Skip to main content

​
Answering in your own language

Respondents can complete a voi study in their own language. Open-ended answers are translated automatically, so you can analyse every response in one place regardless of what language it was given in; the original answer is preserved alongside the translation. Translation runs through the same AI gateway as the rest of the analysis; see Which AI models process your study’s data?

​
Low literacy and audio reading

Voice and video response modes exist so respondents don’t need to type an answer: they can speak or record a short video instead, for any open-ended question. For a subset of languages, voi can also read the question aloud through an audio reader, so a respondent never has to read the question text either.

​
Language support at a glance

Rather than a flag, which stands for a country rather than a language, each one is shown in its own name alongside English.

​
How accurate is transcription and translation?

Accuracy is highest in English, close behind in other widely used digital languages, and lower in languages with a smaller presence online, since the models behind transcription and translation learn from what’s available to train on. It’s worth weighing when you choose a language for a study: for a less widely used language, spot-check a handful of transcripts in Review before you trust the analysis that follows.

​
Closed vs. open questions

Closed questions (scales, single or multiple choice, NPS) are always answered directly, regardless of language or literacy. It’s open-ended questions where the response format (text, voice, or video) matters, and respondents can pick whichever is most comfortable for them.

​
When someone can’t answer unassisted

Language and literacy support removes many barriers, but not every respondent has a reliable connection, a device, or the digital comfort to complete a study alone. See Collection Methods for how voi supports an enumerator running the interview on someone’s behalf.

​
Plan for this before you launch

Confirm your respondents’ languages are covered before you launch, and check whether you want the audio reader available for low-literacy respondents using the table above, since not every supported language has audio-reading support. It’s easier to catch a gap here than after a low response rate tells you something didn’t work.