Amazon product specs scraper: weight, dimensions, GTIN, model and materials

Give it a list of products. Get back the specification table Amazon prints, taken apart into named fields: weight, dimensions, model number, part number, GTIN, colour, material, country of origin, warranty, and the whole raw table alongside.

This is the boring half of a product catalogue and the half that decides what shipping costs. It is also the half Amazon is worst at presenting: the same fact lives under six different labels depending on which page variant you land on, and in at least three different tables.

Twenty-two fields per product, nineteen marketplaces, and no dependence on the buy box, so this Actor works on marketplaces where the price ones return nothing.

Why specifications are hard, and what this Actor does about it

Amazon has eight different detail tables. Which ones a page carries depends on the category, the seller, and which layout you happened to be served. Measured 2026-08-24: five products in a row carried the product overview table with Color and Power Source, and none of them carried the technical table with Item Weight and Manufacturer. A scraper that reads one table goes quiet whenever Amazon picks the other.

This one reads all eight and merges them, then reads the bullet-list layout as well, because on some pages the same facts are a list rather than a table.

Then it maps the labels. Amazon writes weight as "Item Weight", "Product Weight", "Weight", "Artikelgewicht" or "Poids" depending on where you are. It writes the part number as "Manufacturer Part Number", "Item Part Number" or "Mfr Part Number". Each named field in the output collects all of those spellings, so weight is filled whichever one your page used.

Getting that right is ongoing work rather than a solved problem. On 26 August 2026 three fields were still coming back empty on every product, and all three turned out to be our fault rather than Amazon's: manufacturer was being dropped on purpose by a filter that existed for a different reason, and partNumber and warranty were missing the exact labels Amazon uses. Fixed and measured the same day. The raw specifications object is in the output precisely so you can catch the next one yourself.

Every field it returns

Twenty-two per product: eleven named facts, the raw table they were pulled from, and enough identity to match the row against your own catalogue.

Identity and catalogue

Field	What it holds
asin	The product id
title	Full product title, shortened only past 1000 characters
brand	Brand name
manufacturer	Manufacturer, when different from brand
modelNumber	Model number
partNumber	Manufacturer part number
gtin	GTIN, EAN or UPC
marketplace	Which shop this row came from
url	The page this was read from
scrapedAt	Timestamp of the read

Specifications, pulled apart

Field	What it holds
specifications	Every row of every detail table, merged (this is what you are charged for)
productFacts	Amazon's newer facts panel
weight	Weight with its unit
dimensions	Package or product dimensions
colour	Colour
material	Material
unitCount	How many units in the pack
firstAvailable	Date first available on Amazon
countryOfOrigin	Country of origin
warranty	Warranty description
importantInformation	Safety and legal text, when present
isClimatePledgeFriendly	The Climate Pledge Friendly badge

What it charges for

This Actor charges on the specification table, not on the weight.

That is worth explaining, because the first version did charge on weight and it was the wrong unit. Measured on twelve products: a weight on five of them, a specification table on ten. So on half the work this Actor did, it delivered twenty facts about a product and billed nothing, while paying for the page anyway.

It is also the more honest unit. This Actor is called Product Specs and what it sells is the specification set. A product with twenty specifications and no weight is a complete answer, not a miss. A product with no specification table at all is the miss, and that is the row you do not pay for.

Measured across a sixty product run: fifty-five of sixty carried a specification table, so roughly one row in twelve arrived free.

A real row

asin              B0CP9YB3Q4
title             STANLEY Quencher H2.0 Tumbler with Handle and Straw 30 oz
brand             STANLEY
manufacturer      Stanley
modelNumber       10-13425-020
partNumber        10-13425-020
gtin              00041604414831
weight            1.1 pounds
dimensions        3.85 x 3.85 x 11.25 inches
colour            Rose Quartz
material          Stainless Steel
unitCount         1
countryOfOrigin   null
warranty          limited 5 year manufacturer warranty
firstAvailable    null
specifications    14 rows, including Capacity, Included Components,
                  Age Range Description, Item Type Name, Set Name
isClimatePledgeFriendly  false

Fourteen specification rows on that product, of which eleven map onto named fields. countryOfOrigin is empty because Amazon did not print it there, not because the read failed, and you can confirm that yourself by looking at the raw specifications object in the same row.

What is often empty, and why

Being straight about this is more useful than a longer field list.

countryOfOrigin, material and unitCount depend entirely on the category. Measured across eight varied products on 26 August 2026: material on six of eight, both of the others on one. Electronics rarely carry any of the three; kitchenware and clothing usually carry material.

importantInformation only appears on products with safety or legal text, which is mostly food, supplements, batteries and anything with a plug.

firstAvailable and warranty show up on maybe one product in five.

gtin is there when the seller supplied it, which for brand-name products is often and for white-label products is rarely.

None of these emptinesses cost you anything: the charge is on the specification table, and a row with a table and no country of origin is a charged row that told you what Amazon knows.

What to put in

ASINs or product URLs. Paste them in any shape: a bare B08N5WRWNW, a full URL with tracking parameters, a spreadsheet cell that still says ASIN: B08N5WRWNW. Duplicates are removed across everything you give it.

A category or bestseller link, to pull specifications for a whole category at once.

A search term, when the catalogue you want is defined by what shoppers search for.

Settings worth knowing about

Setting	What it does
marketplace	One of nineteen. Labels are read in the shop's own language, so a German page filling "Artikelgewicht" still fills weight.
proxyCountry	Leave empty and it follows the marketplace.
maxItems	Hard ceiling on delivered products. Zero means no ceiling.
maxItemsPerList	How many products to take from each search term or category link.
maxListPages	How deep to walk each list.
lanes	How many products are read at once. Eight by default. Specifications do not come from the buy box, so pushing to sixteen costs less here than on the price Actors.
maxAttempts	Attempts per product, each from a different address. Twelve by default.
residentialFallback	Off by default. Measured 2026-08-26: on it cost roughly six times as much and returned two more products out of sixty.
economical	Leave it off. The detail tables sit at the bottom of the page, so hanging up early is exactly how you lose the field you came for.

How it works, in plain terms

Amazon refuses roughly six requests in ten to an address it has not seen before. Every product here gets up to twelve attempts, each from a different exit address, and the same address is never used twice in a row. Addresses that work are remembered for the rest of the run and tried first. The pace adapts: when refusals climb the runner slows down instead of pushing harder.

Anything still refused at the end goes into a second pass at half speed, using only the addresses that already proved themselves.

The retry rule follows the specification table rather than the price. That distinction cost real money to learn: an earlier version judged every Actor on whether a price had arrived, so this one burned twelve attempts on pages that had already handed over twenty specifications, and then charged for none of them.

Reading the run log

Done: 15 charged (+5 delivered without a price, free), 0 refused by Amazon, 0 other. 359 KB per charged result, 69 attempts in total.
Exit addresses: 39 used, 4 proven good, 10 burned.
Attempts by outcome: {"refused by Amazon":44,"ok":20,"http_500":3,"proxy error":1,"network error":1}

charged is products that came back with a specification table. The bracketed number arrived complete without one and cost you nothing.

If a run ever delivers rows and charges for none of them, the log says so in capitals at the bottom.

Reliability, measured

Three consecutive runs of twelve products on 26 August 2026, at the default settings:

run 1   12 of 12 delivered, 12 with a specification table
run 2   12 of 12 delivered, 12 with a specification table
run 3   12 of 12 delivered, 12 with a specification table

And one run of sixty products the same day: 60 delivered, 55 with a specification table, in 6.3 minutes.

Worth reading out of that: nothing was lost at either size. 36 of 36 on the short runs and 60 of 60 on the long one. That is the shape you want before you schedule something, because a scraper that is reliable only on small batches is not reliable, it is lucky.

The second is the residential switch, and the honest version is less dramatic than it sounds. Measured 26 August on the same sixty products, twice: with the switch off, fifty-one came back with the field this Actor charges on, for $0.0205 of platform usage. With it on, fifty-three came back, for $0.1192. Two more products for roughly six times the cost.

That is why it is off by default. It is in your input form, so if a marketplace is refusing almost everything you can turn it on and pay for the difference.

What you can build with it

A shipping cost model. weight and dimensions across your catalogue, which is what every freight quote actually asks for.

A catalogue match. gtin, partNumber and modelNumber are how you line an Amazon listing up against your own product records or a supplier's.

A compliance sweep. countryOfOrigin and importantInformation where you need them for customs or labelling.

A sourcing brief. material, unitCount and dimensions are what you hand a manufacturer when you want the same thing made.

A listing quality audit. Count how many specification rows your own listings carry against the category leaders. A thin table is a conversion problem you can fix in an afternoon.

Working it into your stack

Every run writes to a dataset you can download as CSV, JSON, Excel or XML, or pull through the API.

One shape note: specifications and productFacts are objects with whatever keys Amazon used, so a strict flat export will either drop them or stringify them. The eleven named fields exist so you do not have to deal with that for the common cases; keep the raw object when you need the rest.

Schedules take a cron expression. Specifications change rarely, so weekly or monthly is usually enough, which makes this one of the cheapest Actors here to keep running.

Marketplaces

Nineteen, from amazon.com to amazon.co.jp, and labels are matched in the shop's own language rather than translated afterwards.

Because none of this comes from the buy box, this Actor works where the price ones do not. Measured on amazon.de, where no product in the run returned a price: twenty specifications per product came back on eight of eight. If you are cataloguing a European marketplace, this is the Actor in the family that will actually give you something.

Pricing

Four dollars per thousand products with a specification table, plus three cents to start a run.

Cataloguing a hundred products costs forty-three cents: a hundred times four tenths of a cent, plus the three cent start fee. You are not charged for a product with no specification table, or for one that never arrived, and platform usage is included.

The rest of this family

Eleven Actors on one engine. Around a specification scraper the useful ones are:

• Amazon Variants Scraper when a listing turns out to be fourteen products and you need the specification set for each.
• Amazon Product Scraper when you want all sixty-four fields, including price, rank and images, rather than the catalogue half.
• Amazon Bestsellers Scraper to build the product list before you catalogue it.
• Amazon BSR Tracker which, like this one, works on marketplaces where the buy box does not arrive.

Questions people ask

Why is weight sometimes empty when specifications is not. Because Amazon did not print a weight on that page. Look in the raw specifications object in the same row and you will see exactly what it did print.

Are the units normalised. No. weight comes back as Amazon wrote it, "1.1 pounds" or "500 g". Converting would mean guessing at ambiguous cases, and the raw string is auditable in a way a converted number is not.

Can I get specifications for every variant of a product. Feed the sibling ASINs in as input. The variants Actor in this family returns that list.

Why is manufacturer sometimes the same as brand. Because Amazon printed the same value in both rows. It differs on products made under licence or by a contract manufacturer, which is exactly when you care.

How often should I run it. Specifications rarely change. Once when you add a product, then monthly if you want to catch relabelling.

How many products can it handle. No built-in ceiling. Sixty products took 6.3 minutes on the measured run, so a hundred is around 10.

What it does not collect

No review text, no reviewer names, no customer questions, no seller names. Those are personal data, and scraping them out of a public website is almost always a GDPR problem in Europe regardless of how public the page is.

Legal note

This Actor reads public pages and takes product facts: weights, dimensions, identifiers, materials. Business information about an item, not information about a person.

You are responsible for what you do with the output. Check Amazon's terms for your own use case.