Amazon Product & Search Scraper

Scrape Amazon search results by keyword or search URL — ASIN, title, price, list price and discount, rating, exact review count, bestseller and Amazon's Choice badges, units bought last month, and image.

Built for price-distribution analysis and competitor catalogue monitoring.

No login, no API key. HTTP-only — no browser.

What you get

{
  "position": 1,
  "page": 1,
  "asin": "B01LP0U5X0",
  "title": "Amazon Basics Extra Thick Exercise Yoga Mat with Carrying Strap",
  "price": 24.18,
  "currency": "$",
  "listPrice": null,
  "discountPercent": null,
  "rating": 4.6,
  "reviewCount": 70384,
  "boughtInPastMonth": 10000,
  "isSponsored": false,
  "isBestSeller": true,
  "isAmazonsChoice": false,
  "imageUrl": "https://m.media-amazon.com/images/I/...",
  "productUrl": "https://www.amazon.com/dp/B01LP0U5X0"
}

Field coverage on a typical 48-result page:

Field	Coverage
asin, title, rating, reviewCount, imageUrl	48/48
boughtInPastMonth	45/48
price	44/48

Prices are missing on multi-variant tiles that show "See options" instead of a figure. That is a real state, not a parse failure, so it stays null.

Review counts are exact. The visible text on Amazon is abbreviated (70.3K); this actor reads the aria-label instead and reports 70384. A test asserts large counts are not suspiciously round.

Category URLs: read this first

A URL only works if it contains a k= keyword. Measured:

URL form	Results
/s?k=yoga+mat	48
/s?k=yoga+mat&rh=n%3A3407731 (category filter)	24
/s?k=yoga+mat&s=price-asc-rank (sorted)	48
/s?k=yoga+mat&rh=p_36%3A1000-3000 (price filter)	60
/s?rh=n%3A3407731 (node only)	0
/s?node=3407731, /b?node=…	0

Amazon serves node-only category URLs as a browse layout with no search results at all for logged-out clients. The actor rejects those up front with an explanation telling you to add a keyword, rather than running and quietly returning nothing.

So: any filter, sort or category refinement works — just keep a keyword in the URL.

Input

{
  "searchTerms": [
    "yoga mat",
    "https://www.amazon.com/s?k=coffee+maker&s=price-asc-rank"
  ],
  "marketplace": "com",
  "maxPages": 3,
  "maxItemsPerTerm": 0,
  "includeSponsored": true
}

Field	Description
searchTerms	Keywords, or Amazon search URLs containing k=
marketplace	com, co.uk, de, fr, it, es, ca, com.au, co.jp, in
maxPages	Up to 7 — Amazon stops offering a next page around there
maxItemsPerTerm	Hard cap per term; 0 for no cap
includeSponsored	Off keeps only organic results, for a cleaner price distribution
delaySeconds	Pacing — Amazon serves CAPTCHAs to traffic it dislikes
proxyConfiguration	Residential strongly recommended

ASINs are de-duplicated across pages, so a product Amazon repeats on page 2 is not double-counted in your price statistics. position is the global rank across pages; positionOnPage restarts each page.

Two fields that need a proxy in the right country

isPrime and deliveryInfo follow the exit IP, not the marketplace. From an Indonesian IP, amazon.com showed zero Prime badges across 48 results and rendered "Ships to Indonesia" with international delivery fees instead. Nothing was broken — Amazon simply does not offer Prime to that destination.

Every row carries shipsTo so you can tell. If it is not the marketplace's own country, treat isPrime: false as unknown, not as not Prime, and rerun with a residential proxy in that country.

Prices are pinned to the marketplace currency regardless of exit IP, via the cookie Amazon honours for this. Without it an Indonesian exit returns IDR 336,939.57 for a $19.99 item, which would silently corrupt price analysis. currency is on every row, so never assume USD.

About brandGuess

Amazon's current search results contain no brand element — no brand span, no data-brand attribute, no brand JSON. Verified across three queries. The a-size-base-plus class that older scrapers read for brand is the title element on today's markup.

So brandGuess is derived from the leading words of the title, and it is named that way on purpose. It gets Gaiam and BLACK+DECKER right, and it will get Amazon where the brand is really Amazon Basics. Use it for rough grouping; do not treat it as authoritative. There is no field here claiming to be a scraped brand, because there is no such data to scrape.

Known limits

• ~7 pages per term. Amazon stops serving a next-page link; the actor stops with it rather than burning requests on filler pages.
• No brand, for the reason above.
• Prime and delivery are geo-dependent — see above.
• Sponsored placements are auctioned per impression, so the sponsored subset of a page differs between runs. isSponsored reflects that impression.
• CAPTCHAs are reported as CAPTCHA_ESCALATION rows, never solved.

See CRAWLING_METHOD.md for the fingerprint matrix, the category-URL measurements and the parsing pitfalls.