How answers get produced in another language
Your knowledge base can stay in one source language โ English, in most cases โ and the assistant answers a question asked in Spanish, Portuguese, Arabic or any of the other supported languages by drawing on that same English material and replying in the customer's language. There's no separate translated knowledge base to keep in sync, which is the main reason this scales to more markets without more maintenance work.
This also means an edit to your English source material updates the answer in every language immediately โ the same re-indexing behaviour that applies to any other knowledge-base change applies here too, with nothing extra to do per language.
What this is good at
Factual, well-documented questions โ order status, how a feature works, what a policy says โ translate reliably, because the answer is a fact being restated in another language, not a judgment call that depends on cultural context. This is most of what a support queue is made of, which is why source-language knowledge bases work for the large majority of conversations without any localisation effort.
When to translate the source instead
Legal and compliance content is the clearest case for maintaining a genuinely translated source rather than relying on answer-time translation โ terms of service, region-specific policy, anything where the exact wording carries legal weight. A translated-at-answer-time response is a faithful rendering of the English source, but if the English source itself doesn't reflect a region's actual legal requirements, translating it accurately just produces a wrong answer in another language.
Brand voice and idiom are the second case. If tone matters more than fact in a given market โ a consumer brand leaning on humour or a specific style โ a source written and reviewed in that language directly will read more naturally than one translated on the fly, even when the translation is accurate.
Translated messages are a metered, shared allowance
Translation is included from the Team plan up, with a monthly allowance shared across your whole team rather than metered per person โ 3,000 messages on Team, 7,000 on Growth, 25,000 on Scale. One person handling most of a language's volume can use as much of that shared allowance as they need; it's there to catch runaway usage, not to ration ordinary use.
Check translated transcripts occasionally, not just English ones
It's easy to review transcripts in your own language and assume the same resolution rate holds everywhere โ but a knowledge gap or an ambiguous article can translate differently than intended, and that's easiest to catch by actually reading a sample of conversations in your other supported languages, not by inferring quality from the English-language numbers alone.
