How to run a multilingual meeting
A meeting in several languages works when you decide how people will understand each other before the call, not during it. This guide covers the options, a checklist and the mistakes we see most.
A multilingual meeting is any call where not everyone speaks or reads the same language well. It could be two colleagues who prefer different languages, a customer call across a border, or a town hall for a hundred people in six countries. The right setup is different for each, and picking the wrong one is the main reason these meetings go badly.
There are four ways to bridge the gap: one shared language, translated captions, AI spoken translation and human interpreters. This guide explains each, says when it fits, gives a step by step checklist and lists the mistakes to avoid. Product details and prices mentioned here were checked on 1 November 2026. Vendors change plans often, so read the linked page before you buy. For tool by tool comparisons, see Zoom vs AI vs human interpreters and the best live translation apps for meetings.
The four ways to bridge a language gap#
1. One shared language#
The cheapest option is to agree on a language everyone can handle, often English. It needs no software. It also puts the whole cost on the people who speak it less well. They talk more slowly, say less and miss jokes and detail. Use it when everyone is comfortable in that language, and be honest about whether they really are.
2. Translated captions#
Each person reads text in their own language at the bottom of the screen. Speech recognition turns the speaker's words into text, and machine translation converts it. Captions are quiet, cheap and leave a transcript behind. They suit technical meetings and people who read faster than they listen. They are weaker for people who cannot look at the screen, and they lag the speaker by a second or two.
3. AI spoken translation#
A listener picks a language and hears a synthetic voice say what the speaker said. Meaning has to form before it can be translated and spoken, so expect a delay of a few seconds. It works for talks and presentations where one person speaks at a time. It struggles when people interrupt each other, in noisy rooms and with names or industry terms. Quality differs a lot between languages.
4. Human interpreters#
Professional interpreters listen and speak in real time, usually from a remote booth. This is the standard for legal, medical, diplomatic and board meetings, and for anything where a wrong word has a cost. It is also the most expensive and needs planning: interpreters work in pairs, take turns, and want your slides and word lists in advance.
How to choose#
Ask four questions.
- What happens if a sentence is translated wrongly? If the answer is "someone laughs and asks again", captions or AI are enough. If it is "we sign the wrong contract", book a person.
- Who is listening? Readers are happy with captions. People on the move, older audiences and anyone with low vision may need a voice.
- How many languages? Two languages is easy. Six is a logistics problem for humans, because each language needs its own interpreter. For machines, six costs about six times as much as one.
- How often? A weekly team call needs something that costs almost nothing per use. A yearly conference can pay for specialists.
A common mix is AI for routine internal calls and human interpreters for the few events that carry risk. Some platforms let you do both in the same meeting. If you want the exact trade-offs and prices, read the Zoom vs AI vs human interpreters comparison.
A checklist for the week before#
Work through this list at least a few days ahead, not on the morning of the meeting.
- Fix the language list. Write down which languages are spoken and which are only listened to. Check that your tool supports all of them. Language counts are marketing numbers until you check the pair you need.
- Check what each plan includes. Translation is often a paid add-on. A host on a free plan may not be able to switch it on. Zoom, Google Meet and Microsoft Teams all tie translated captions or speech to specific paid plans, as we describe in live translation for meetings.
- Name the speaking language per person. Most tools ask each speaker to set the language they talk in. If that is wrong, everyone else gets wrong text.
- Send a short glossary. Product names, people's names, acronyms. Give it to interpreters, and use it yourself to check what the AI does with them.
- Ask for the slides early. Interpreters need them. Captions read better when the speaker follows the slides.
- Run a two minute test with a real speaker of each language, on the same devices and connections people will use.
- Decide who moderates. One person watches the language channels and the chat, and tells the speaker to slow down when needed.
How to speak so that it works#
Most of the quality comes from the speaker.
- Speak in short sentences and pause at the end of each one.
- One person at a time. Overlapping speech is the failure mode of every system we know.
- Avoid idioms, sarcasm and jokes that depend on a pun. Say "we are behind schedule", not "we are up the creek".
- Spell out names and numbers once, then repeat them.
- Use a headset. A good microphone beats any setting.
- Share the slides and write key numbers in the chat so nobody depends on hearing them.
Listeners can help too. Keep captions on even when you hear a voice, so you can check a number or a name. Put questions in the chat when the speaker is going quickly.
Setting it up in a few common tools#
Zoom. Interpretation by humans is a scheduled meeting feature, and captions can be translated on some plans. See our Zoom comparison for the detail.
Google Meet and Microsoft Teams. Both offer translated captions on specific paid plans, and Google offers spoken translation on eligible plans. Our translation roundup has the plans and limits as read from their pages.
RespireSoft Meet. The free live interpreter lets two or more people try spoken translation without an account. It supports five languages today, English, Spanish, French, Arabic and Persian, in a room of up to 4 people, with free minutes every day. The live captions tool does the same for captions. For your own product, the API offers captions and spoken translation in a room. Room translation languages are English, Dutch, German, French, Spanish, Arabic, Persian, Pashto, Turkish and Urdu, and a room can use at most 5 at once. That list is short next to what the big vendors claim, so check that your languages are in it. The call screen itself comes in 18 languages, which is a separate thing. Developers can read the captions and translation guide.
Mistakes we see most often#
Treating machine translation as a record. A caption that was wrong for two seconds is usually harmless, but a transcript can end up in minutes and emails. Have a person read it before it is shared.
Skipping the test. The first time you hear that the Persian speaker's microphone is silent should not be in front of the client.
Counting languages, not pairs. A tool may translate from English to 50 languages and not between Dutch and Arabic. Check the pair.
Forgetting the people who cannot read the screen. Offer a voice or an interpreter for anyone who needs it, and ask in the invitation.
Putting a long agenda on a translated call. Translation roughly doubles the time each point takes. Cut the agenda or add time.
Ignoring privacy. Captions and translation send speech to speech services. If the meeting has personal or confidential content, read what the vendor says about where audio goes and whether it is kept. For a deeper look, see GDPR and EU hosted video conferencing.
Questions people ask#
Can a meeting be fully automatic? For low risk talks, yes. For anything with legal or medical weight, use a human interpreter, and use AI as a backup for the languages you cannot staff.
Is there a free way to try this? Yes, small tools exist. Ours has a short daily allowance. Small tools like it are good for a quick conversation and not for a conference.
How long is the delay? A second or two for captions and a few seconds for a voice. Every vendor warns of a lag, so plan for it.
What if someone speaks two languages in one sentence? Most systems are set to one speaking language per person and will stumble. Ask speakers to stay in one language per turn.
Do we need a transcript? If you do, turn transcription on before the call starts. Many tools do not save captions unless transcription is enabled.
If you build meetings into your own product, the docs explain how rooms, tokens and translation fit together, and the pricing page lists per minute prices. Billing is not switched on yet, so you are not charged for usage today, and those prices may change.