Amazon Reviews Scraper

Scrape Amazon product review previews from public product pages and export Amazon review data to JSON, CSV, or a spreadsheet. This Amazon reviews API alternative accepts product links or ASINs and returns one flat record per visible review, including rating, title, text, date, purchase verification, variant, images, and helpful votes.

The actor is designed for comparing feedback across many products. Amazon currently redirects anonymous visitors from the dedicated review pages to sign-in, so this actor reads the review cards that Amazon places on each public product page. That distinction matters: it can collect the visible preview for each product, but it cannot promise every review ever written for one ASIN. In the verified run, four products yielded 37 reviews, with eight to thirteen review cards per product.

What data can you extract from Amazon?

Each dataset item is a review. The fields stay flat so a row can be used directly in a table, export, or downstream analysis. Nullable fields may be empty when Amazon does not show the corresponding detail on a review card.

Field	Description	Example
productAsin	ASIN of the requested product	B0CWXNS552
productUrl	Canonical product page used for extraction	https://www.amazon.com/dp/B0CWXNS552
reviewId	Public identifier of the review	R3NGMNN8H0IJ52
ratingScore	Rating on the five-star scale	5
reviewTitle	Headline displayed on the card	Excellent to find my things
reviewUrl	Link to the review	https://www.amazon.com/gp/customer-reviews/R3NGMNN8H0IJ52?ASIN=B0CWXNS552
reviewReaction	Displayed helpful-vote wording	4 people found this helpful
reviewedIn	Full displayed place and date line	Reviewed in the United States on August 13, 2026
reviewDate	Date parsed into YYYY-MM-DD	2026-08-13
reviewDescription	Text of the review card	The location is quite accurate...
isVerified	Whether a verified-purchase badge is present	true
variant	Variant label, when shown	Style: Single
reviewImages	Images attached to this specific review	An array of image URLs
helpfulVotes	Helpful-vote count parsed from the displayed wording	4
position	Order of the card in the product page preview	1
reviewerName	Public display name, only when explicitly requested	A display name or null

The image list is tied to the review card. A product's general gallery is not copied into every review record. A missing helpful-vote sentence gives helpfulVotes a value of zero; reviewReaction stays empty, so you can distinguish a displayed zero from no displayed sentence. The full reviewedIn line is preserved alongside the normalized reviewDate because the wording can include geographic context that a date alone does not convey. Review IDs allow deduplication when you combine exports from different runs.

How to scrape Amazon reviews

1. Copy one or more Amazon.com product URLs. A bare ten-character ASIN also works. Product pages, standard /dp/ links, and recognized review URL paths are accepted as input.
2. Paste them into Product URLs or ASINs. The default example uses one product, but you can add a list to compare several products in one run.
3. Set a per-product limit if you need fewer records. Add rating, keyword, or date filters when you want to narrow the visible preview. Leave the filters at their defaults to capture all visible review cards.
4. Run the actor. It first tries a low-cost datacenter request, then uses a US residential connection when Amazon blocks that request.
5. Open the dataset and download JSON, CSV, or spreadsheet output. Each row represents one review, so it can be joined to a product catalog by productAsin.

For a first run, start with several products that have substantial review activity. A single product normally contributes only the short public preview. The maxReviews setting is a ceiling on extracted reviews, not a request for Amazon to expose more cards.

Input

Name	Type	Default	Description
productUrls	List of URLs or ASINs	One sample URL	Products to inspect. This field is required.
maxReviews	Integer	100	Maximum matching reviews saved per product. Zero removes this cap, but does not extend the public preview.
maxItems	Integer	0	Total cap across all products. Zero means no overall cap.
sort	helpful or recent	helpful	Keep page order or arrange visible cards by review date.
filterByRatings	List of rating groups	allStars	Keep selected stars or the positive and critical groups.
reviewsFilterByKeywords	List of strings	Empty	Keep a visible card if its title or text contains any keyword.
reviewsCutoffDate	Date string	Empty	Keep cards dated on or after a YYYY-MM-DD date.
includeGdprSensitive	Boolean	false	Include the public reviewer display name when explicitly enabled.
proxyConfiguration	Proxy settings	Platform proxy	Override proxy settings if your account needs a specific configuration.

A realistic input for comparing four products is:

{
  "productUrls": [
    { "url": "https://www.amazon.com/dp/B0CWXNS552" },
    { "url": "https://www.amazon.com/dp/B08BHHSB6M" },
    { "url": "https://www.amazon.com/dp/B0D54JZTHY" },
    { "url": "https://www.amazon.com/dp/B0GJTFXNRX" }
  ],
  "maxReviews": 100,
  "maxItems": 0,
  "sort": "helpful",
  "filterByRatings": ["allStars"],
  "reviewsFilterByKeywords": [],
  "includeGdprSensitive": false
}

The verified run used these four URLs and returned 37 items. The result count can change as Amazon updates product pages or the preview selection. URLs that resolve to the same ASIN are deduplicated within the input list. Invalid URL formats cause an input error before extraction begins. A product that fails to load is reported in run logs; records from other products can still be saved.

Rating filters operate on the cards returned by each product page. fiveStar means five-star cards, while positive covers four and five stars and critical covers one through three stars. If groups overlap, an individual review still appears once. Keyword matching ignores letter case and checks title plus text. Multiple keywords use OR logic: a match on any one keyword keeps the card. The cutoff applies to the parsed date. A card with no parsable date cannot satisfy an active cutoff.

The sorting choice also applies after collection. helpful preserves the page's order; recent sorts the visible cards by date. Neither option changes which cards Amazon chooses for the product page preview. For a historical or statistically complete review collection, treat this actor's output as a sample of public page content rather than the full review archive.

Output example

This shortened item is from the verified 37-item run. The review body and image URL are shortened here for readability; the dataset stores the complete values returned by the page.

{
  "productAsin": "B0CWXNS552",
  "productUrl": "https://www.amazon.com/dp/B0CWXNS552",
  "reviewId": "R3NGMNN8H0IJ52",
  "ratingScore": 5,
  "reviewTitle": "Excellent to find my things",
  "reviewUrl": "https://www.amazon.com/gp/customer-reviews/R3NGMNN8H0IJ52?ASIN=B0CWXNS552",
  "reviewReaction": "4 people found this helpful",
  "reviewedIn": "Reviewed in the United States on August 13, 2026",
  "reviewDate": "2026-08-13",
  "reviewDescription": "The location is quite accurate and gives me a lot of peace of mind...",
  "isVerified": true,
  "variant": "Style: Single",
  "reviewImages": ["https://m.media-amazon.com/images/I/51TStuZBJaL._SY500_.jpg"],
  "helpfulVotes": 4,
  "position": 1
}

One item represents one review, rather than a product summary. There is no synthetic average rating in the output: averaging a small page preview would suggest broader coverage than the actor actually has. When you need a product-level summary for your own workflow, group items by productAsin and label any computed measure as a summary of the visible cards.

Use cases

• Product research: compare recurring praise and complaints across a shortlist of products. Review titles, text, rating, and date appear together in each row.
• Quality monitoring: schedule repeat runs for the same ASINs and compare review IDs over time. A new ID indicates a newly visible card, although a card can also move into or out of the preview without being new to the product.
• Merchandising: examine the review language associated with several candidate products before changing catalog descriptions. Keep the sampled nature of the data in mind.
• Variant checks: inspect the variant field to see whether a card mentions a particular size, style, or other variation. The field may be absent and is not a guarantee that the page shows all variants.
• Customer support research: find visible reviews that contain specific terms, then follow the review URL for source context. Keyword filtering can reduce the exported rows but does not search all reviews on Amazon.
• Media review collection: identify review cards with attached images and retrieve their image URLs without confusing them with the main product gallery.

How much does it cost to scrape Amazon reviews?

The configured result charge is $0.001 per saved review, or $1 per 1,000 results. The charge follows the number of dataset items saved, not the number of product URLs entered. The platform's standard plan and usage terms still apply to your account.

At that rate, the 37-item verified run corresponds to $0.037 in result charges. Saving 500 reviews across many products corresponds to $0.50. Saving 5,000 reviews corresponds to $5.00. These calculations describe result charges; actual item counts depend on how many visible cards match your filters and how many products are supplied. A product with no matching public preview contributes no review items.

A small per-product preview changes planning. For example, requesting 100 reviews from each of ten products does not ensure 1,000 items. If each page exposes eight cards and all pass the filters, the run yields about 80 items. Estimate your result count from observed cards per product, then choose enough products to cover the comparison you want to make.

Tips for best results

Start with valid Amazon.com product pages. A canonical /dp/ASIN URL is easy to audit, and a bare ASIN is accepted when you already maintain a product catalog. The actor normalizes recognized paths to /dp/ASIN before fetching the page. It does not fetch arbitrary URLs outside the supported domain.

Use several products when you need more than a handful of records. The public product page preview is the binding limit, and the verified products showed eight to thirteen cards each. Changing maxReviews above that count does not unlock additional pages. If you only need a few cards per product, lower maxReviews to shorten your dataset.

Apply filters carefully. A narrow rating group, rare keyword, or recent cutoff may leave zero matching cards even when a product has many reviews overall. That means the visible preview lacks a match; it does not prove that no matching review exists elsewhere. For discovery, begin with allStars and no keyword or date cutoff, inspect the records, and then refine the input.

Leave reviewer names disabled unless you have a legitimate need and an appropriate basis to process them. Ratings, titles, text, dates, and review IDs remain available with the default setting. The optional display name comes from the same public card and may differ from a person's legal name.

For repeat monitoring, retain reviewId and productAsin as a compound key. Review position can change when the page preview changes, so it is unsuitable as a stable identifier. The reviewDate gives a normalized date for filtering; the reviewedIn line preserves the exact displayed context. Review text can be edited or removed on the source site, so preserve extraction timestamps in your own workflow if historical changes matter.

Limits

Amazon redirects anonymous visitors from dedicated product review and individual review routes to sign-in in the tested environment. This actor does not log in. It reads the public cards embedded in product pages. The tested pages yielded eight to thirteen full cards each; another product can expose fewer or none. It cannot guarantee an exact number for a given ASIN or reach the full history behind the sign-in wall.

The actor attempts a datacenter connection first. Amazon blocked that path during development, so a US residential connection supplied the successful pages. Amazon can vary its defenses by time, network, or product. A blocked product is logged and skipped after retries. Large runs may use more proxy traffic than a simple page count suggests because product pages are substantial.

Some cards omit a variant, image, helpful-vote sentence, or full text. reviewImages may be an empty array. reviewReaction and variant can be null. A missing helpful-vote sentence is represented as zero in helpfulVotes, which should not be interpreted as a confirmed displayed zero. Dates are parsed from the English date line used on the tested US pages; a changed date format may leave reviewDate empty while reviewedIn retains the original wording.

Filters and sorting run on visible cards. They do not query Amazon's complete review index. The actor accepts Amazon.com ASINs and URLs only; it does not claim equivalent coverage for other country stores. Product availability, removed pages, redirects, and layout changes can reduce results. If zero items are returned, check the run log and retry with an unfiltered input for a currently accessible product.

Use with automated clients

The input and output schemas give automated clients clear field names and limits. A client can submit several product URLs, wait for the run, and consume the default dataset without trying to interpret a nested product object.

Example request: “Collect the visible public reviews for these four Amazon product URLs, then group the saved rows by ASIN and show the most common topics in one-star through three-star cards.” Another request: “Run the same product list weekly, compare review IDs to the previous dataset, and show newly visible cards with their rating and review date.” Both examples use the public preview and should be described as such in any resulting report.

FAQ

Is scraping Amazon reviews legal?

Public review information can be used for legitimate research, but you are responsible for complying with applicable law, the site's terms, and privacy rules. Avoid collecting or redistributing personal information without a proper basis. Reviewer names are disabled by default. Review text also belongs to its source context; keep links and attribution when you use excerpts publicly.

Do I need to configure proxies?

The actor attempts the platform proxy automatically. It starts with a datacenter request and uses a US residential connection when the site blocks it. The tested golden run succeeded through that path. Advanced proxy settings are available in the input if your account requires them.

How fast is a run?

Time depends on the number of products, page size, proxy response, and blocks. The verified four-product run completed in about 35 seconds after build. A run that needs retries or a browser fallback can take longer. The actor processes product pages sequentially so each item stays associated with its ASIN and the run can stop at the total item cap.

Can I get every review for one product?

No. Anonymous dedicated review pages redirected to sign-in in testing. This actor captures the full review cards visible on the public product page, which was eight to thirteen cards for the verified products. Increasing the requested maximum does not change that source limit.

Can I sort by newest or filter by rating?

Yes, within the visible preview. The recent setting rearranges cards with parsed dates; rating groups keep cards whose displayed stars match. The options do not request hidden pages or change the source's card selection.

Can I schedule monitoring and export to a spreadsheet?

Yes. Schedule repeat runs on the platform, then export the dataset as JSON, CSV, or a spreadsheet. Preserve reviewId when comparing successive exports. A change in preview membership is not necessarily a newly posted or deleted review.

Why is a field empty or why did I get no items?

Amazon may not display a field on a particular card. Filters can also remove all visible cards. If the run has no items, check the logs for blocked or unavailable products, then test without filters. When a field is absent from the page, the actor leaves its nullable output field empty rather than inventing a value.

Integrations

The default dataset can be fetched through the platform API and connected to scheduled workflows or webhooks. Export rows to a spreadsheet, send them to a data warehouse, or feed them into a catalog comparison process. Use productAsin to join review rows to a product list and reviewId to avoid duplicates between runs. Downstream dashboards should label counts and averages as measurements of the public preview, not of every review on the product.

Support

Open an issue on the Issues tab with a product URL, run ID, and the behavior you expected. We reply within 24 hours and add fields on request when the public page supplies them.