Ecommerce Localization and International SEO: How to Translate and Localize Product Catalogs for Search and AI

1. 01

   Shoppers buy in their own language: in a CSA Research survey of 8,709 consumers in 29 countries, 76% preferred product information in their native language and 40% never buy from sites in other languages.

2. 02

   A translated keyword is often a keyword nobody types. UK shoppers search "trainers", Irish shoppers say "runners", Italians search "piumino" and Dutch shoppers search "regenjas". Every market needs its own keyword research.

3. 03

   Sizes and units have to be local too, along with prices and attributes: size_system, centimeters or inches, decimal commas, local currency, and the attribute names shoppers actually filter on.

4. 04

   International SEO and AI visibility rest on the same foundation: one complete, native-quality product page per market, connected with hreflang and mirrored in a localized product feed.

5. 05

   AI agents now answer in the shopper's language. Google's AI Mode added more than 35 languages in October 2025, and an agent can only recommend what your local product data actually says.

Why translated keywords miss local demand

Translated keywords fail in predictable ways. Synonyms split by country ("sweater" here, "jumper" there). Dutch and Italian shoppers often search the English word ("sneakers", "hoodie"). And Italians search "piumino" for a down jacket, a word no translation tool produces from "down jacket".

Plenty of localization advice gets this wrong, insisting on "proper" local language. If Dutch shoppers type "sneakers", the Dutch title should say sneakers. Search volume decides.

Same product, different searches

Treat these as starting points and validate each term with country-level search volume.

Keyword research per market, step by step

1. Start from the product (type, material, use case, audience), not your English keyword list.
2. Collect native seed terms from local competitors and marketplaces, and from autocomplete in the target country.
3. Pull search volume per country. "Jumper" in the UK and in Ireland are separate data points.
4. Map one primary keyword per product type and market, plus modifiers (gender, material, fit, occasion).
5. Add local questions such as "welke regenjas is echt waterdicht". AI agents answer these too.
6. Write the keywords into titles and H1s, descriptions, then category names. Our ecommerce SEO guide covers placement.

English is not one market

The UK and Ireland share a language but not a currency (GBP and EUR). Irish shoppers mostly use UK clothing and shoe sizes, so size labels can stay. Prices and some vocabulary can't. A separate en-IE version with euro prices, Irish delivery times and words like "runners" serves Irish shoppers better than the UK page. Hreflang tells Google which to show where.

Choose a site structure

This follows Google's guide to managing multi-regional and multilingual sites. Google prefers separate URLs per language over cookies or browser settings, and advises against automatic language redirects. It ignores the lang attribute and reads language from visible content, so keep each page in one language, navigation included. On Adobe Commerce, each locale usually runs as its own store view, and hreflang typically comes from an extension or custom code; our Magento SEO guide covers that setup.

Country domains still get sold as the "strongest signal". For most brands we'd pick subfolders anyway. A new brand.it starts with zero authority, and you pay to run it every month for a signal hreflang mostly gives you already.

How hreflang works on a product page

Each product page lists every version of itself, including itself. For the fictional Harbor jacket in four markets, every version carries this set:

• <link rel="alternate" hreflang="en-GB" href="https://brand.com/en-gb/products/harbor-jacket" />
• <link rel="alternate" hreflang="en-IE" href="https://brand.com/en-ie/products/harbor-jacket" />
• <link rel="alternate" hreflang="it-IT" href="https://brand.com/it-it/products/harbor-jacket" />
• <link rel="alternate" hreflang="nl-NL" href="https://brand.com/nl-nl/products/harbor-jacket" />
• <link rel="alternate" hreflang="x-default" href="https://brand.com/products/harbor-jacket" />

Google's hreflang documentation sets the rules. Link tags in the HTML work the same as HTTP headers or XML sitemaps. "If two pages don't both point to each other, the tags will be ignored." Each version must list itself, codes are ISO 639-1 languages with optional ISO 3166-1 Alpha 2 regions, and URLs must be fully qualified. The UK's region code is GB, so en-GB, never en-UK.

Hreflang won't make a weak Italian page rank. It only swaps in the right version once a page ranks.

It also goes wrong a lot, according to Ahrefs.

Missing x-default tops that list, but we rarely start there. Broken return tags and redirected targets do more damage, because Google then ignores the tags.

Ecommerce-specific hreflang rules

• Only connect true equivalents. If a product isn't sold in Ireland, leave en-IE out rather than pointing it at a category.
• Point hreflang at canonical URLs only, never at filtered URLs like ?color=navy.
• Never canonicalize /it-it/ pages to the English URL. That tells Google to drop the Italian page.
• Retire discontinued products in all markets at once, so no tag hits a 404.
• On large catalogs, hreflang in XML sitemaps is easier to maintain than tags in the head.
• Add a visible language and country selector, because geotargeting, in Google's words, "isn't an exact science".

Structured data has to match the market too, with Offer price and priceCurrency in local terms.

In October 2025 Google expanded AI Mode to more than 35 new languages and over 200 countries and territories. In the US, AI-referred retail visitors converted 60% better than non-AI traffic in July 2026 (Adobe). As that spreads to European languages, the data agents read in each language decides who gets recommended.

How a local-language prompt gets answered

An Italian shopper asks: "Mi consigli un piumino leggero da uomo per andare in bici in città, sotto i 200 euro?" The agent splits this into sub-queries (lightweight down jacket, men's, cycling, under €200) and pulls candidates from Italian pages and feeds. A page that says "piumino leggero", gives the weight in grams and mentions cycling can be matched. A translated title on an English-only spec sheet can't. Irish English works the same way.

ChatGPT

Demo illustrativa

ChatGPT

ChatGPT

It could say all that only because the en-IE page stated it, in euros, using the word Irish shoppers use.

What agents need from a localized catalog

• Explicit attributes. "Waterdicht tot 10.000 mm" is matchable; "perfect voor elk weer" is not.
• Local use cases, like bike commuting in Amsterdam or wet Irish winters. See how to write product descriptions.
• Local FAQ, phrased the way shoppers ask. A product description template per category keeps the structure identical across languages while the wording changes.
• Market-aware feeds. The OpenAI feed spec has no language field, so text must already be local; target_countries and store_country set where the offer applies. More in getting recommended by ChatGPT.

This is Agentic Commerce Optimization applied market by market: the same four dimensions (completeness, keywords, intents, metadata), scored and fixed in every language. See also GEO for ecommerce.