How Meeto's AI summary handles meetings in several languages

Why one language per meeting does not work

Plenty of meetings are not held in a single language. A team in Lisbon talks in Portuguese while a colleague from Berlin answers in English. A sales call starts in Spanish and moves to English for the technical part. A developer stand-up is held in Russian, yet half the words are product names and code. A note-taker that asks for one meeting language before the call starts mishears everyone who speaks another one.

Meeto has no meeting-language setting for that reason. You start AI summary and people talk the way they normally would; the language is worked out from the speech itself.

How does Meeto tell which language each person speaks?

Language is tracked separately for every speaker, never for the room as a whole, so a meeting can hold any number of languages at once. Each participant's own voice is matched against their own history:

• A new speaker starts with no assumption. Every clear phrase is accepted in whatever language it is, and after two clear phrases Meeto knows the language that person speaks.
• Switching is immediate. A clear sentence in a new language switches that speaker at once, and going back to a language they already used needs no confirmation.
• Unclear moments wait briefly. A short, unclear phrase that sounds like a new language is held until the speaker's next words show whether they really switched. It is not thrown away: it appears once confirmed, or when the speaker pauses.
• Short replies stay replies. A quick “Yes” or “Okay, sure” in the language a colleague is using is kept as said and does not change what the speaker is expected to speak.
• Echoes and noise are filtered. A phrase in an unexpected language that overlaps another participant speaking that language at the same moment is treated as their voice picked up by this microphone, and stock lines that speech recognition tends to invent during silence, such as subtitle credits, are dropped.

What does the transcript look like?

Every line stays in the language it was spoken, with the speaker's name. Meeto does not translate the transcript, so nothing said in one language is replaced by a paraphrase in another, and a sentence that mixes languages — a Spanish question about an English product name, say — is kept as it was said. While AI summary runs, the transcript fills in as people talk; live captions over the call video are switched off for now.

After the meeting the transcript sits next to the summary, and both can be exported as Markdown, plain text or a PDF summary sheet, or shared by a link that the owner can switch off at any time.

Which language is the summary written in?

The summary is written in the language most of the meeting was held in. Meeto works that out from the transcript — which alphabet most of the text is written in and which common words appear most — and tells the AI model which language to write in, rather than leaving it to guess.

Two details make this hold up in real meetings. First, product names and code do not tip the balance: a Russian conversation about pull requests, APIs and dashboards is still a Russian conversation, even when many of its words are written in Latin letters. Second, the summary is checked when it comes back. If it arrived in a different language from the meeting, Meeto asks for it once more, naming the language again.

What happens to the parts in other languages?

They are part of the meeting, not noise. When some of a conversation was held in other languages, Meeto names those languages to the model and asks it to cover what was said in them as fully as the rest, written in the summary's language. A question a German-speaking guest asked in German appears in the English summary, and an action item someone accepted in French is listed with the others.

A few words on their own never count as another language: an “okay” or a product name in English does not make a Russian meeting bilingual. Only phrases long enough to show their language on their own are taken into account.

Tips for a multilingual meeting

• Let everyone speak the language they are most comfortable in; there is nothing to configure.
• When switching language, a full sentence switches immediately, while a single word may wait a moment for the next phrase.
• In a shared room, headphones keep one person's voice off another person's microphone and make the transcript cleaner.
• Read the summary before sending it. Check names, numbers, dates and who owns each action item, as you would with any AI-written notes.

Who can use AI summary?

AI summary is part of Pro and Team: 10 hours a month for each account or paid seat. After that a full transcript and a basic summary stay available until the allowance resets. The host starts it, it never captures video, everyone in the call is shown a notice and an indicator, and meeting content is not used to train AI models. It works inside Meeto calls only — no bot joins Zoom, Google Meet or Microsoft Teams. The pricing page has the current allowances, and the recording guide explains how recording and AI summary start together.