Amazon Review Insights & Sentiment

Turn a list of Amazon products into a list of answers about what customers actually think. One row per product — not one row per review — carrying the full star breakdown, the share of ratings that are critical, and the themes buyers keep raising, each with how many customers mentioned it and how that split between praise and complaint. Give it your own catalogue to find out what to fix, or a competitor's to find out what their customers keep grumbling about.

What you can do with it

• See what customers praise and complain about, ranked by how many of them said it, instead of reading a hundred reviews yourself.
• Find a competitor's weak spot — the theme with the most complaints is usually the feature to beat them on.
• Screen a product before you source it by looking at the share of critical ratings rather than the headline star average.
• Prioritise a fix using the number of customers who raised each issue, so the biggest complaint gets worked on first.
• Compare a shortlist side by side — every product returns the same columns, so a hundred candidates sort straight into a spreadsheet.
• Track how opinion shifts over time by running it on a schedule and letting the rows accumulate.

What you get

One row per product. An abridged row:

{
  "asin": "B079VP6DH5",
  "parentAsin": "B0CT5WBY8Q",
  "title": "Bounty Paper Towels Quick Size, White, 16 Family Rolls",
  "brand": "Bounty",
  "url": "https://www.amazon.com/dp/B079VP6DH5",
  "rating": 4.8,
  "ratingsCount": 235305,
  "starHistogram": { "five": 90, "four": 8, "three": 1, "two": 0, "one": 1 },
  "fiveStarPercent": 90,
  "oneStarPercent": 1,
  "positiveSharePercent": 98,
  "criticalSharePercent": 1,
  "ratingScope": "variant-family",
  "reviewsAnalysed": 13,
  "reviewsForThisProduct": 6,
  "reviewsForOtherVariants": 7,
  "verifiedSharePercent": 100,
  "themeSource": "amazon-aspects",
  "themeCount": 8,
  "topTheme": "quality",
  "topThemeMentions": 2696,
  "positiveThemes": "quality, durability, size, functionality, softness, cleaning, brand",
  "negativeThemes": "",
  "mixedThemes": "thickness",
  "mostCriticisedTheme": "quality",
  "mostCriticisedThemeNegativeMentions": 256,
  "mostCriticisedThemePositivePercent": 91,
  "themes": [
    {
      "label": "quality",
      "sentiment": "positive",
      "mentions": 2696,
      "positiveMentions": 2440,
      "negativeMentions": 256,
      "positivePercent": 91,
      "summary": "Customers appreciate the quality of the paper towels…"
    }
  ],
  "customersSayHeading": "Customers say",
  "customersSaySummary": "Customers find these paper towels durable, noting they stay strong even when wet…",
  "collectedAt": "2026-08-07T10:00:00.000Z"
}

Where each number comes from

The row mixes three sources of different sizes, and every figure says which one it came from — because reading a theme count as if it were a review count, or a sample statistic as if it covered everything, is exactly how review analysis goes wrong.

Group	Based on	How big
rating, ratingsCount, the star breakdown and every …SharePercent	Amazon's own star breakdown	Every rating ever left on the product family — by far the largest signal on the row
themes, topTheme, positiveThemes, negativeThemes	Amazon's own analysis of what customers mention	Thousands of customers on a popular product; the mention counts are Amazon's, not counted from the reviews on the row
reviewsAnalysed, verifiedSharePercent, averageSampleRating, sampleReviews	The reviews Amazon publishes on the product page	A dozen or so, ranked by relevance rather than by date

ratingsCount is how many people left a star rating. It is not the number of written reviews, which Amazon does not publish, and the two are never mixed.

For the small number of products where Amazon has published no theme analysis, themes are worked out from the wording of the published reviews instead, and the row says so with themeSource: "review-text". Those mention counts are much smaller by nature, because they can only count the reviews on the page.

Input reference

Setting	Type	Default	Description
Amazon product URLs or ASINs	list of text	—	Required. The products to analyse. Full Amazon links or bare ASINs both work, mixed freely. Duplicates are removed.
Extract recurring themes	yes / no	yes	Include the themes buyers raise, split into positive, negative and mixed. Turn it off for a rating-only summary.
Attach the reviews analysed	yes / no	no	Include the individual published reviews the sample figures were built from, so the numbers can be checked.
Minimum mentions for a theme	whole number	2	A theme must have been raised by at least this many customers to be reported. Maximum 20.
Maximum products	whole number	100	Stop after this many products. Maximum 2,000.
Amazon marketplace	choice	United States	Which country's Amazon site to read. Ratings and reviews differ between marketplaces.
Parallel targets	whole number	4	How many products to work on at the same time. Maximum 10. Leave it at 4 unless you have a reason.

Output fields

Field	Type	Description
asin	string	The product's ASIN.
parentAsin	string	The family this product belongs to, when it has one.
title / brand	string	Product title and brand.
url	string	Clean link to the product.
marketplace	string	The marketplace this reading came from.
rating	number	Average star rating.
ratingsCount	number	How many star ratings the product has.
starHistogram	object	The five star levels, as percentages.
fiveStarPercent … oneStarPercent	number	Each star level on its own column.
positiveSharePercent	number	Share of ratings at 4 or 5 stars.
neutralSharePercent	number	Share at 3 stars.
criticalSharePercent	number	Share at 1 or 2 stars.
ratingBasis / ratingScope	string	What the rating figures are based on, and that they cover the whole product family.
reviewsAnalysed	number	How many published reviews were on the page.
reviewsForThisProduct	number	How many of them are about this exact product.
reviewsForOtherVariants	number	How many are about a sibling variant.
verifiedPurchaseCount / verifiedSharePercent	number	Verified purchases among the published reviews.
verifiedShareBasis	string	States that the verified share is a sample statistic.
averageSampleRating	number	Average rating of the published reviews — usually higher than the overall average, since they are chosen for relevance.
sampleRatingCounts	object	Star counts within the published reviews.
localReviewCount / internationalReviewCount	number	Split of the published reviews by origin.
themeSource	string	amazon-aspects, review-text, none or not-requested.
themeCount	number	How many themes met the minimum.
themes	array	Each theme with its label, sentiment, mention count, positive and negative split, and Amazon's own one-sentence summary.
positiveThemes / negativeThemes / mixedThemes	string	The theme labels in each sentiment group, comma separated.
topTheme / topThemeMentions	string, number	The theme the most customers raised.
mostCriticisedTheme	string	The theme with the largest number of negative mentions.
mostCriticisedThemeNegativeMentions	number	How many of those there were.
mostCriticisedThemePositivePercent	number	How positive that same theme is overall — a popular theme can carry the most complaints and still be praised by most people.
customersSayHeading / customersSaySummary	string	Amazon's own written summary of customer opinion, when it publishes one.
quotedReviewCount / quotedReviewIds	number, array	Reviews quoted by the theme analysis, including ones beyond those published on the page.
sampleReviews	array	The published reviews themselves, when you ask for them.
collectedAt	string	When the reading was taken.

Pricing

Pay per event, with a single event:

What	When you pay
Product analysed	Once for every product row returned. A product that is unavailable, or that could not be read, produces no row and costs nothing.

There is no per-review charge and no second event. Analysing a product with thousands of ratings costs exactly the same as one with a handful.

Worked example. A shortlist of 200 products costs about $0.30. A full catalogue sweep of 2,000 products costs about $3.00.

Limits & what this actor cannot do

• Amazon shows only a selection of customer reviews on a public product page. This actor returns that public selection together with the full star breakdown, not every review ever written. Any tool promising every review of a product is not describing what Amazon publishes.
• Those published reviews are pooled across a product's variant family, so some of them are about a sibling size or colour. The row tells you exactly how many are about the product you asked for and how many are not.
• The published reviews are ranked by relevance, not by date, so they are not "the latest reviews" and their average rating is usually flattering compared with the overall star breakdown. Treat every sample figure as a sample.
• Ratings, the star breakdown and the theme analysis are published by Amazon for a whole product family. Two variants of the same product return the same figures, so this actor does not offer per-variant sentiment — it would be the same answer wearing a different ASIN.
• The ratings count is not a count of written reviews. Amazon does not publish the second number anywhere on a public product page, so it is not reported and is never substituted.
• Customer questions and answers are not publicly listed and are not returned.
• Themes come from Amazon's own analysis wherever it exists. Where it does not, they are worked out from the wording of the published reviews, which is a much smaller basis, is English-oriented, and is marked as such on the row.
• Opinion data is a snapshot at the moment of collection and keeps changing afterwards.
• Ratings and reviews differ between Amazon marketplaces; results reflect the marketplace you select.
• Products that are removed, restricted or unavailable in the selected marketplace are reported as unavailable rather than returned empty.
• Speed depends on how many products you analyse and on Amazon's own response times. No fixed throughput is promised.
• Amazon's terms govern automated access. You are responsible for using the data lawfully and in line with the source site's terms, and for handling any personal data in line with applicable privacy law.

FAQ

Do I need an Amazon account? No. It reads only what Amazon shows the public.

Does it need my login or password? No, and it will never ask for one.

Can I schedule it? Yes. Running it weekly or monthly on the same list is a good way to watch opinion shift over time.

Is the data complete? The rating distribution is complete — it covers every star rating the product family has received. The theme analysis is Amazon's own and draws on far more reviews than the page displays. The individual reviews attached to the row are the selection Amazon publishes publicly, which is around a dozen per product, so anything derived from them is labelled as a sample.

Why does the theme mention count dwarf the number of reviews on the row? Because they measure different things. The mention counts come from Amazon's analysis of all its reviews; the reviews on the row are only the ones Amazon chose to display. Both are reported so the difference is visible.

Why is a theme both "most criticised" and positive? Because a theme everybody talks about collects the most complaints in absolute terms while still being praised by the large majority. That is why the row carries the number of complaints and the positive percentage side by side.

Why is negativeThemes often empty? Because Amazon labels a theme negative only when opinion genuinely leans that way. Themes people disagree about show up under mixedThemes, and the complaint counts inside themes let you rank pain points even when nothing is labelled negative.