A multilingual chatbot is an AI chatbot that reads a visitor's message in their language and replies in that same language, from one set of business content. You do not need a copy of your website in every language. You do need to test the languages your customers really use, set a default language, and keep your team on hand for conversations that need a person.

This guide is for a business that gets questions from customers who write in more than one language. It explains how the reply language is chosen, when one agent is enough, how to test it, and where the limits are. It ends with a worked example and a short test plan you can run this week.

Key takeaways

  • A multilingual chatbot answers customers in more than one language without you building a separate chatbot for each.
  • If you want a human to step in on the hard cases, see our guide to the best AI chatbot with human handover.
  • People prefer to buy in their own language.
  • Language support is not a paid add-on in Dante AI.

What is a multilingual chatbot?

A multilingual chatbot answers customers in more than one language without you building a separate chatbot for each. A visitor writes in Spanish and gets a Spanish answer. The next visitor writes in Polish and gets a Polish answer. Both answers come from the same content: your website, your documents and the question-and-answer pairs you add.

Modern AI chatbots can do this because the language model behind them can read and write many languages. The chatbot looks up the facts in your content, then writes the answer in the visitor's language. That is different from older chatbots, which needed a hand-built script for every language.

Dante AI states on its website that agents support 100+ languages and answer in the language the visitor writes in, on every plan. Coverage is not the same as quality in every language, so the rest of this guide is about checking the languages that matter to you.

How does a multilingual chatbot pick the reply language?

Three things decide the reply language, in this order:

  1. Your instructions. If your agent's instructions say "always reply in French", it replies in French, whatever the visitor writes. Only add a rule like this if you mean it.
  2. The visitor's latest message. If you set no language rule and the visitor clearly writes in another language, the agent mirrors the language of their most recent message. If they switch from English to Italian halfway through, the reply switches too.
  3. The agent's default language. When neither of the above applies, for example a one-word message that could be several languages, the agent uses its default.

One detail matters a lot: your content can be in a different language from the answer. If your returns policy is written in English and a visitor asks in German, a Dante AI agent finds the English policy and translates the facts into a German answer. It does not switch into English just because the source page was English.

You can read more about how instructions shape replies in the instructions guide.

Do you need your content in every language?

No. For most small and mid-sized businesses, one well-kept set of content in your main language is the right start. Translating every page first is slow, and two copies of the same policy drift apart.

There are three cases where extra content in another language helps:

  • Words that must be exact. Legal wording, safety instructions and regulated terms should not be machine-translated on the fly. Write the approved version in each language and add it as a question-and-answer pair.
  • Local differences. If delivery times, prices or opening hours differ by country, say so in your content, country by country. The chatbot can only be as specific as your sources.
  • Local words for your products. Customers search in their own terms. Adding the common local name for a product helps the agent match the question to the right page.

If you have not trained a chatbot before, start with our guide on how to train a chatbot on your own data.

Should you run one agent or one per language?

Start with one agent. Move to one agent per language or market only when the content or the team behind it is truly different. This table compares the common setups.

SetupHow it worksBest forWatch out for
One agent, one set of contentContent in your main language; replies mirror the visitorSame products, prices and policies everywhereQuality in less common languages; check them
One agent plus translated key answersSame as above, with approved Q&A pairs in each key languageExact wording on returns, legal or safety topicsKeeping each translated answer up to date
One agent per language or marketSeparate agents, each with its own content and instructionsDifferent prices, laws or teams per countryMore to maintain; Dante AI includes 1 agent on Free and Starter, 3 on Advanced, 10 on Pro
Translated website, no chatbotVisitors read translated pages and contact you by emailVery low question volumeNo answer outside office hours; visitors must find the page
Team answers every language by handPeople reply to every message, with or without translation toolsComplex, high-value conversationsSlow replies when the team is busy or offline

In practice most businesses combine the first two rows with a person for the conversations that need one. The chatbot answers the routine questions in every language, and your team takes over when a customer asks for a human or the question needs judgement.

What does this look like for a real business?

Take a bike rental shop in Lisbon. Its website is in English. Most customers are tourists, and they write in English, German, French, Spanish and Portuguese.

  1. Setup. The shop adds its website and its rental terms PDF to one Dante AI agent. The default language is English. The instructions say nothing about language, so replies mirror each visitor.
  2. A German visitor asks: "Kann ich das Fahrrad an einem anderen Ort zurückgeben?" (Can I return the bike at a different location?) The agent finds the English rental terms and answers in German with the shop's real return rule.
  3. A French visitor asks about a deposit. The deposit amount sits in the English PDF. The agent answers in French and keeps the amount exactly as written.
  4. The exact-wording fix. The owner wants the damage policy worded carefully. She writes an approved version in each of the five languages and adds them as question-and-answer pairs. Q&A pairs take priority over website content for matching questions, so the agent answers from them, in its own tone.
  5. The handover. A Spanish visitor has a complaint about a broken bike. The shop is on a plan with Human Handover, so the visitor taps "Talk to a human" and a team member picks it up in the Inbox.

The shop did not translate its website. It translated the handful of answers that must be exact, and it kept a person in the loop for complaints.

How do you test a multilingual chatbot before customers use it?

Run this short plan for each language that brings you real customers:

  1. Write five real questions in that language, taken from emails or calls you have actually received.
  2. Ask them on your live site, on a phone and on a laptop.
  3. Have a fluent speaker read the answers. Machine-fluent is not the same as natural. Ask them whether the tone fits your brand.
  4. Check the facts that must not change: prices, dates, product names, addresses and phone numbers.
  5. Test the handover in that language, so you know how a request for a person looks to your team.
  6. Repeat after every big content change. Our AI customer support accuracy test gives you a simple way to score the answers.

After launch, read what people actually ask. In Dante AI, the Top Questions panel in Advanced Insights lists repeated visitor questions and works in any language, including non-Latin scripts such as Arabic, Greek or Chinese. Advanced Insights is available on the Advanced and Pro plans.

What are the limits to know about?

  • Quality varies by language. Widely used languages tend to read better than rare ones. Test before you promise a language to customers.
  • Very short messages are hard to place. "Ok" or a single product name could be many languages. The agent may then fall back to its default language.
  • Some features listen for common phrasings. In Dante AI, the lead capture form can open when a visitor asks to be contacted. Common phrasings are recognised, not every wording or language, so try yours in the preview.
  • Regulated wording needs a person. Have someone on your team approve anything legal, medical or financial in each language.

If you want a human to step in on the hard cases, see our guide to the best AI chatbot with human handover.

Why does answering in the customer's language matter?

People prefer to buy in their own language. CSA Research surveyed 8,709 consumers in 29 countries for its 2020 "Can't Read, Won't Buy" study. It found that 76% of online shoppers prefer to buy products with information in their native language, and 40% will never buy from websites in other languages.

A multilingual chatbot does not fix a site that is only in one language. It does mean a visitor can ask a question in their own words and get an answer they can read, at any hour. That keeps them on the page long enough to decide.

What does a multilingual chatbot cost on Dante AI?

Language support is not a paid add-on in Dante AI. Every plan, including Free, can answer in the visitor's language. What changes between plans is how many credits, agents and team features you get. You can check the current plans on the pricing page.

The easiest test is free: add an AI agent to your website, then ask it your five real questions in each language your customers use.

Sources