Product Matching in E-commerce: What It Is and How AI Does It (2026)

Last updated: 09 Nov 2026

Product Matching in E-commerce: What It Is and How AI Does It (2026)

Key takeaways

  • Product matching links the same product across your catalog, supplier files and competitor sites when titles, SKUs and pack sizes are written differently.
  • No shop platform maps CSV titles and SKUs for you: Shopify, WooCommerce and Adobe Commerce only update a row when the handle, ID or SKU is exactly the same.
  • Match before you import: EAN or GTIN first, then cleaned titles and attributes, then a short review of the unsure pairs.
  • Keep pack sizes and variants apart. A 2 kg tub is not a match for a 500 g bag, even when the titles look the same.

Product matching is the process of finding and linking the same product across different data sources, such as supplier catalogs, marketplaces, your own shop and competitor sites, even when titles, SKUs or images differ. In e-commerce it is what makes like-for-like price comparison, clean catalogs and correct stock counts possible. Most teams match on identifiers (EAN/GTIN) first and use text, attribute and image matching for the rest.

Short answer for CSV files: if you have a CSV of your own titles and SKUs and want to map it to another list (a supplier feed, a marketplace export or a competitor site), no ecommerce platform does the mapping for you. Clean the titles first, match on EAN or GTIN where both sides have one, match the rest on title and attributes with a confidence score, and check the unsure pairs by hand before you import. Product matching software runs those steps for you and hands back one row per pair.

Here is what that looks like for one SKU from a CSV file against four competitor listings (example data):

Your CSV row Competitor listing After cleaning Result
WHEY-VAN-500G, "Whey Protein Vanilla 500g" "Whey Protein Isolat Vanille 500g Beutel" brand, vanilla, 500 g Match
WHEY-VAN-500G "Vanilla Whey 0,5 kg Pouch" brand, vanilla, 500 g (0,5 kg = 500 g) Match
WHEY-VAN-500G "Whey Protein Vanilla 2 kg Dose" brand, vanilla, 2 kg Not a match: different pack size, kept as a variant
WHEY-VAN-500G "Vanilla Whey Protein" brand, vanilla, no size Unsure: sent to review

Exact matching on the title would find none of these. Fuzzy matching alone would wrongly accept the 2 kg tub. The cleaning step (units, pack size, brand) is what makes the difference. See the product matching use case for the full row format with confidence scores, and price per unit across pack sizes for the cleaning side.

We use this same flow for bebo.ee, an Estonian grocery price comparison built on Scrapewise: scrape, clean and match across the big Estonian grocery chains, so the same product can be compared from shop to shop.

What Is Product Matching in E-commerce?

Product matching in e-commerce means deciding which listings in two or more product lists are the same item. It compares identifiers, titles, attributes and images, gives each candidate pair a confidence score, and returns linked pairs. The output is a mapping: your SKU next to the matching supplier SKU, marketplace ID or competitor URL.

The same product shows up as "Blue Sofa XL" on a marketplace, "XL Azure Couch" on your own site and "Large Blue Sectional" in your ERP. People see at a glance that these are one sofa. A spreadsheet doesn't. A VLOOKUP on the title finds nothing, and a VLOOKUP on the SKU finds nothing either, because every system has its own SKU.

There are three common jobs:

  • Catalog matching. Your own catalog against supplier feeds, so new products don't get created twice.
  • Channel matching. The same product across your shop, marketplaces and your ERP, so stock and prices line up.
  • Competitor product matching. Your SKUs against competitor listings, so you compare prices on the same item.

If you want the matched rows rather than the method, product matching covers what a delivered row looks like, and competitor price tracking puts the matched prices side by side.

How Does a Product Matching Algorithm Work?

A product matching algorithm checks the strongest signal first and falls back to weaker ones. It matches on a shared identifier such as EAN or GTIN, then compares cleaned titles, brand and model, then attributes like size and colour. Each signal adds to a confidence score, and that score decides whether a pair is accepted, reviewed or rejected.

Signal What it compares How strong Where it goes wrong
Identifier EAN, UPC, GTIN, or brand + manufacturer part number Strongest when both sides carry it Missing or wrong codes, regional codes, leading zeros lost in Excel
Title text Cleaned, tokenised titles; fuzzy string similarity Medium Marketing words, other languages, near-identical titles for different sizes
Attributes Brand, model, size, weight, colour, pack count Strong when complete Units written differently ("0,5 kg" vs "500g"), missing fields
Image Visual similarity of product photos Useful as a tie-breaker Same photo used for several variants, lifestyle shots
Confidence score All of the above, weighted Decides the outcome Thresholds set too loose let wrong pairs through

GS1, which issues the codes, describes the GTIN as a number that gives products and packaging a globally unique identity, and it can identify different packaging levels, from a single item to a case or a pallet. That is why identifiers come first. It is also why a multipack can carry a different code from the single item inside it. The ways identifier matching breaks are covered in why EAN matching fails.

Text matching

Titles are lowercased, marketing words are removed, and units are written one way ("64 Gigabyte" becomes "64GB"). The title is split into tokens, and fuzzy string matching handles typos and word order. Synonyms ("couch" and "sofa") and other languages need either a synonym list or a language model.

Attribute matching

Brand, model number, size, weight, colour and pack count are pulled out of the title or the product page and compared field by field. Unit conversion matters here. A rule like "sizes must be equal after conversion" is what keeps a 2 kg tub away from a 500 g bag.

Image matching

Computer vision compares product photos. It helps when titles are thin, but it is risky on its own because shops often reuse one photo for every colour or size. Treat it as confirmation, not proof.

Confidence Scoring and Review

Each candidate pair gets a score. Pairs above a high threshold are accepted, pairs in the middle go to a person, and pairs below a low threshold are dropped. The thresholds are a business choice: for pricing decisions, a missed match is cheaper than a wrong one, so set the auto-accept bar high.

What Is AI Product Matching, and How Is It Different From Rules?

AI product matching uses language models and image models to decide whether two listings are the same product, instead of only fixed rules like "titles must share 80% of words". It copes better with synonyms, other languages and messy titles. It still needs clean units, identifiers where they exist and a human review of unsure pairs.

Rules-based matching AI product matching
How it decides Exact keys, fuzzy thresholds, synonym lists Models trained on titles, attributes and images, plus rules
Synonyms and languages Only what is in your lists Understands meaning, for example "couch" and "sofa", "Vanille" and "vanilla"
Setup Fast for one clean source More setup, but scales better across many sources
Upkeep Grows with every new shop and language Learns from review decisions
Typical weak spot Misses matches with different wording Can accept look-alike variants if size and pack rules are missing

Search vendors describe the same shift on the shopper side. Bloomreach says AI search reads what the shopper means rather than matching the letters they typed. Matching two product lists is the same problem, applied to listings instead of search queries.

In practice the best setups mix both. Use hard rules where they are safe (same EAN, different pack size means no match) and models for the fuzzy middle.

What Is Competitor Product Matching?

Competitor product matching links each of your SKUs to the same product on competitor sites, so a price comparison is like for like. Competitor pages rarely show your SKU and often hide the EAN, so matching leans on brand, model, size and pack count. Without it, a price gap can just be two different products.

Price comparison sites such as Prisjakt, PriceRunner and idealo show many shops' offers on one product page, and that only works once the listings are matched. See Prisjakt and PriceRunner for Nordic price monitoring for what those sites can and can't tell you.

Pack sizes matter even more in Europe. Under the EU Price Indication Directive, shops must show both the selling price and the unit price, meaning the price per kilogram, litre or metre. If your matching treats a 500 g bag and a 2 kg tub as the same item, your unit price comparison is wrong from the start.

The pricing side is covered in product matching for price monitoring.

What Are the Signs Your Catalog Needs Better Matching?

You need better product matching when the same product appears more than once, stock counts don't agree between systems, competitor price gaps look too big to be true, or your team spends hours each week pairing products by hand in Excel. Each of these traces back to listings that were never linked.

  1. Duplicate listings. The same item shows up two or three times in search, with slightly different titles and prices. Shoppers don't know which one is real, and your own listings compete with each other.
  2. SKU conflicts. Your ERP, your marketplace account, your supplier and your warehouse each use their own code. Without a mapping, stock doesn't reconcile and orders go wrong.
  3. Inconsistent product information. The same product has different prices, specs or photos on different channels.
  4. Returns from wrong expectations. Colours, sizes, bundles or regional versions get mixed, so customers receive something other than what the page showed.
  5. Manual matching in spreadsheets. In my experience, this is the one that hurts most. Excel files go back and forth by email, several people add their part, and days pass before anyone can act on the prices.

What Are Advisory Tools for Product Matching?

Advisory matching tools suggest a match and let a person decide, instead of applying it silently. Each candidate pair comes with a confidence score and the reason for it, such as a shared EAN, a similar title or matching attributes. A reviewer approves, rejects or corrects the pair, and wrong matches stay out of your pricing data.

Price intelligence tools with a matching module are the usual place to find this. Price2Spy, for example, has offered matching suggestions you approve or reject, one by one or in bulk, since 2017.

When you compare providers, check for:

  • Explanations that show which signals (identifier, title, image, attributes) drove each suggested match
  • Thresholds you can adjust, so you decide what is approved automatically and what goes to review
  • A review queue with bulk approve and reject for unclear pairs
  • Learning from review, where reviewer decisions retrain or re-weight the matcher
  • Quality reports with false-positive and false-negative rates over time, by category

If you want to see what an advisory verdict looks like before you shortlist anything, our free competitor price checker runs the same pattern on two product URLs. It compares brand, model codes and title words, returns a High, Medium or Low confidence label, and warns you when the two pages are not the same product.

How Do You Choose a Product Matching Solution?

A product matching solution is any tool or service that takes two product lists and returns linked pairs with a confidence score. You can build it, buy a matching or PIM tool, or use a managed data service. The right choice depends on how many sources you match and who will keep the rules up to date.

Option What you get Works well when Watch out for
Build it in-house Your own rules, fuzzy matching and review queue You have engineers, a stable set of sources and clean identifiers Upkeep grows with every new source and language
Buy a matching or PIM tool Ready matching logic inside a catalog or pricing product Most of your matching is your own catalog against supplier feeds Competitor and marketplace data still has to be collected first
Use a managed data service Collected competitor data delivered already matched to your SKUs You need matched competitor prices and do not want to run the pipeline You depend on the provider's review process, so ask how low-confidence pairs are handled

Questions to ask any vendor: which identifiers are matched first (EAN, GTIN, MPN), what happens to products without an identifier, how multipacks and variants are treated, and whether you can see and correct the low-confidence pairs. Ask them to run your own sample, not their demo data.

ScrapeWise, which publishes this blog, is one managed option. Its Managed Data service scrapes the shops you name, matches each listing to your SKU (by EAN or SKU when both sides have it, otherwise by title, brand, size and picture, with unsure pairs going to review) and delivers one table by REST API or CSV/Excel, at most once a day. Self-serve competitor price tracking runs the same matching: you upload your product list as the master file, connect the columns, and review the unsure pairs yourself.

If you want the method step by step first, the free course Matching products across sites covers identifiers, matching without a barcode, multipacks and review queues in six short lessons.

Which Ecommerce Platform Is Best for Automating Product Mapping From CSV Titles and SKUs?

No ecommerce platform does this well on its own. Shopify, WooCommerce and Adobe Commerce (Magento) only update a product when an exact key such as the handle, ID or SKU matches. The best setup is a matching step before the import: match on EAN or GTIN, fuzzy-match titles and attributes, review low-confidence pairs, then import the clean mapping.

This is what each option does with a CSV file:

Option How it maps CSV rows Handles different titles? Best for
Shopify product CSV import Overwrites products whose URL handle matches No, exact handle only Bulk edits to your own catalog
WooCommerce product CSV importer Updates existing products with the same ID or SKU No, exact ID or SKU only Stores with clean, consistent SKUs
Adobe Commerce (Magento) import Add/Update import keyed on SKU No, exact SKU only Large catalogs with a strict SKU master
PIM or feed management tool Rule-based field mapping and attribute cleanup Partly, through rules and synonyms Sending one catalog to many sales channels
AI product matching tool Scores title, brand, EAN/GTIN and attributes, then links records across sources Yes, with a confidence score per pair Matching supplier or competitor products that use other titles and SKUs

A workflow for CSV titles and SKUs:

  1. Clean the file first. Lowercase the titles, remove marketing words, split out brand, size, colour and pack count, and write units one way ("64 Gigabyte" becomes "64GB").
  2. Match on hard identifiers. Rows that share an EAN, UPC or GTIN, or the same brand and manufacturer part number, are near-certain matches.
  3. Fuzzy-match the remaining rows on title and attributes, and give each candidate pair a confidence score.
  4. Approve high-confidence pairs automatically and send medium-confidence pairs to a short human review (see Confidence Scoring and Review above).
  5. Export the approved mapping as a clean CSV, with your SKU next to the matched SKU or URL, and import it on the platform's exact key.

The platform import stays the last step. All of the matching happens before it.

How Do You Roll Out Product Matching?

Start with one category where you already know the matching is bad. Run the new matching next to your current manual process for a few weeks, compare the two, and set your confidence thresholds from what you see. Then add categories one at a time and keep a person on the review queue.

Pilot. Pick one category with known duplicates or a competitor set you already track by hand. Check a sample of matches yourself, including the ones the tool rejected.

Scale. Add categories one by one. Tune thresholds per category, because fashion leans on images and colour, while electronics leans on model numbers.

Measure. Track four things over time:

  • False positives: wrong pairs that were accepted
  • False negatives: real pairs that were missed
  • Review queue size compared to what your team can handle
  • Hours spent on matching compared to before

Connect. Feed the matched pairs into your PIM, your pricing tool and your reports, so every team works from the same mapping.

Where to Start

Product matching is not glamorous, but every price comparison, stock count and catalog import depends on it. Begin with your identifiers, clean your units and pack sizes, score the rest and keep a person on the unsure pairs. Whether you build it, buy a tool or have it done for you, ask to see the review queue before you trust the numbers.

If you want competitor data to match against, sign up for ScrapeWise and run a few of your competitors' product pages: the first 5 requests are free, no card needed. You get the price, title, brand and EAN where the shop publishes it, ready for the matching steps above. If you'd rather have the matching done for you, Managed Data covers that.

Stop checking competitor prices by hand

Paste a competitor product URL. ScrapeWise checks it every day and matches it to your product. 5 free requests, no card.

Not ready to sign up? See 40 rows of Amazon Best Sellers data →

5 free requests no credit card · pay per page, no monthly plan · book a 15-min call →

FAQ

Frequently asked questions

Product matching in ecommerce - how the algorithm, AI matching and competitor matching work

Product matching is finding and linking the same product across different lists, such as your catalog, supplier feeds, marketplaces and competitor sites, even when titles, SKUs or images differ. It compares identifiers like EAN or GTIN, cleaned titles, attributes such as brand, size and pack count, and sometimes pictures. The result is a mapping with your SKU next to the matched SKU or URL, which makes like-for-like price comparison, clean catalogs and correct stock counts possible.

A product matching algorithm checks the strongest signal first. It links rows that share an EAN, GTIN or brand plus manufacturer part number, then compares cleaned titles, then attributes like size, colour and pack count, and sometimes product images. Each signal adds to a confidence score. Pairs with a high score are accepted, pairs in the middle go to a person for review, and pairs with a low score are dropped.

AI product matching uses language and image models to decide whether two listings are the same product, instead of only fixed rules such as exact keys or word-overlap thresholds. It handles synonyms, other languages and messy titles better than rules alone. It still needs clean units and pack sizes, identifiers where they exist, and a human review of unsure pairs, because look-alike variants such as a 500 g bag and a 2 kg tub can fool any model.

Competitor product matching links each of your SKUs to the same product on competitor sites, so price comparisons are like for like. Competitor pages rarely show your SKU and often hide the EAN, so matching relies on brand, model, size and pack count. Without it, a price gap can simply be two different products. Managed services such as ScrapeWise Managed Data deliver competitor prices already matched to your SKUs.

Yes. Without EAN or GTIN codes, matching uses the cleaned title, brand, size, colour and pack count, and product pictures where titles are not enough. Each pair gets a confidence score, and the unsure ones go to a person to confirm. Pack sizes and variants are kept apart, so a 2 kg tub is not matched to a 500 g bag. Expect more pairs in review than with codes, so plan time for that.

No ecommerce platform does this well on its own. Shopify matches on the product handle, WooCommerce on ID or SKU, and Adobe Commerce (Magento) on SKU, so a row only updates a product when that key is identical. The best setup is a matching step before the import: match on EAN or GTIN, fuzzy-match titles and attributes, review low-confidence pairs, then import the clean mapping.