Does voi record a respondent’s voice or video?
Only if they choose that way to answer. Text, voice, and video are all offered for open questions, so people can respond however they’re comfortable. A recording is stored in private cloud storage on EU servers. The recording is kept as part of the study results so your organisation can play it back, and it can appear in a report if the respondent agreed to that use. At the end of the interview, a respondent can still hold the recordings back: if they choose that, only the transcripts are submitted and the audio and video files are never sent.Does an AI ever listen to or watch a recording?
The recording goes to one automated transcription service, which turns speech into text and returns nothing else. The models that translate answers, generate follow-up questions, or find patterns across responses work from that text and never receive the original file. Transcription runs without a person listening in, though your organisation can play back recordings in its own workspace. voi runs no emotion recognition and no voice-based inference of any kind; every model downstream works only from the transcript.Which AI models process your study’s data?
Text answers, translations, follow-up questions, safety screening, theme clustering, analysis summaries, the Explore feature, and report drafts are handled by AI models reached through Opper, an EU-based AI gateway operated by Opper Technology AB in Sweden. Only voice recordings go to ElevenLabs for transcription, and ElevenLabs also receives question text when a question is read aloud. The two paths stay separate: the language models never receive raw audio or video, and the transcription service never receives anything beyond the recording and the question text. See AI models & providers for what an AI gateway is, which companies are behind it, and how they’re chosen.Is study data used to train AI models?
No. voi operates on a no-training basis with every provider it works with. None of them use your study data to develop, train, or improve any AI model, and that obligation is written into voi’s contract with each provider. The AI gateway keeps request and response content for 14 days inside the EU to trace processing errors and monitor output quality, after which it’s deleted. Separately, leonardo derives anonymized statistics from structured results such as indicator values and closed-ended answers, and uses them for benchmarking and to improve its methods. Free-text answers, audio, and video are never part of that, and the statistics relate to no identifiable person or organisation.Will a respondent’s voice or video appear in a report?
It can. Reports can carry direct quotes and short clips from recordings, and your organisation decides who receives that report, sometimes including a funder. What respondents are told is the consent message itself: your organisation, as the data controller, declares there how their data will be used. The stronger safeguard sits at the point of recording: if a respondent answers by voice or video, they explicitly consent that the recording can be used for reporting before it is ever collected. Answering by text keeps recordings out of a report entirely. See Report for how consent shapes what a report can include.This section explains voi’s approach to data protection. It isn’t legal advice, and it doesn’t replace your organisation’s own privacy policy or the Data Processing Agreement you accept when you create a workspace.