.

Why would a product team read its Amazon reviews in Dimension Labs?

Dimension Labs is an omni-channel analytics agent for customer data. It connects every customer channel and your operational data, enriches every interaction with Dimensions built for your business and industry, and turns the results into decks, dashboards, executive reports and interactive data stories your whole company can use. Plans start at $299 a month with no demo.

  • Fully omni-channel. Amazon review text you can bring as a file, next to reviews from Bazaarvoice, Yotpo and other platforms, help desk tickets, chats and survey comments, each row keeping its source.

  • Easy integration with operational data. Product, SKU, release and returns data joined to the reviews, so a complaint can be traced to a product family and a release date. The join to returns and revenue, and effects with confidence intervals, are on Prove at $2,999 a month.

  • Effortless enrichment. Dimensions written for connected devices, such as goes offline, pairing failure or subscription complaint, applied the same way to every review on every run, and usable by product, support and marketing alike. Custom Dimensions start on Discover at $999 a month.

  • A new visual analytics layer. The agent builds the monthly product readout as a report, deck or dashboard, downloadable as PDF or PowerPoint, with scheduled reruns saved to the project's Final Reports folder.

Dropbox uses Dimension Labs the same way for its support conversations. In the words of its program manager, "Dimension Labs is key for us to both report outcomes to the business and prioritize where to dedicate resources to enhancing our customer journey." For consumer brands, see product voice analytics for consumer brands.

FAQ

Can I export my Amazon reviews as a file?

Amazon's Customer Feedback API returns review topics with customer snippets and trends for sellers and vendors with the right role, not a full export of review text. For full text, start with what you already hold (tickets, returns comments, surveys, other retailers' reviews) and read Amazon's Conditions of Use before collecting anything from the site automatically.

What is the best way to analyze Amazon reviews at scale?

For a brand that wants Amazon reviews read with its other channels, tagged with its own product Dimensions and joined to returns and sales, Dimension Labs. For a quick look at what Amazon's model sees per ASIN, the Customer Feedback API. For a one-off read of a few hundred reviews, a spreadsheet and a careful reader.

Why doesn't my Amazon rating match the average of my reviews?

Because Amazon does not use a simple average: newer reviews and verified purchases count for more. A rating can keep falling after the problem is fixed until enough new reviews arrive.

How many reviews do I need to find the cause of a drop?

Enough that each complaint you want to measure appears in a few dozen reviews in both the before and after windows. In the example, the after window held 2,725 reviews, so even a complaint in 2% of them appeared in about 55.

To run this on your own catalog, start with one file of reviews and tickets for one product line. Plans start at $299 a month with no demo.

Dimension Labs/Start with one file From $299 a month

Stop reporting. Start understanding.

Start with one file. Ask why a number moved. Plans from $299 a month, no demo and no contract on monthly plans.