Amazon Reviews Scraper with Coverage Report

Scrapes Amazon product reviews, and tells you exactly what you got.

Every run reports, per product, how many reviews Amazon has, how many came back, and the reason it stopped. You are charged per review delivered and for nothing else, so a run that comes back empty costs you nothing.

Why this one

Amazon review scrapers are easy to start and hard to finish. The common failure is not an error message, it is a run that reports success, hands you 20 rows, and never mentions that the product has 146 reviews. You find out when the analysis is already built on top of the gap.

This scraper is built around removing that silence.

Every product gets a coverage record.

{
  "asin": "B08N5WRWNW",
  "requestedReviews": 500,
  "retrievedReviews": 100,
  "reviewsWithTextOnAmazon": 317,
  "ratingsOnAmazon": 1482,
  "missingReviews": 217,
  "complete": false,
  "pagesRead": 10,
  "duplicatesDropped": 3,
  "stopReason": "amazon-pagination-limit",
  "stopReasonExplained": "Amazon stops paginating its reviews listing after 10 pages, for every caller"
}

The same summary goes on the run's status message, so you can see it without opening anything:

340 of 1,206 reviews from 4 products. 2 complete, 2 partial: 2 stopped because Amazon stops paginating its reviews listing after 10 pages, for every caller.

You are only charged for reviews that reach your dataset. There is no start fee. If Amazon blocks the run, or the ASIN is wrong, or the product has no reviews, the run costs nothing.

Every URL you paste is scraped. Each product gets its own review budget and its own coverage record, so a long first product cannot eat the allowance of the ones behind it, and one product being blocked does not truncate the rest.

Two numbers that are not the same

Amazon publishes ratings and reviews, and they are different totals:

• Ratings counts every star given, including the many left with no written text.
• Reviews counts only the ones with text.

Only reviews with text exist as individual records anywhere on Amazon. Star-only ratings are counted in the histogram and are never published one by one, so no scraper can return them and any tool implying otherwise is measuring itself against a number it cannot reach. This actor reports both totals and measures its own shortfall against the reviews figure, which is the honest denominator.

Amazon's ten-page limit, and how to get more anyway

Amazon's reviews listing stops after 10 pages of 10 reviews for every caller. That is a hard ceiling of roughly 100 reviews per filter combination, and it is Amazon's, not this actor's. When a run hits it, the coverage record says amazon-pagination-limit rather than pretending the reviews ran out. When Amazon blocks the listing outright and only the product page is readable, it says listing-blocked-product-page-only instead.

The way past it is to run each filter separately, because Amazon applies the ten-page limit to each one on its own:

• Run once per star rating (5 star only through 1 star only) for up to five times the reviews.
• Split further by Most recent and Most helpful.
• Split again by Verified purchases only.

The coverage record on each run tells you which slices are full and therefore worth splitting again.

Input

Field	What it does
productUrls	Amazon product or reviews URLs, any marketplace. Mix amazon.com and amazon.co.uk freely: the marketplace is read from each URL.
asins	Ten-character codes such as B08N5WRWNW, if pasting full URLs is inconvenient.
domain	Which marketplace the ASINs belong to. 21 marketplaces supported. URLs ignore this.
maxReviewsPerProduct	A budget per product, not for the run.
sortBy	Most recent, or most helpful.
filterByStar	All, a single star rating, or positive against critical.
reviewerType	All reviewers, or verified purchases only.
mediaType	All reviews, or only those with photos and video.
formatType	All variants of the product, or only the one you linked.
language	An Amazon locale code such as en_US.
proxyConfiguration	Residential by default, and strongly recommended.

Bad entries are reported in the log rather than silently dropped, so a typo in one URL is visible instead of appearing as a product with no reviews.

Output

One row per review:

Field
asin, domain, productUrl	Which product the review belongs to
reviewId, reviewUrl	Stable identifier and a direct link
rating	The star rating as a number, half stars included
title, body	The review as written, with Amazon's own interface stripped out. No trailing "Read more", no accessibility teaser, no star rating glued to the front of the title. Whitespace is cleaned up and paragraphs are kept
publishedAt	An ISO date, so it can be sorted and filtered
dateText	Amazon's original sentence, kept as written
reviewedInCountry	Which marketplace the reviewer wrote from
author, authorProfileUrl	The public display name Amazon shows on the review
verifiedPurchase	Whether Amazon marked it a verified purchase
helpfulVotes	Helpful count as a number
variant	The size, colour or format reviewed
images	Review photos, at full resolution rather than as thumbnails
scrapedAt	When this row was collected

Coverage records go to a separate coverage dataset and to the COVERAGE key in the run's key-value store, so the main dataset stays clean for a CSV or Excel export.

Limits, stated plainly, and measured

These are results from real runs on 2026-09-03, not estimates.

Amazon blocks its paginated reviews listing from datacentre IP addresses. It answers with an anti-bot interstitial that returns HTTP 200 and looks like a normal page. This actor detects it, retries on fresh sessions, and when the listing stays blocked it falls back to the reviews Amazon embeds in the product page itself. On a free Apify plan, which has no residential proxy, expect roughly 8 to 15 reviews per product and a coverage record saying listing-blocked-product-page-only. The largest measured run was four products and returned 34 reviews with every core field populated on every row. One of those four came back with nothing at all, its coverage record named the reason as blocked, and it was not charged for.

A residential proxy is what unlocks the paginated listing. Amazon treats residential addresses very differently from datacentre ones. That path has not yet been measured here, so this listing does not claim a number for it.

Amazon's reviews listing stops after 10 pages of 10 reviews for every caller, so around 100 per filter combination is the ceiling even when the listing is reachable. Split by star rating and by sort order to go past it.

Star-only ratings cannot be returned by anything, because Amazon does not publish them individually. See the section above.

Only public review pages are read. The reviewer's public display name and their review text, which is what Amazon itself shows on the page. No email addresses, no contact details, no attempt to identify anyone behind a display name.

Whatever happens, the coverage record says which of these you hit, and you are charged only for reviews that reached your dataset.

Pricing

Charged per review written to your dataset. No monthly fee, no start fee, no charge for a run that returns nothing.