Multilingual ecommerce chatbots: what to evaluate

A multilingual chatbot is not simply a translator placed on top of chat. It has to understand requests, find products, and return information consistent with the language, market, and content actually available in the store.

Start with languages you genuinely support

List the languages present in the site, catalog, and policies. An interface language does not imply that every product page is complete in that language. Define what happens when a translation is missing.

Keep catalog terminology consistent

Category names, attributes, and variants should be connected across languages. Literal translation is not enough: shoppers in different markets may express the same need with different terms.

Valen imports the product catalog from Shopify and enriches its attributes, as demonstrated on a demo store.

Check search and product questions

Prepare equivalent queries in each language and verify that they lead to comparable selections. Include synonyms, common mistakes, and sentences that mix local and international terms.

Valen lets shoppers search the catalog and ask questions about products within the conversation, as demonstrated on a demo store.

Adapt tone without changing facts

Tone may vary across markets, but characteristics and policies should remain consistent with the source. When a translation makes an attribute ambiguous, the chatbot should ask for clarification rather than choose an interpretation.

Verify the interface across devices

Test longer text, special characters, right-to-left writing where applicable, and mobile keyboards. Make sure buttons and product cards remain readable. The Valen chat is available on mobile devices and desktop computers during browsing, as demonstrated on a demo store.

Launch with a clear scope

You can begin with one language and expand after content and tests are complete. Keep configurations separate when different markets need different functions. Valen keeps a registry of features that can be enabled separately for each store, as demonstrated on a demo store.

Build a test matrix

For each language, prepare the same request families: exact search, open-ended need, comparison, unanswered question, and ambiguous term. Do not expect identical sentences. Check instead that facts, constraints, and products remain consistent. Include someone fluent in the language to identify wording that is grammatically correct but unnatural.

Repeat the matrix when the catalog or policies change. Keeping previous cases prevents an improvement in one language from introducing a regression in another.

Where Valen fits

The limitation of a multilingual chatbot is the gap between a translated sentence and the market's actual data. Addressing it requires a structured catalog, product-linked questions, and an interface tested on shoppers' devices. Valen imports the catalog from Shopify, as demonstrated on a demo store. It lets shoppers search that catalog and ask product questions in the conversation, as demonstrated on a demo store. Its chat is available on mobile and desktop while browsing, as demonstrated on a demo store. Valen therefore connects conversation with the storefront rather than remaining an isolated chat; language coverage and translation quality still need validation for each market.

Evaluate accuracy, consistency, and the ability to state limitations in every language. A localization is not complete merely because the answer is grammatically correct. Book a demo to test multilingual requests against your catalog.