Amazon Review Aggregates & Extended History Scraper

Collect Amazon customer reviews and rating summaries without a login and without any proxy setup. Point it at product URLs, bare ASINs, or search keywords and it merges three public sources — the product page's own review block (desktop and mobile), Amazon's public extended review feed for the same ASIN, and optionally the same product's pages on other Amazon marketplaces — then deduplicates everything and returns one clean review dataset plus a per-product rating summary.

What this actor does

• Extended review history — the product page alone shows only ~8–13 "top reviews"; the public extended feed adds up to 100 reviews per star/sort bucket, and walking every bucket yields 100–850+ unique reviews for popular US products (measured maxima: 856), merged and deduplicated by review ID and content fingerprint.
• Multi-marketplace merge — collect the product page on additional Amazon marketplaces and union the deduplicated reviews, since each storefront carries a different (partially overlapping) set.
• Rating aggregates — product title, average rating, global ratings count, the full star histogram with derived per-star counts, and Amazon's AI-generated "Customers say" summary with per-aspect sentiment.
• Filters that actually narrow the data — star (including positive/critical), verified purchases, keywords, date, media-only, plus sort by most helpful or most recent.
• Resilient by design — the review wall is probabilistic, so every product page is retried across fresh sessions and both desktop/mobile paths; blocked pages escalate to a residential connection automatically.
• Fields a review or product page doesn't provide (no reviewer images, no Vine badge, no helpful votes) are simply left out of that record — empty fields are omitted, never sent as null.

Output per review

Field	Description
productAsin	Amazon ASIN of the requested product.
reviewedProductAsin	Variant ASIN the review was written for, when the page exposes it.
ratingScore	Star rating, 1–5.
reviewTitle / reviewDescription	Review headline and full text.
reviewUrl	Direct link to the individual review, when the source exposes one.
reviewReaction	Raw "N people found this helpful" text, when present.
reviewedIn	Raw "Reviewed in <country> on <date>" text.
isVerified	Verified Purchase flag.
isVineVoice	Amazon Vine reviewer flag, when evidenced.
variant	Product variant reviewed (e.g. size/color), when shown.
reviewImages	Reviewer-uploaded image URLs, when includeImages is on.
reviewVideos	Reviewer-uploaded videos (url, thumbnailUrl), when present.
position	Rank of this review within its product's returned set.
reviewId	Amazon's own review identifier, when the source exposes one.
helpfulCount	Parsed numeric helpful-vote count.
reviewDate	Parsed ISO date.
reviewLocation	Parsed reviewer country/region.
sentimentHint	Derived positive / neutral / negative label.
wordCount	Review length (character count for CJK scripts).
hasImages / imageCount / hasVideos / videoCount	Derived media counts.
reviewHash	Stable content fingerprint used for cross-source deduplication.
source	Which pipeline produced the row: productPage or extendedHistory.
sourceMarketplace	Amazon domain the row came from.
reviewerName / reviewerProfile / reviewerAvatar	Reviewer identity fields, only present when includeGdprSensitive is true.
scrapedAt	UTC timestamp of the scrape.
sourceUrl	The product URL that was scraped.

Output per product (rating aggregate)

When includeAggregates is on (default), one additional record per product carries:

Field	Description
type	Always "amazon_review_aggregate".
productAsin	Amazon ASIN.
productTitle	Product title from the same page.
averageRating	Overall star rating shown on the product page.
globalRatingsCount	Total number of ratings, not just the reviews collected.
starsBreakdown	Percentage per star, 5 through 1.
ratingCountByStar	Derived count per star (percentage × globalRatingsCount, rounded).
customersSay	Amazon's AI-generated "Customers say" summary, when Amazon shows one.
customersSayAspects	Per-aspect sentiment breakdown (label, sentiment, mentionCount, summary).
productUrl	Canonical /dp/{ASIN} product URL.
scrapedAt / sourceUrl	Scrape timestamp and source URL.

Status records

If a run can't return real reviews for a product, the dataset gets one typed status record instead of being left empty, identified by its type value: amazon_reviews_walled_us, amazon_reviews_no_star_match, amazon_reviews_blocked, or amazon_reviews_no_products_discovered. A blocked/checkpointed page is always reported in the run's status message — the actor never fails silently.

Input

Field	Type	Default	Description
productUrls	Array of strings	–	Amazon product URLs, or bare ASINs expanded using country. Leave empty to use keywords/categoryUrls instead.
keywords	Array of strings	[]	Search keywords to auto-discover products, used only when productUrls is empty.
categoryUrls	Array of strings	[]	Amazon search/category URLs to auto-discover products from, used only when productUrls is empty.
maxProductsToDiscover	Integer	10	Maximum products to discover from keywords/categoryUrls (1–50).
extendedHistory	Boolean	true	Merge Amazon's public extended review feed (typically 100–600+ unique reviews for popular US products) with the product-page sample.
mergeMarketplaces	Array of domains	[]	Extra Amazon marketplaces to fetch the product page from and merge deduplicated reviews.
includeAggregates	Boolean	true	Emit one rating-aggregate record per product.
country	String (select)	"amazon.com"	Amazon marketplace domain (19 supported).
maxReviews	Integer	100	Hard cap per product after merging and deduplication (1–2000).
sortBy	String (select)	"helpful"	helpful or recent.
filterByStar	String (select)	"" (all)	Restrict to one star rating ("1"–"5"), positive (4–5), or critical (1–2).
verifiedPurchasesOnly	Boolean	false	Only return Verified Purchase reviews.
filterKeywords	Array of strings	[]	Keep only reviews whose title or text contains at least one keyword.
reviewsNewerThan	String (date)	–	Keep only reviews dated on or after this date.
mediaReviewsOnly	Boolean	false	Keep only reviews with at least one image or video.
includeImages	Boolean	true	Extract reviewer-uploaded images and videos.
includeGdprSensitive	Boolean	false	Include reviewer name, profile link, and avatar.

Example: extended history for one product

{
    "productUrls": ["https://www.amazon.com/dp/B09X7MPX8L"],
    "extendedHistory": true,
    "maxReviews": 300,
    "sortBy": "recent"
}

Example: only critical reviews with images

{
    "productUrls": ["B09X7MPX8L"],
    "filterByStar": "critical",
    "mediaReviewsOnly": true,
    "verifiedPurchasesOnly": true,
    "maxReviews": 100
}

Example: merge several marketplaces

{
    "productUrls": ["https://www.amazon.de/dp/B09X7MPX8L"],
    "country": "amazon.de",
    "mergeMarketplaces": ["amazon.co.uk", "amazon.fr", "amazon.it"],
    "maxReviews": 200
}

Example: keyword discovery with a date window

{
    "keywords": ["wireless earbuds"],
    "maxProductsToDiscover": 5,
    "reviewsNewerThan": "2026-01-01",
    "filterKeywords": ["battery", "sound"],
    "sortBy": "recent"
}

Use cases

• Voice-of-customer analysis — pull hundreds of review texts per product line for sentiment, topic, or keyword analysis.
• Competitive benchmarking — compare averageRating, starsBreakdown, and "Customers say" summaries across competing ASINs.
• Review monitoring — schedule runs on a watch-list to track new reviews, ratings, and helpful-vote counts over time.
• Trust and quality signals — surface isVerified/isVineVoice flags and reviewer-uploaded media for product-quality dashboards.
• Localization research — merge reviews for the same product across marketplaces to compare region-specific feedback.
• Critical-review triage — filter to critical reviews with media to jump straight to the problems customers photograph.

FAQ

How many reviews can I get per product? The product-page block alone shows ~8–13 top reviews. With extendedHistory on (default), the public extended feed adds up to 100 reviews per star/sort bucket; walking all buckets measured 100–856 unique reviews across a 35-ASIN survey (17 ASINs had feed data; the rest fall back to the product-page sample). Set maxReviews high enough (e.g. 500–1000) to receive the full depth, or keep it lower for a fast, cheap run.

Is a proxy or login required? No. The actor works without any credentials and without any proxy setup; it chooses a working connection automatically and only escalates to a residential connection for a request that Amazon actually blocks. It never uses your Amazon account or cookies.

Why did one product return only a handful of reviews? The extended feed does not cover every ASIN — in a 35-ASIN survey, 17 had feed data and 18 exposed few or no rows (the feed carries Amazon.com reviews only). When that happens the actor still returns the product-page sample plus the rating aggregates, and the run's status message reports what was collected.

Why are some fields missing on some reviews? Every source exposes a different subset: the extended feed has no review permalink or review ID (rows are deduplicated by a content fingerprint), some marketplace pages hide reviewer profiles for cross-posted reviews, and Amazon only renders the AI "Customers say" summary on some products and mostly on amazon.com. Missing fields are omitted, never filled with placeholders.

What does source mean? It tells you which pipeline produced the row: productPage (the product page's inline block) or extendedHistory (the public extended feed). sourceMarketplace tells you which Amazon storefront it came from.

Is ratingCountByStar exact? It is derived: Amazon publishes the histogram as percentages plus one global ratings count, so the per-star counts are that percentage applied to the total and rounded (validated on 61 live pages: percentages always sum to 100 and the derived counts are within ±1 of the global total). starsBreakdown holds the exact percentages.

Can I get the complete review history? The actor returns everything Amazon publishes to anonymous visitors: the product page's sample plus the public extended feed (and any merged marketplaces). Amazon keeps the fully paginated list behind a sign-in wall, and this actor never asks for an account, cookie, or password — so it does not read it.

Do reviewer name, profile, and avatar always appear? No — only when includeGdprSensitive is true. Make sure you have a lawful basis before enabling it.

Is this affiliated with Amazon? No. This is an independent, third-party actor that reads publicly available Amazon product and review pages and is not operated, endorsed, or sponsored by Amazon.com, Inc.