How to Choose Which Languages to Offer

Nearly every organization we talk to picks its languages the same way: someone looks at a membership list, a census figure or a registration form, names the three biggest groups, and that becomes the plan. It is a reasonable instinct, and it is wrong in both directions at once.

It over-counts, because a language somebody speaks at home is not necessarily the language they want your talk in — plenty of people follow the room perfectly well and would find a translated voice in their ear more distracting than helpful. And it under-counts, because the people most likely to use translation are the ones least likely to be on your list: the visitor, the new arrival, the family who came because a friend brought them.

There is a better method, and it works because of a property of the pricing that is easy to miss.

Being wrong in one direction is free

A translation only starts when its first listener arrives. Enable a language, have nobody select it, and no translation session opens and nothing is charged for it. That single fact changes the whole exercise: you are not making a forecast you have to defend, you are opening a door and seeing who walks through it.

One exception, and it is worth knowing before you plan around this. If you switch on full-session recording, the first language in your list is translated for the whole broadcast whether or not anybody is listening — that is the only way a recording of it can exist. With recording on, put a language you are sure of first.

The other direction is the expensive one, and the cost is not money. A person who came, could not follow, and left does not fill in a survey about it. You never learn what that language was worth, because the evidence walked out.

So the rule is: offer the language you are unsure about. The downside of a guess nobody uses is nothing at all.

What you can actually turn on

A single broadcast carries up to five output languages. That is a firm limit, not a plan tier, so it is worth using them deliberately. Two things are worth knowing before you do:

• Captions in the language being spoken. A separate switch, and the one people forget. It is not for translation at all — it is for anyone in the room who is deaf or hard of hearing, and for the person at the back who simply cannot make out the words. See accessibility and live captions for what it does and, just as importantly, what it does not replace.
• A language a human interpreter covers still takes a slot.An interpreter takes one output language and speaks it themselves, and no AI runs on it while listeners stay with them — but it is still one of the five, because the audience needs a channel to listen on. Five is five whether the voice on a language is a person or a model.

The full catalogue is on the supported languages page, written in each language's own script, so you can look for the one you need rather than take our word for how many there are.

A method that uses evidence instead of guesses

1. Ask, but do not only ask

Asking is cheap and worth doing. It is also unreliable on its own: people under-report needing help, especially in front of a group, and especially about language. Treat the answers as one input among several, not as the answer.

2. Start with three or four, not one

Pick the two you are confident about and add one or two you are merely curious about. The curious ones are free unless somebody uses them, and they are the only way you will discover a need nobody told you about.

3. Run it three times before you conclude anything

The first time you offer translation, the numbers measure how well you advertised it, not how much it was needed. Half the room does not know the code exists. Run it a second and third time, with the link somewhere people will see it, before you read anything into the figures.

4. Then read the numbers properly

During a broadcast the console shows how many people are on each language right now. Afterwards, each session records how many listener-minutes every language used, so a language's audience is a fact rather than an impression. Two things to watch for:

• Minutes flatten a short burst. A language with four listeners for twenty minutes of a two-hour service looks small beside one with a steady trickle throughout — and those twenty minutes may have been the entire reason a family stayed. Glance at the live counts during the event too, not only the totals afterwards.
• Zero over three sessions is a real answer. One zero is noise. Three is a language you can retire without guilt — and turn back on the week you have a reason to.

5. Prune, then re-spend the slot

Five is the cap, so retiring a language is what buys you the chance to test the next one. This is the step people skip: the list gets set once and then never revisited, and two years later it still reflects who was in the room the month somebody set it up.

Two things to watch out for

Quality is not uniform. Widely spoken languages are the strongest; a language with less training material behind it will be less fluent, most visibly on names, places and specialist terms. That last part you can fix yourself — see fixing names and terms with a glossary. But do not promise a smaller language the same polish you would expect in Spanish — and if it matters enough, that is the language to put a person on.

Printed material outlives your language list — if you print the right address. Use your organization's permanent link on a poster or a programme, not a session code: a session code belongs to one broadcast and resolves to an "ended" page afterwards, while the permanent link points at whatever you are broadcasting now. The languages behind that link are whatever you have enabled on the day, which is a feature — you can change the list without reprinting — but it does mean the poster should not name specific languages unless you intend to keep them.

The short version

• Offer the language you are unsure about — an unused language costs nothing.
• Spend the five slots deliberately; captions in the spoken language are free of them.
• Run it three times before you read the numbers.
• Read peak, not average, and treat three zeroes as a decision.
• Retire something before you add something.

For how this fits a specific room, see live translation for churches, live translation for conferencesor live translation for events. For what each language costs while somebody is listening to it, see the pricing page.