Amazon Reviews Scraper — Ratings Histogram & Customers Say

Get reviews for any Amazon product by ASIN or URL, from the public product page, with no login, no cookie pool and no CAPTCHA solving. One row per review: reviewId, rating, title, body, reviewerName, reviewerId, date, reviewCountry, verifiedPurchase, vineReview, helpfulVotes, variant, images, hasVideo — and on every row the product's productAverageRating, productTotalRatings, productRatingHistogram (the 5→1-star percentages) and the AI "Customers say" summary with its per-aspect sentiment (customersSayAspects: aspect, positive/negative/mixed, mention counts, evidence snippets linked to review ids). That last field is decoded from the page's own serialised state and shipped on every row at no extra charge.

Scope: up to ~13 reviews per product per marketplace (13 on a well-reviewed listing; thin or new listings return fewer — 1 to 8 in our 60-ASIN load test) — the marketplace's own "Top reviews" block plus "Top reviews from other countries". That is everything Amazon serves without a login, on every marketplace and every proxy rung we measured (details below). If you need the full paginated review history of a product, read "Read this before you buy rows" first: it explains why no logged-out scraper can deliver that today, and what the ones claiming to are doing.

{
  "productUrls": ["B09B8V1LZ3", "B0BDHWDR12", "B00FLYWNYQ"],
  "marketplaces": ["amazon.com"],
  "maxItems": 100
}

Click Try for free and hit Start — that block is the input the Actor ships with. A run on those shipped defaults returned 34 reviews from 3 product pages in 17 seconds for $0.10, with rating, title, body, date and the "Customers say" aspects filled on 100% of rows.

---

The wedge: reliability you can schedule, with the histogram and Customers-say aspects on every row

The leader in this lane (4 GB browser build) documents that "residential proxies return results more stably", runs at a 3.4% failed / timed-out / aborted rate over its last 77,045 runs, charges a $0.50 minimum per run, and is rated 3.92; the runner-up is rated 3.88. Both ratings were read off the Store on 2026-08-22. Every rated Actor in the lane sits under 4.0 — review scraping on Amazon is a reliability problem, not a parsing problem.

This Actor takes the other trade: only the public page, so there is no account to get banned, logged out or cookie-expired — a blocked session is simply retried on a fresh one. Plain HTTP, one request per product, a fresh proxy session on every retry, a 4 GB browser nowhere in sight. What you give up is depth (up to ~13 reviews per page instead of a paginated history). What you get in exchange:

• customersSayAspects — Amazon's own AI reading of all reviews of the product, not just the ones on the page: for each aspect ("sound quality", "battery life", "fit"…) the sentiment label, how many reviews mention it, how many of those agree with the label, the percentage, and up to 4 evidence snippets each tied to a reviewId. On the shipped sample that is 7–8 aspects per product built from thousands of underlying reviews (e.g. "performance: positive, 26,997 mentions, 97%"). For voice-of-customer work this is a better summary of the product than 500 scraped rows, and it costs the same as the 8 rows it arrives with.
• productRatingHistogram — {"5": 82, "4": 11, "3": 3, "2": 1, "1": 3}, the star distribution behind the average, with productTotalRatings (198,409 for the sample Echo Dot).
• Marketplace fan-out — give a bare ASIN and a list of marketplaces and you get each marketplace's own top-review set for the same listing, deduplicated on reviewId across the run. Reviews carry reviewCountry and language so a mixed dataset stays sortable.
• No minimum charge, no start fee, no browser compute. $0.003 per review delivered; a product that returns nothing costs nothing.

---

Read this before you buy rows

Five things that would otherwise turn into a refund request.

1. The page is the ceiling. Amazon's full review list (/product-reviews/<ASIN>, its star filters, the "see more reviews" link, the in-page AJAX endpoints and the lazy-widget stream) is closed to every logged-out client — measured 2026-08-22 on direct, datacenter and residential exits, on amazon.com, .co.uk and .de: the paginated list and its star filters 302 to /ap/signin on 27 of 27 calls, the medley AJAX endpoints return Amazon's 404, and the lazy-widget stream returns 0 bytes. No route returned a review. This Actor will never log in, never replay a session cookie and never solve a CAPTCHA, so maxReviewsPerProduct cannot raise the per-page count above what the page shows (up to ~13). If another scraper returns hundreds of reviews per ASIN, it is reaching a surface this Actor deliberately does not touch — check how before you depend on it.
2. ASINs are marketplace-specific. The Echo Dot is B09B8V1LZ3 on amazon.com and amazon.ca but a different ASIN on amazon.co.uk and amazon.de, where B09B8V1LZ3 is an HTTP 404. A 404 is logged, counted and not billed, but it also returns no rows — paste the marketplace's own product URL if you want that marketplace's reviews.
3. reviewerId / reviewerProfileUrl are usually absent on "other countries" reviews (measured 76.5% fill on the sample: 100% on local reviews, 20% on the "other countries" block). Amazon usually renders foreign reviews without a profile link. reviewerName is always present.
4. helpfulVotes is a real 0, not a missing value, and images is an empty array for the ~77% of reviews with no photo. variant (e.g. "Color: Deep Sea Blue | Configuration: Device only") only exists on products with variations — 61.8% on the sample.
5. Filters remove rows you would otherwise have paid for, and that is the point. filterByRatings, verifiedOnly and includeInternationalReviews: false are applied before billing; a page whose 8 reviews all fail your filter delivers 0 rows and charges $0.

---

What you get

One row per unique review. rating is an integer 1–5. date is ISO YYYY-MM-DD parsed from Amazon's "Reviewed in the United States on June 29, 2026" line (English, German, French, Spanish, Italian, Portuguese, Dutch, Swedish, Polish and Japanese date formats are handled; dateRaw always keeps the original text). scrapedAt is a full UTC timestamp.

Group	Fields	Example
Identity	reviewId, reviewUrl, asin, reviewedAsin, marketplace, position, isInternational	R3UN7L2595T26X · B00FLYWNYQ (review written on variant B06Y1YD5W7) · amazon.com
Review	rating, title, body, language	5 · "Very useful and easy to use" · 3 paragraphs joined with \n · en-US
Reviewer	reviewerName, reviewerId, reviewerProfileUrl	Daimer Castillo · AGW3CMD6VFOG25F4S2EPAETVLP3A
When / where	date, dateRaw, reviewCountry	2026-06-29 · United States
Trust signals	verifiedPurchase, vineReview, helpfulVotes	true · false · 2
Media / variant	images[], hasVideo, variant	2 image URLs · Size: 3 Quarts
Product summary (on every row)	productUrl, productTitle, productAverageRating, productTotalRatings, productRatingHistogram{}	4.7 · 173,570 · {"5":84,"4":9,"3":3,"2":1,"1":3}
Customers say (on every row)	customersSaySummary, customersSayAspects[]	8 aspects with sentiment, mentions, %, snippets
Provenance	scrapedAt	2026-08-22T17:41:05.107Z

The dataset ships with a Reviews table view (ASIN, stars, title, body, reviewer, date, country, verified, helpful votes, variant, product average, total ratings, Customers say, link) so the Console and the Store page render without configuration.

---

Field fill — measured on 34 reviews

The shipped default run (3 products on amazon.com, 2026-08-22), sorted by fill so the sparse fields are impossible to miss. Booleans are reported as the share that is true.

Field	Fill	Notes
reviewId / reviewUrl / asin / marketplace	100%
rating	100%	integer 1–5, read from the star icon's class, not from locale text
title / body	100%	full text; "Read more" is already expanded in the HTML
reviewerName	100%
date / dateRaw / reviewCountry	100%
productAverageRating / productTotalRatings / productRatingHistogram	100%	all five star buckets
customersSaySummary / customersSayAspects	100% on the 3-product sample; ~88% of rows on a wider 26-product run	6–8 aspects; null on listings Amazon has not generated it for (6 of 26 products)
verifiedPurchase = true	100%
helpfulVotes > 0	64.7%	0 is a real 0
reviewerId / reviewerProfileUrl	76.5%	100% on local reviews, 20% on "other countries" reviews
variant	61.8%	only on products with variations
isInternational = true	29.4%	the "Top reviews from other countries" block
images (≥1)	23.5%
vineReview = true	0%	Vine badges are rare in top-review slots
hasVideo	0%

The headline that could mislead you: customersSayAspects summarises the whole product, not the rows in your dataset. Its mention counts come from Amazon's model over every review of the product (26,997 mentions of "performance" on the Instant Pot), while the dataset holds the up-to-13 reviews the page shows. Treat it as product-level metadata that happens to ride on each row.

Every run prints its own measured fill in the log.

---

How to run it

1. ASINs or product URLs (the usual choice)

{
  "productUrls": [
    "B09B8V1LZ3",
    "https://www.amazon.com/dp/B0BDHWDR12",
    "https://www.amazon.co.uk/dp/B0BDHWDR12"
  ],
  "marketplaces": ["amazon.com"],
  "maxItems": 500
}

A bare ASIN is read on every marketplace in marketplaces; a full URL is read on the marketplace in the URL, whatever the list says. Accepted URL shapes: /dp/…, /gp/product/…, /gp/aw/d/…, /product-reviews/…, anything with ?asin=; anything else with a 10-character ASIN segment is tried too. amzn.to short links are not resolved.

2. The same ASIN on several marketplaces

{
  "productUrls": ["B0BDHWDR12"],
  "marketplaces": ["amazon.com", "amazon.ca", "amazon.com.au"],
  "maxItems": 200
}

Each marketplace shows its own top-review set, so this is the cheapest way to get more reviews for one listing without a login — provided the ASIN exists there (gotcha 2). Supported: amazon.com, .co.uk, .de, .ca, .fr, .it, .es, .com.au, .in, .co.jp, .com.mx, .com.br, .nl, .se, .pl, .ae, .sa, .sg, .com.be, .ie. The proxy country follows the marketplace automatically.

3. Critical reviews only

{
  "productUrls": ["B00FLYWNYQ"],
  "filterByRatings": ["1", "2", "3"],
  "verifiedOnly": true
}

Keeps only 1–3-star verified-purchase reviews from the page. Remember the page is the ceiling: a product whose top reviews are all 5-star returns 0 rows here (and charges nothing).

4. Local reviews only

{ "productUrls": ["B00FLYWNYQ"], "includeInternationalReviews": false }

Drops the "Top reviews from other countries" block (the 5 rows flagged isInternational: true on the Instant Pot sample).

Scheduling and integrations

Save the input as a Task (Console → the Actor → Save as Task), attach an Apify Schedule (0 6 * * 1 for a weekly Monday refresh of a product list) and read the dataset over the REST API or the standard integrations (Zapier, Make, n8n, webhooks, MCP). Starting a run from the API:

curl -X POST "https://api.apify.com/v2/acts/scrapersdelight~amazon-reviews-scraper/runs?token=YOUR_TOKEN" \
  -H 'Content-Type: application/json' \
  -d '{"productUrls":["B09B8V1LZ3"],"marketplaces":["amazon.com","amazon.ca"]}'

---

Sample row

A real row from the shipped-default run, 2026-08-22 (body and aspects truncated here; full length in the dataset).

{
  "reviewId": "R3UN7L2595T26X",
  "reviewUrl": "https://www.amazon.com/portal/customer-reviews/srp/-/R3UN7L2595T26X/ref=cm_cr_dp_d_rvw_ttl",
  "asin": "B00FLYWNYQ",
  "reviewedAsin": "B06Y1YD5W7",
  "marketplace": "amazon.com",
  "rating": 5,
  "title": "Very useful and easy to use",
  "body": "I bought this Instant Pot to save time when cooking, and it has been a great purchase. It cooks food much faster than traditional methods and is very easy to use.\nI have used it for rice, soups, and meat, and everything comes out well cooked…",
  "language": "en-US",
  "reviewerName": "Daimer Castillo",
  "reviewerId": "AGW3CMD6VFOG25F4S2EPAETVLP3A",
  "reviewerProfileUrl": "https://www.amazon.com/gp/profile/amzn1.account.AGW3CMD6VFOG25F4S2EPAETVLP3A",
  "date": "2026-06-29",
  "dateRaw": "Reviewed in the United States on June 29, 2026",
  "reviewCountry": "United States",
  "isInternational": false,
  "verifiedPurchase": true,
  "vineReview": false,
  "helpfulVotes": 2,
  "variant": "Size: 3 Quarts",
  "images": [
    "https://m.media-amazon.com/images/I/51FZBJue3hL._SY500_.jpg",
    "https://m.media-amazon.com/images/I/51t5lJQauRL._SY500_.jpg"
  ],
  "hasVideo": false,
  "position": 1,

  "productUrl": "https://www.amazon.com/dp/B00FLYWNYQ",
  "productTitle": "Instant Pot Duo 7-in-1 Electric Pressure Cooker, Slow Cooker, Rice, Steamer, Sauté, Yogurt Maker, Warmer & Sterilizer, Includes App With Over 800 Recipes, Stainless Steel, 6 Quart",
  "productAverageRating": 4.7,
  "productTotalRatings": 173570,
  "productRatingHistogram": { "5": 84, "4": 9, "3": 3, "2": 1, "1": 3 },
  "customersSaySummary": "Customers find this pressure cooker to be the best kitchen appliance ever, praising its ease of use with features like \"set and forget\" functionality…",
  "customersSayAspects": [
    {
      "aspect": "performance",
      "sentiment": "positive",
      "summary": "Customers find the pressure cooker performs exceptionally well, describing it as the best kitchen appliance ever.",
      "mentions": 26997,
      "sentimentMentions": 26207,
      "mentionsPercentage": 97,
      "snippets": [
        { "reviewId": "R3C1PP7MFJNS5R", "text": "...There is really nothing I don't like about this product. Five stars!" },
        { "reviewId": "R1K2DIKH3XRRY4", "text": "Love it but wish I had waited a bit since I also found a larger one for not much more money…" }
      ]
    }
    // … 7 more aspects
  ],
  "scrapedAt": "2026-08-22T17:41:05.107Z"
}

Fields people misread:

• reviewedAsin is the variant the reviewer bought; asin is the one you asked for. They differ whenever a listing has variations (3 Quarts vs 6 Quarts).
• mentions counts reviews mentioning the aspect; sentimentMentions is how many of those carry the labelled sentiment; mentionsPercentage is the ratio. sentiment is positive, negative or mixed.
• position restarts at 1 for the international block; use isInternational + position together.

---

Input

Field	Type	Default	What it does
productUrls	string list	3 sample ASINs	ASINs or product URLs, one per line. Empty = the documented sample.
marketplaces	string list	["amazon.com"]	Marketplaces every bare ASIN is read on.
includeInternationalReviews	boolean	true	Keep the "Top reviews from other countries" block.
filterByRatings	multi-select	[] (all)	Keep only these star ratings. Applied before billing.
verifiedOnly	boolean	false	Verified-purchase reviews only. Applied before billing.
maxReviewsPerProduct	integer	0 (all on page)	Cap per product × marketplace. Cannot exceed what the page shows.
maxItems	integer	100	Your hard cost cap across the run. 100 = $0.30. 0 = no cap (an internal ceiling of 100,000 rows still applies).
requestConcurrency	integer	4	Product pages in flight (1–6).
proxyConfiguration	proxy	Apify RESIDENTIAL	Country follows the marketplace unless you pin one.

---

Pricing

$0.003 per review delivered — $3 per 1,000. Charged on the review-scraped event. No start fee, no minimum charge, no monthly fee from this Actor.

Run	Reviews	Cost
The shipped default (3 products, amazon.com)	34	$0.10
100 products, ~10 reviews each	~1,000	~$3.00
1,000 products × 3 marketplaces	~30,000	~$90

• You are charged for rows delivered. A review removed by your filters, a duplicate seen on a second marketplace, a product that 404s or a page that failed every attempt is never charged.
• Rows are charged as they are pushed (Actor.pushData(items, 'review-scraped')), so a budget cap stops on whole rows, never a half-billed dataset.
• maxItems is your hard cost cap (default 100). It counts unique delivered reviews.
• The product summary and the "Customers say" aspects ride on every row at no extra charge.

---

Honest limits

• At most ~13 reviews per product page, and fewer on thin listings. Amazon's paginated review list is behind a login on every marketplace and every proxy rung measured (27/27 calls redirected to sign-in). This Actor does not log in, and no input changes that. It is a top reviews + product sentiment tool, not a review-history exporter.
• Datacenter proxies are unreliable here. Measured 2026-08-22 on 7 valid ASINs: direct 7/7, Apify datacenter 2/7 (5 CAPTCHA interstitials), Apify RESIDENTIAL 6/7 (one transient proxy error). The default is RESIDENTIAL and the Actor retries every failure on a fresh session (up to 4 attempts); a page that fails all four is reported, not billed.
• A CAPTCHA interstitial is never solved. It is treated as a per-session block and retried on a new session. If Amazon challenges every session of a run, you get 0 rows, a status message saying so, and no charge.
• ASINs are marketplace-specific (see gotcha 2) — a 404 on amazon.de for a US ASIN is Amazon's catalogue, not a scraping failure.
• reviewerId is usually absent on international reviews (2 of 10 carried one on the sample); reviewerName is always present.
• "Customers say" is not on every product. Amazon only generates it for listings with enough reviews; on new or thin listings customersSaySummary / customersSayAspects are null — on a wider 26-product run that was 6 of 26 products. The three sample products all carry it (7–8 aspects each).
• Review text is returned in the language it was written in (language tells you which); the Actor does not use Amazon's translate button.

---

How it works, and what it cost to make reliable

One plain-HTTP request per product × marketplace to /dp/<ASIN>, with a desktop-Chrome header set and an Accept-Language matched to the marketplace. The 1.5–3 MB page is stripped of script, style and SVG before parsing; reviews are read from the review-card containers (each carries the review id, the ASIN it was written on, locale and marketplace), the histogram from the star-filter links and progress meters, and the "Customers say" aspects from the base64 JSON Amazon embeds in that block's own HTML comments. No browser, no JavaScript execution, no API key.

Transport ladder, measured 2026-08-22 through Apify, one attempt per ASIN, fresh session each:

Rung	Result (7 valid ASINs)
Direct, no proxy (reported and then ignored — no Apify container has a home IP)	7/7 pages with reviews
Apify proxy, datacenter / auto	2/7 — 5 CAPTCHA interstitials (native validateCaptcha, no third-party vendor), 1 stripped page with no review block
Apify RESIDENTIAL, country matched to the marketplace	6/7 — 1 transient proxy 590, cleared on retry
amazon.ca via RESIDENTIAL country-CA	2/2

Routes that are closed to a logged-out client (all rungs, 2026-08-22): /product-reviews/<ASIN> page 1, page 2 and ?filterByStar= → 302 to /ap/signin (27/27); /hz/reviews-render/ajax/medley-reviews/get/ and /medley-filtered-reviews/get/ → Amazon's generic 404 (decommissioned); the /portal/customer-reviews/ajax/lazy-widgets/stream loader → HTTP 200 with 0 bytes; /dp/<ASIN>?filterByStar= and ?sortBy= → the same unfiltered page. The ladder stops there on purpose: the next rungs are a session cookie or a CAPTCHA solver, and this Actor uses neither.

Proof runs, through Apify RESIDENTIAL (platform runs, 2026-08-22): see the run log of any run — every run prints HTTP: N attempt(s) — ok / CAPTCHA / stripped / sign-in / 404. Measured on the proof runs for this build: prefill, {} and {"helloWorld":123} runs on build 0.1.1 (2026-08-22, 1024 MB, timeout 300 s) each fetched 3/3 product pages clean on the first attempt — 9/9 through RESIDENTIAL, 0 CAPTCHA, 0 stripped, 34 reviews each in 17 s / 8 s / 12 s; the offline validation run on the same defaults through RESIDENTIAL was 3/3 as well (12/12). Two unbounded 60-ASIN runs at 1024 MB fetched 23 and 26 product pages clean through RESIDENTIAL before the container hit its memory limit — see the memory note under "When a run fails".

robots.txt on amazon.com does not disallow /dp/. Defaults stay gentle: concurrency 4, one request per product, fresh session per attempt.

---

Duplicates — measured

Dedupe key is reviewId, run-wide, applied before billing and checked after: if the dataset ever held two rows with the same review id the run reports it in the log and status message. Measured on the shipped default: 34 rows, 34 unique ids, 0 duplicates. Duplicates do arise by design when the same ASIN is read on several marketplaces (a review can appear in one marketplace's local block and another's international block) — those are dropped and logged.

---

When a run fails

• 0 rows → the run succeeds with a status message naming why (all pages 404 on that marketplace / all attempts CAPTCHA-challenged / every review removed by your filters) and charges nothing. The log carries the per-attempt HTTP tally.
• Invalid ASIN or URL → warned and skipped; if nothing usable remains the documented sample runs instead.
• Time budget → the Actor stops starting new pages ~30–60 s before its timeout, pushes what it has and sets a status message with the number of pages left. It never ends TIMED-OUT on its own account.
• Duplicate leak → the run logs the error and sets a status message naming it; the rows are already pushed. Dedupe on reviewId in your own store as a belt-and-braces check.
• Large lists need more memory. At the default 1024 MB, concurrency 4, a 60-product run peaked at 1.0 GB and was killed by the platform (exit 137) partway through — rows already delivered stay billed. For lists above ~40 products raise the run memory to 4096 MB or split the list.

---

Who buys this

• Brands and agencies doing voice-of-customer on Amazon — customersSayAspects is the sentiment breakdown, with counts, for a whole product in one row.
• Sellers tracking competitors — schedule a product list weekly; the histogram + total-ratings trend is the early signal, the top reviews are the evidence.
• Marketplace analytics and repricing tools — rating, histogram and review velocity per ASIN per marketplace, with no account to maintain.
• AI agents — a single deterministic call per ASIN returns the product's rating summary, aspect sentiment and the reviews Amazon itself ranks highest. No minimum charge, pay-per-event.
• Review-monitoring and alerting — 1–2-star filter + verifiedOnly, deduped on reviewId across runs by your own store.

---

FAQ

Does this need an account, a login or cookies? No. It reads the public product page only. There is no login, no cookie jar, no CAPTCHA solving service and no browser automation.

Why only ~10 reviews per product? Because that is what Amazon serves to a logged-out client. The paginated review list redirects to sign-in on every rung we measured. Read "Honest limits".

How do I get more reviews for one product? Read it on every marketplace where the ASIN exists (marketplaces), and paste variant ASINs — variants with their own listing page carry their own top-review block.

What is "Customers say"? Amazon's AI summary of all reviews of a product, with per-aspect sentiment and mention counts. This Actor decodes it from the page and ships it on every row.

Do I need a residential proxy? Yes, realistically: 2/7 on datacenter vs 6/7 on residential in the ladder above. RESIDENTIAL is the default and the country follows the marketplace.

Two runs — will I get duplicates? Within a run, never — every reviewId is deduped run-wide before push. Across runs, yes if the same page shows the same top reviews — dedupe on reviewId in your store.

Do I get charged for rows a filter removed, or for a 404? No. Only delivered rows are charged.

---

Legal & fair use

This Actor reads public product pages on Amazon marketplaces without logging in, without cookies and without touching any authenticated surface. Amazon's Conditions of Use restrict automated access and direct commercial catalogue use to its Product Advertising API (which publishes ratings but not review text); you are responsible for complying with Amazon's terms and with privacy law in your jurisdiction — rows contain reviewer display names and, for local reviews, public profile ids.

Amazon® is a trademark of Amazon.com, Inc. This Actor is not affiliated with, endorsed by, or connected to Amazon.

---

Feedback

Found a missing field or a marketplace whose date format is not parsed? Open an issue on the Issues tab, and if the Actor earns it, a review on the Reviews tab helps other buyers find it.