If your website gets visitors from more than one country, some of them will type their question in another language. Modern AI models can often read and write several languages, which makes it tempting to say "our chatbot speaks everything". That promise is easy to make and hard to keep accurately.
This guide helps you decide which languages to support properly, how to keep answers accurate in each, and how to test before you tell visitors the chatbot can help in their language. I am not claiming any particular product supports a given language, so verify language behaviour with any vendor, including us, through your own testing.
Should Your Chatbot Support More Than One Language?
Support another language when you have real visitors writing in it and someone who can stand behind the answers. A language you can neither check nor staff is a risk, because wrong answers in a language you cannot read go unnoticed.
- Check your analytics: which countries and browser languages send meaningful traffic?
- Check your inbox: do customers already write to you in other languages?
- Check your capacity: can someone review answers and handle handoffs in that language?
- Start small: one extra language done well beats five done badly.
Is Machine Translation Good Enough?
It is a good starting point and a poor finish. Machine translation is usually fluent but can misjudge tone, idioms, units, legal terms and product names. A fluent wrong answer is worse than a clumsy right one, because it sounds trustworthy.
Treat machine translation as a first draft. Have a fluent speaker review the answers that matter most: prices, policies, safety information and anything legal. For less critical answers, a lighter review may be acceptable.
How Do You Write Answers for Each Language?
Write the facts once, carefully, then adapt them for each language and market, not just translate them. Prices, currencies, dates, units, phone formats and policies can differ by country. A return window that applies in one region may not apply in another.
| Element | What to check per language or market |
|---|---|
| Currency and prices | Show the right currency and whether tax is included |
| Dates and times | Format, time zone and opening hours |
| Units | Metric or imperial, sizes, measurements |
| Policies | Returns, delivery and legal terms that differ by region |
| Contact routes | Phone numbers, hours and support languages available |
| Tone | Formality and phrasing normal in that language |
What About Handoff to Your Team?
A chatbot that answers in a language your team cannot read creates a dead end when it needs to hand off. Before you offer a language, decide who will reply when a visitor asks for a person. If nobody can, say so honestly in that language and offer a channel that works, such as email in English.
A shared inbox helps because every conversation lands in one place with the visitor's details attached. See the shared inbox and human handoff guide for how to set this up.
How Do You Test Before You Promise?
- 1List the ten most important questions and write them as a native speaker would phrase them.
- 2Ask them in the chat, in a private window, in each language you plan to support.
- 3Ask a native speaker to read the replies for accuracy and tone, not just fluency.
- 4Try mixed-language messages, typos and informal phrasing.
- 5Check that the "I don't know" and handoff replies work in that language.
- 6Only then tell visitors that the chatbot supports that language.
Our pre-launch testing guide covers the general checklist, and the same rigour applies in every language.
How Should You Describe Language Support?
Say exactly what you support and be honest about limits. "Answers available in English and Spanish" is clear. "Speaks all languages" invites visitors to rely on something you have not checked. If you offer a language with limited coverage, say that some questions will be passed to your team.
Keep a record of what you tested and when. If you add a language later, repeat the checks rather than assuming it works like the others.
Pro Tip: Look at the unanswered-question list by language. A cluster of failed questions in one language usually means that language needs more or better answers, not that the visitors are unusual.
What Pitfalls Should You Watch For?
Language problems often appear in the details that a quick test misses. Include these in your test list for every language you plan to support.
- Mixed-language messages: visitors often combine languages or paste product names in English.
- Regional differences: the same language varies by country in spelling, vocabulary and formality.
- Names, addresses and special characters: check they display and save correctly in the inbox and any emails.
- Scripts that read right to left: check the widget's layout and any text it displays, if you plan to support them.
- Legal and policy wording: translated terms may not mean the same thing legally, so have them reviewed.
How Do You Keep Several Languages Up to Date?
Every change doubles or triples the work, so keep the structure simple. Maintain one master version of each fact, mark which languages have been updated and review all languages whenever a price or policy changes.
A short table in your maintenance document, with one row per answer and one column per language showing the last reviewed date, prevents the quiet problem of an old price lingering in a language nobody checks.
Related reading: shared inbox and human handoff, how to test a chatbot before launch and knowledge base maintenance.
Conclusion
Support another language only when real visitors need it and someone can stand behind the answers. Treat machine translation as a draft, adapt facts to each market, plan human handoff in that language, test with native speakers and describe language support honestly.
Key takeaways
- Support a language only when real visitors need it and you can check the answers.
- Machine translation is a first draft, not a finished answer.
- Adapt prices, units, policies and tone for each market.
- Plan who handles handoffs in each language before offering it.
- Test with native speakers and describe support honestly.
Quick Summary
A multilingual website chatbot should support a language only when real visitors need it and someone can check the answers. Use machine translation as a draft, adapt prices, units, policies and tone to each market, plan who handles handoffs in that language, and test with a native speaker before promising support. Describe exactly which languages you offer, and verify any vendor's language behaviour through your own tests.
Want to see grounded answers on your own site?
Start free, add your Q&A pairs and paste one script tag. No credit card required.

Founder & CEO
Writes about grounded AI chatbots for websites: how they work, how to measure them and what to check before you buy.
Topics
Get new posts by email
Guides on AI chatbots for websites, straight to your inbox.
