Multilingual AI Chatbot for an Ecommerce Website: How It Works in 2026

TL;DR: A multilingual AI chatbot for an ecommerce website answers each shopper in the language they write in, even when the store's pages and products exist only in English, because the AI reads the content once and writes the answer in the visitor's language. Chatonbo answers in 50+ languages, auto-detected from what the visitor writes, including right-to-left ones such as Hebrew and Arabic; the chat interface itself is localized in 14 languages, and you can fix one language per assistant when you prefer. Live Chatonbo assistants already run in Hebrew, Spanish, Dutch and German as their fixed language, alongside the English majority.

How does a multilingual chatbot work on an English-only store?

Older chatbots needed a separate script per language, so a store that wanted Spanish support had to write the Spanish flows by hand. An AI assistant works differently. It indexes your pages, products and policies in whatever language they are written in, and when a shopper writes in Spanish, it answers in Spanish from that same content. The knowledge is one copy; the language is chosen per conversation.

Two things decide how good that feels to the shopper:

• Answer language. The AI model's ability to write a natural reply. Chatonbo answers in 50+ languages, auto-detected from what the visitor writes.
• Widget language. The buttons, the placeholder in the input box, the greeting, the "talk to a human" label. These are not generated; they are translated once by the vendor, so the list is shorter. The chat interface itself is localized in 14 languages and follows the visitor's browser by default.

A chatbot that answers in Portuguese but surrounds the answer with English buttons feels like a translation layer. One that gets both right feels like a local store.

What should you check before choosing one?

The [multilingual assistant page](/solutions/multilingual-agent) lists the interface languages and shows the layout in both text directions.

Where does multilingual chat actually pay off?

Three situations come up again and again on live stores.

Cross-border shoppers on a single-language store. A Dutch store gets German and French visitors; an Israeli store gets English-speaking tourists. Without a multilingual assistant those visitors either guess from the pictures or leave. With one, they ask about shipping to their country in their own language and get the answer from your shipping page.

Stores in a non-English market. Chatonbo assistants fixed to Hebrew, Spanish, Dutch and German are live today. For these stores the question is not whether the AI can translate, it is whether the chat window and the right-to-left layout are native. That is an interface question, not a model question, which is why the interface language list matters.

Support outside business hours. Across live assistants on store and business sites, 39% of conversations start outside 9:00 to 18:00 in the visitor's local time. For a store with international visitors the share is naturally higher, because their daytime is your night. A multilingual assistant covers both the hour and the language.

How do you set it up?

1. Create the assistant from your store URL. Chatonbo reads the pages and products as they are, in whatever language they are written.
2. Leave the language setting on automatic so the assistant follows each visitor, or fix one language on the bot's settings page.
3. Pick the interface language, or leave it on "follow the browser".
4. Test it: open the chat and write in two or three languages your visitors use. Product cards should show the same products each time with the surrounding text in the right language.

The same assistant handles lead capture, product cards and, on Shopify, order lookup, in every language. There is nothing multilingual-specific to maintain afterwards.

What does it cost?

Chatonbo pricing is flat and the language support is the same on every plan: free to start, Pro at $19 a month for 5,000 conversations, Business at $59 a month for 10,000. There is no per-language fee and no per-message meter. See the [pricing page](/pricing).

What does the same question look like in three languages?

A shopper on an English-language store asks about delivery to their country. The assistant reads one shipping page and answers each visitor in their own language:

• In Spanish: the shopper asks whether the store ships to Spain and how long it takes; the assistant answers in Spanish with the delivery window and cost from the shipping page, and links the page.
• In Dutch: the same question gets the same facts in Dutch, with the product card for the item they mentioned, in the same layout.
• In Hebrew: the answer arrives in Hebrew with the layout flipped to right-to-left, including the buttons and the input box, because the interface detects the text direction.

Nothing was translated by hand. The facts come from one English page; only the language of the answer changes, and the product cards link to the same products every time.

What are the common mistakes when a store goes multilingual?

• Translating the FAQ instead of the assistant. Stores sometimes keep a Spanish FAQ page and an English one and expect the chatbot to pick the right one. An AI assistant does not need that; it needs one accurate source and answers in whatever language the visitor writes. Two versions of the same page drift apart and the assistant ends up quoting the stale one.
• Checking the model, not the interface. The demo answers beautifully in Portuguese, then the live widget shows "Type your message" in English above it. Test the buttons and the greeting in the languages you care about, not only the replies.
• Forgetting right-to-left layouts. Hebrew and Arabic answers inside a left-to-right chat window read wrongly even when every word is correct. Open the chat on a phone, write in Hebrew, and look at the alignment.
• Locking a language when visitors are mixed. Fixing the assistant to German is right for a German-only store and wrong for one with Austrian, Swiss and Dutch visitors who write in three languages. Leave it on automatic unless the team needs every answer in one language.
• Measuring nothing. The conversation list shows the language of each chat. After a month, look at which languages visitors actually used; stores are regularly surprised by the second and third ones.

How do you measure whether it is paying off?

Compare the share of conversations in each language with the share of orders or leads from those countries. If Spanish conversations are a noticeable share of chats but Spanish-speaking visitors rarely leave contact details or reach the cart, the problem is usually in the content the assistant has, for example a shipping page that does not mention Spain at all, not in the language. Add the missing fact and the assistant answers it in every language from then on.

Try it in your visitors' languages

Enter your store URL, then write to the drafted assistant in any language your shoppers use. [Try it on my store](/try-it).