Amazon Product Scraper - Prices, ASIN & Demand Signal

Product research and price monitoring on Amazon without the Product Advertising API. No seller account, no API credentials, no approval process — point it at a search and get typed rows back.

Two things make this one different. The price always carries an explicit currency, and any row that came back in the wrong currency is flagged rather than quietly mixed in. And every row includes Amazon's bought in past month figure — the closest thing to a public demand signal that Amazon exposes.

Common uses: sizing a category before sourcing, monitoring competitor prices and discounts, tracking demand on a shortlist of ASINs, and feeding product data into a dashboard.

What you get per product

Field	Example	Notes
asin	B0GP99ZYSH	Amazon's own id, use it to deduplicate across runs
title	Rambler 20 oz Stainless Steel Vacuum Insulated Tumbler
brand	YETI	Separated from the title
price / currency	24.99 / USD	Number plus the currency it was quoted in
priceRaw	$24.99	Original string, kept for auditing
listPrice / discountPercent	34.99 / 29	When Amazon shows a struck-through price
currencyMismatch	false	True means the request was geolocated elsewhere — see below
rating	4.8
reviewCount	148600	148.6K expanded to a real number
boughtInPastMonth	2000	From 2K+ bought in past month
badge	Overall Pick	Also Best Seller, Amazon's Choice
isPrime, isSponsored	true / false
variants	3 sizes
deliveryText	FREE delivery Wed, Aug 19
imageUrl, productUrl
searchKeyword, pageNumber, amazonDomain, proxyCountry, scrapedAt		Provenance of every row

The currency problem, and what this actor does about it

Amazon prices by geography. Measured: requesting amazon.com from outside the United States returns prices like VND 919,222 with local delivery estimates — not USD. A scraper that hands that back as a plain number silently corrupts every comparison built on top of it.

This actor:

1. Pins the marketplace's own currency with a cookie before the first request.
2. Matches the proxy country to the marketplace you selected.
3. Derives the expected currency from amazonDomain and sets currencyMismatch: true on any row that still disagrees, plus a warning in the run log and a count in the run diagnostics.

So a bad run is visible instead of quietly wrong.

Input

{
  "keywords": ["yeti rambler 20 oz tumbler", "stanley quencher"],
  "amazonDomain": "www.amazon.com",
  "proxyCountry": "US",
  "sortBy": "relevance",
  "maxProducts": 500
}

Filters: marketplace, sort order, price range, Prime only, and optionally dropping sponsored placements. Sponsored rows are kept by default because ad presence is useful competitive data — every row carries isSponsored either way.

The result cap is real

maxProducts stops the run at exactly that number, and duplicates are dropped on ASIN, so the same product is never delivered — or billed — twice. The run also stops on its own if three consecutive pages return nothing usable.

Runs fail loudly, not silently

Every run writes a DIAGNOSTICS record with block rate, per-field fill rates and the currency-mismatch count:

{ "itemsExtracted": 500, "blockRate": "0%", "currencyMismatchRows": 0,
  "fieldCoverage": { "price": "97%", "boughtInPastMonth": "41%" } }

If nothing was extracted, or more than half the pages were blocked, the run is marked failed with the reason.

Typical uses

• Price and discount monitoring across a category
• Product research ranked by the bought-in-past-month signal rather than guesswork
• Brand share of shelf: who owns the first pages for a keyword, and how much is paid
• Feeding a repricing or catalogue pipeline with typed rows

Notes and limits

• boughtInPastMonth and badge only appear where Amazon shows them, and are null elsewhere rather than estimated.
• This actor reads search results. Full review text lives behind a login on Amazon, so it is deliberately out of scope.
• Keep maxConcurrency low; Amazon rate limits per session.

Support

If a field stops filling or you need a filter that is not exposed, open an issue on the actor page with the input you used.

What isPrime and isSponsored really tell you

isSponsored is detected from the card's markup, not from its text. Amazon marks ad cards with an AdHolder class and a sponsored-label element; both were measured agreeing on the same cards. The older text-only check silently returned zero sponsored rows on pages that actually carried 24 of them, because this actor blocks stylesheets to save bandwidth and that changes what innerText contains. Every run now logs sponsoredSignals so you can tell a genuinely ad-free page from a broken selector.

isPrime is usually false, and that is honest rather than missing. Measured on amazon.com/s: zero result cards carried any Prime marker at all. Amazon only paints that badge for some sessions and locales, so treat false as "no badge shown to this request" rather than "not a Prime item". Runs report primeBadgesSeen for the same reason.

FAQ

Do I need an Amazon API key or a seller account? No. This reads Amazon's public search pages. There is nothing to register for and no approval process — you only need an Apify account.

Why would a price come back in the wrong currency? Amazon prices by IP country. Requested from a Vietnamese IP, amazon.com returns figures like VND 919,222 with local delivery instead of USD. This actor pins the marketplace currency with the i18n-prefs cookie, matches the proxy country to the marketplace, and then verifies each row — anything that still comes back mismatched is flagged with currencyMismatch: true instead of being averaged in with real USD prices.

Is "bought in past month" reliable? It is Amazon's own figure, copied as shown and parsed to a number (2K+ becomes 2000). It appears on most but not all listings — roughly three quarters in the runs measured — and is null where Amazon does not display it, never estimated.

Can it scrape reviews or product detail pages? No. This actor covers search results only. Amazon redirects /product-reviews/ to a sign-in page, so a review scraper would need an authenticated session and is a different product.

How do I stop it from running up a bill? maxProducts is a hard stop. The run also fails loudly on zero items or a block rate above 50%, so you are never charged for a dataset that quietly came back half empty.