Data handling

    What happens to a respondent’s voice answer

    A plain description of how Voice Capture handles audio and transcripts, written for research and compliance teams. It describes how the product works today.

    When recording starts

    • Nothing is recorded until the respondent presses the record button. The browser then asks for microphone permission.
    • Next to the button the widget shows a short note in the survey language: “Your voice is not stored — only the transcribed text”.
    • The respondent can always type instead: the text box stays available.
    • Voice Capture does not collect a separate consent statement. If your study needs one, put it in your survey.

    What happens to the audio

    • The audio is sent over HTTPS to the Voice Capture API and held in memory, not written to disk or to file storage.
    • It is sent to a speech-to-text provider: Groq (Whisper Large v3 Turbo) as the primary path, OpenAI (gpt-4o-mini-transcribe) as the fallback.
    • The audio is processed in memory and never written to storage, also when transcription fails. Only the text is saved. Voice Capture holds no audio in storage today.

    What is stored

    • The transcript text, or the typed text if the respondent typed.
    • Response details: project, session ID and question ID sent by the survey, detected language, duration, file size and timestamps.
    • The domain of the survey page (origin only, not the full URL) and the survey platform detected from it.
    • Which transcription provider was used and how long it took.

    Where it is stored

    Transcripts and response details are stored in a Supabase (PostgreSQL) database in the AWS eu-central-1 region, in Frankfurt, Germany.

    Vendors that touch audio or text

    Voice Capture does not store the audio, but the transcription providers apply their own terms. Per OpenAI’s API terms, abuse-monitoring logs are kept for up to 30 days and data sent to the API is not used to train its models. Per Groq’s terms, inference requests are not retained by default, and reliability and abuse-monitoring logs are kept for up to 30 days. Groq may not use inputs or outputs to train models unless the customer allows it.

    • Groq and OpenAI: receive the audio to transcribe it.
    • Supabase: stores transcripts, response details and account data.
    • Railway: hosts the Voice Capture API that receives the audio. Insight Genius LLC and Railway have a signed data processing addendum, effective 9 October 2026.
    • Resend: sends account emails. The notice of a project’s first response includes the text of that transcript.
    • Anthropic: only when an account user asks the dashboard to generate a transcription prompt, up to 15 recent transcripts of that project are sent to write it. The help chat also uses Anthropic.
    • Stripe: payments only. It receives no survey data.

    OpenAI: your data · Groq: your data · Groq: services agreement

    How long it is kept and how to delete it

    • Transcripts are kept until you delete the project or your account.
    • Deleting a project deletes all of its responses.
    • Deleting your account (Settings, confirmed through a link sent by email that expires in one hour) deletes your projects and their responses.
    • To ask for specific responses to be deleted, or for anything else on this page, write to us.

    Contact: hello@insightgenius.io

    This page describes technical facts about how the product works. It is not a legal certification and does not state compliance with any specific law, such as the LGPD or the GDPR. Your legal team decides how it fits your obligations.

    Related documents: Privacy Policy · Data Processing Agreement