← All posts

Same product, treated as different ones: what a marketplace sale reveals about product data

During a big marketplace sale, the same product shows up in dozens of shops. Whether an AI can tell they're the same product comes down to model numbers, GTINs and how specs are written. What manufacturers and retailers should check.

2 min read

By ZAG

日本語版 →

Early December is when Japan's big online marketplaces usually run their large sales events, Rakuten Super SALE among them. A popular product can appear under the same model number in dozens of shops.

A person sees right away that those listings are the same product, however differently each shop words the title. An AI pulling information together across shops needs something to go on to decide whether two listings are the same product. Without it, one product gets treated as several, or specs from different products get mixed together.

Here are three things that give the AI something to go on. Every example is fictional.

1. Model number and GTIN: the strongest signal

Say a fictional handheld fan, EXAMPLE ZC-4520H, is listed in three shops like this:

  • Shop A: title "[SALE] EXAMPLE handheld fan ZC4520H white". Model number and GTIN fields blank
  • Shop B: title "EXAMPLE ZC-4520H handheld fan, white". Model number field ZC-4520H, GTIN field 2001234567893
  • Shop C: title "Quiet powerful USB handheld fan". Model number and GTIN fields blank

Shops, product, model number and GTIN are all fictional.

Shop B puts the model number and GTIN in their own fields, so it can be matched. Shop A has the model number in the title, without its hyphen and buried in the text. Shop C gives nothing that identifies the product.

A model number mixed into free text isn't recognized as a SKU or GTIN. What works best is a split of duties: the manufacturer settles on one way of writing each model number and GTIN and hands it to retailers, and retailers put it in the dedicated fields.

2. Specs: in attributes, not in the title

During a sale, titles collect words like "SALE", "free shipping" and "powerful". When specs like capacity or weight are packed into the title as well, an AI can't tell where the product name ends and the spec values begin.

Take specs out of the title and write them as attribute, value and unit. The steps are in our post on PDF catalogs.

3. Consistency: does your site match the marketplace?

Say the manufacturer's site lists a model at "weight 250 g" and one shop lists the same model at "approx. 300 g". The AI will pick one or use neither. Either way, what the manufacturer meant doesn't come through.

Most mismatches come from old catalog values still sitting with retailers. When the manufacturer revises a spec, the data sent to retailers gets updated the same day. It's unglamorous, and it's the most reliable way to stay consistent.

Checklist for manufacturers and retailers

Manufacturers

  • One agreed way of writing each model number (hyphens, case)
  • A GTIN for every model number, passed on to retailers
  • Specs sent to retailers as attribute, value and unit
  • Retailer data updated the same day a spec is revised

Retailers

  • Model number and GTIN in their dedicated fields
  • Specs in attributes, not in the title
  • The latest values from the manufacturer in use

Mid-sale is not when anyone has time to fix pages. Read this as homework for the next one.