Amazon Reviews Scraper - Thousands, Not Hundreds

How does Amazon Reviews Scraper work?

This Actor collects Amazon product reviews past the limit that stops every other scraper, and tells you what percent of them you received.

Amazon returns at most ten pages of ten reviews behind any single filtered view. Ask for a product's reviews and you get a few hundred, whatever the product actually has. This Actor splits the review pool into hundreds of small filtered views, each one under that ceiling, and collects them all.

1. Input parameters. Provide one or more ASINs, or any Amazon product or review URLs, up to 100 per run. Choose the marketplace, and optionally how complete you want the result, a source budget, or specific star ratings.
2. Collection. The Actor reads how many written reviews Amazon reports for each star rating, then collects across many narrow filtered views, so that no single view runs into Amazon's cap.
3. Deduplication. The same review appears in many views. Duplicates are removed before anything is written, and you are never charged for them.
4. Totals you can check. Alongside the reviews, the Actor writes one row per product giving Amazon's own written review count, how many of them you received, the split by star rating, and why collection stopped for each one.

Reviews are delivered to your dataset and can be downloaded as JSON, CSV, Excel, XML or HTML, or pulled straight from the Apify API.

This Actor needs a paid Apify plan. It buys review rows from a separate Store Actor, and that Actor does not serve free-plan accounts. On a free plan this run stops immediately and tells you so without charging you for any reviews.

Input data

Field	Required	Default	What it does
asins	Yes		Bare ASINs, or any Amazon product or reviews URL. Up to 100 per run.
marketplace	No	com	Which Amazon domain to read. Reviews differ per marketplace.
targetCoveragePercent	No	85	Stop collecting a star rating once this much of it is held.
maxSpendUsd	No	5	Hard cap on what the run spends buying rows, enforced in code before every call.
stars	No	all five	Collect only certain ratings, for a cheaper focused run.
collectionHints	No	none	Words you know appear in this product's reviews. Helps on products with distinctive vocabulary. Does not filter your results.
excludeReviewIds	No	none	Review IDs you already hold. Suppressed before anything is written, so you are not charged for them.
useSavedState	No	true	Remember what was delivered per product and send only new reviews next run.
maxCollectionRounds	No	40	Set to 0 for a faster, cheaper run that collects less.

A sample input:

{
  "asins": [
    "B0747FGGVC",
    "https://www.amazon.com/dp/B0BFSM1GHJ"
  ],
  "marketplace": "com",
  "targetCoveragePercent": 95,
  "maxSpendUsd": 5.0,
  "stars": ["five_star", "four_star", "three_star", "two_star", "one_star"],
  "collectionHints": ["battery", "accurate", "app"],
  "useSavedState": true
}

Output data

Two record types land in the same dataset, separated by record_type. Use the Reviews dataset view for the reviews alone and the Review totals view for the per-product counts.

A review record, taken from a live run:

{
  "record_type": "review",
  "review_id": "R2FY6UHDSQLZA7",
  "asin": "B0747FGGVC",
  "rating": 5,
  "star": "five_star",
  "title": "I like real time tracking.",
  "text": "I like real time tracking.",
  "date": "2026-05-12",
  "country": "United States",
  "verified": true,
  "vine": false,
  "helpful_votes": 1,
  "image_urls": [],
  "video_urls": [],
  "variant": null,
  "variant_asin": null,
  "review_url": "https://www.amazon.com/gp/customer-reviews/R2FY6UHDSQLZA7",
  "word_count": 10,
  "found_in_window": "five_star/recent"
}

Every field on a review record:

Field	Type	What it is
record_type	string	review or coverage_report.
review_id	string	Amazon's review ID, unique across the marketplace.
asin	string	ASIN the review was collected for.
rating	number	Star rating, 1 to 5.
star	string	Star filter the review belongs to, for example five_star.
title	string, null	Review title. Null when the reviewer left none.
text	string, null	Review body, exactly as Amazon serves it.
date	string, null	Review date, normalized to YYYY-MM-DD.
country	string, null	Country the review was written in.
verified	boolean	Verified purchase.
vine	boolean	Vine Voice review.
helpful_votes	integer	Helpful votes at the time of collection.
image_urls	array	Images attached to the review.
video_urls	array	Videos attached to the review.
variant	string, null	Product variant reviewed.
variant_asin	string, null	ASIN of that variant.
review_url	string	Permalink to the review.
word_count	integer	Words in title plus body.
found_in_window	string	Which Amazon filter combination surfaced it.

Reviewer names and profile links are never collected, in any field.

The totals record, one per product:

{
  "record_type": "coverage_report",
  "asin": "B0747FGGVC",
  "amazon_written_review_count": 750,
  "reviews_delivered": 701,
  "coverage_percent": 93.5,
  "per_star": [
    { "star": "five_star", "amazon_written_count": 519, "held": 473,
      "coverage_percent": 91.1, "missing": 46, "stopped_reason": "spend cap reached" },
    { "star": "four_star", "amazon_written_count": 107, "held": 104,
      "coverage_percent": 97.2, "missing": 3, "stopped_reason": "target reached" }
  ],
  "run_status": "stopped-short"
}

If a run reaches 71%, the report says 71%, names the ratings that fell short, and says why each one stopped. When your budget ends a run early it says so, rather than presenting a partial result as a complete one.

Supported Amazon domains

Seventeen marketplaces:

amazon.com, amazon.co.uk, amazon.de, amazon.fr, amazon.it, amazon.es, amazon.ca, amazon.com.mx, amazon.com.au, amazon.co.jp, amazon.in, amazon.nl, amazon.com.br, amazon.sg, amazon.ae, amazon.se, amazon.pl

Reviews differ per marketplace, so each one is a separate run.

How much does Amazon Reviews Scraper cost?

There are two charges, and you should know about both before your first run.

This Actor charges per unique review delivered, plus a flat fee for the per-product totals record. Duplicates are never charged, however many filtered views they turn up in. Current prices are on the pricing tab of this listing.

The underlying row fetcher is a separate Store Actor, billed to your own account at its own price. This is the larger of the two charges. The maxSpendUsd input caps it in code before every call, so a run cannot spend past the figure you set, and a run that reaches the cap stops cleanly and still reports everything it collected.

Estimating the total up front is difficult, because what a product costs depends on how its reviews are distributed rather than simply on how many it has. Start with the prefilled budget on one product, read your actual usage on the run detail page, and scale from there.

What to consider when using this Actor

1. Small products come back nearly complete. A few hundred reviews finishes in the nineties. A product with thousands levels off near 85%, and that is a property of Amazon, not of this Actor. The last slice is one-line reviews whose only words are the commonest ones, so they exist only inside views Amazon truncates. The report tells you this instead of hiding it.
2. Amazon's headline number is ratings, not reviews. A product showing 38,000 ratings may hold 12,000 written reviews. Only written reviews have text, and the percentages count only those.
3. Repeat runs are cheap. With useSavedState on, the Actor remembers what it already delivered per product and sends only new reviews next time, so you are not billed twice for the same row.
4. Set a budget you are comfortable with. maxSpendUsd is enforced in code before every call, not estimated afterward. Pair it with the run's own maximum cost setting to cap both halves.
5. No credentials, ever. This Actor sends no cookie, signs in to nothing, and never collects reviewer identity.
6. A paid Apify plan is required. The row source will not serve a free-plan account. If yours is free, the run stops on its first call, says exactly that, and charges you nothing for reviews.

Reliability and data handling

This Actor holds no credential. It sends no cookie, signs in to nothing, and stores no account of any kind. There is nothing here to share and nothing here to suspend. That is a deliberate limit rather than an oversight: it is also why the Actor cannot reach anything Amazon shows only to signed-in visitors.

No reviewer identity, ever. Display names, profile links and Amazon account IDs are dropped before anything is written. There is a test in the suite asserting it, so it cannot regress quietly. If you need reviewer identity, this is the wrong Actor.

Proxies, geo targeting and captchas are handled, by a third party we do not audit. Row fetching is done by a separate, long-established Store Actor that carries its own proxy pool and solves blocks and captchas under the hood, billed to your account at its own price. You never configure a proxy. How that supplier obtains its data inside its own infrastructure is not something we can verify for you, and we would rather say so than imply an audit we have not performed.

Review text is never altered. No escaping, no trimming, no normalizing of the body or title. What Amazon served is what lands in your dataset, which is the only honest thing to do with text you intend to analyze. See the spreadsheet note at the end for the one consequence of that.

The documented input matches the shipped input. A test asserts that every default in the input schema equals what the code actually uses, so the table above cannot drift away from the Actor.

Use cases

1. Ongoing review monitoring, the cheapest way to run this. Collect a product once, then schedule the same run on a timer with useSavedState left on. Reviews you already hold are never charged again and still count toward your coverage percent, so every run after the first bills only for what is genuinely new while the report keeps telling you where you stand. This is how to watch a catalogue without paying for it twice.
2. Brand review monitoring. Track everything customers write about your products, not the most recent hundred, and know what fraction of the record you are looking at.
3. Sentiment analysis with a stated denominator. Sentiment drawn from a few hundred reviews of a twelve thousand review product is a sample nobody measured. This gives you the sample size.
4. Competitor teardown. Pull the full written record on a rival product and find the complaints that repeat, including the ones buried past the first hundred.
5. Product quality and defect discovery. Low-star reviews are where failure modes live, and star filters let you buy only those for a fraction of a full run.
6. Market and academic research. Research that cites review data has to say how much of the record it actually read. The totals record is designed to be quoted.
7. Training and evaluation datasets. Large, deduplicated, text-bearing review corpora with verified-purchase and helpfulness signals attached.
8. Due diligence. Before acquiring a brand or a listing, read the whole review history rather than the page-one summary.
9. Review-led content. Comparison pages and buying guides built on the full record.

Opening the data in a spreadsheet

Review text is delivered exactly as Amazon serves it. A few reviews begin with =, +, - or @, which Excel, Google Sheets and LibreOffice read as the start of a formula.

Prefer the JSON or the API for anything automated, where no formula evaluation happens. If you need a spreadsheet, import the CSV rather than double-clicking it and set the title and text columns to Text: in Excel use Data, then From Text/CSV, then Transform; in Google Sheets untick "Convert text to numbers, dates, and formulas."

This Actor deliberately does not alter those characters, because prefixing them would store a character Amazon never wrote and corrupt the text for every consumer reading the JSON.