Voice Interviews for Qualitative Research: How AI Voice Interviewing Works (2026)
Written by Fengming Liu (UCL) · Reviewed by Prof. Shubin Yu (HEC Paris) · October 2026 · Updated: 2026-10-12 · 8 min read
Short answer: an AI voice interview lets participants speak their answers instead of typing them, in the browser, without an app. QualiTaTi transcribes each answer and asks follow-up questions from your interview guide. You can also let the AI speak its questions aloud, or run a real-time spoken conversation.
Four ways to run a voice interview
| Mode | Participant | AI interviewer | Credits per minute |
|---|
| Voice to Text | Speaks | Writes | 1 |
| Text to Voice + Text | Types | Speaks and writes | 1 |
| Voice to Voice + Text | Speaks | Speaks and writes | 1 |
| Realtime voice (experimental) | Talks freely | Talks back in real time | 5 |
A text-only interview also costs 1 credit a minute. The first three modes take turns, like a chat with sound. Realtime voice is a continuous spoken conversation that the participant can interrupt. It is newer and has its own link and settings.
When voice is the better choice
- Participants who find typing slow: older adults, people on phones, and people more fluent in speech than in writing.
- Topics where stories matter more than precise facts.
- Studies that need to hear how something is said, not only what is said.
When text is the better choice
- Sensitive questions. Schober et al. (2015) found that text interviews on smartphones produced more disclosure of sensitive information and more precise answers than voice interviews, with human and automated interviewers alike.
- Participants who prefer writing. In Lenzner and Höhne (2022), people unwilling to answer by voice most often said they preferred written communication. Willingness to use voice was highest among younger, frequent smartphone users.
- Noisy settings or shared rooms, where participants cannot speak freely.
When in doubt, pilot both modes with a few participants from your population. Our own comparison of voice and text AI interviews covers the trade-offs in more detail.
Setting up a voice interview
- Create an AI Interviewer project and write or generate your guide.
- In the project settings, choose the interaction mode. For realtime voice, open the Realtime Voice Mode panel and use its own link.
- Pick a voice and listen to the preview in your interview language.
- Turn on Record Audio if you want the recordings as well as the transcripts. Participants then see that the interview is being recorded.
- Run a test interview yourself on a phone and on a laptop before sharing the link.
What participants experience
- No app or account. The interview opens in the browser on a phone or computer.
- A microphone check before the first question, and a consent step on the entry screen.
- On iPhones, audio starts after the participant taps the start button, which the entry screen asks for.
Transcription and its limits
Every spoken answer is transcribed automatically, and the transcript is what you code. Automatic transcription makes mistakes, especially with names, accents, technical terms and background noise. Transcription errors and uneven follow-up quality are known limits of AI voice interviewers (Tirumala et al., 2025). Read through each transcript and correct key passages before quoting them.
Consent, recording and data
Tell participants in your information sheet that the interview is conducted by an AI, that their voice is transcribed, and whether the audio is kept. The ethics approval kit has text for each of these. Data is stored in the European Union. The trust page explains where audio and transcripts are processed and lists the providers involved.
Analysing voice interviews
Voice transcripts go into the same QDA workspace as text interviews. Code them by hand or with AI suggestions you review, invite a second coder, and export to NVivo, ATLAS.ti or MAXQDA with a .qdpx file if you prefer to finish there. See the NVivo hand-off guide.
Frequently asked questions
Do participants need to install anything?
No. A browser with microphone access is enough.
Which languages work by voice?
Voice interviews run in the interview language you set for the project. Preview the voice in that language before you launch, because voice quality varies between languages.
Is a voice interview more expensive?
Not in the three standard modes, which cost 1 credit a minute like text. Realtime voice costs 5 credits a minute, so a 30-minute realtime interview uses 150 credits. See the pricing page.
Are AI voice interviews as good as human interviews?
Early studies are encouraging but small. Wuttke et al. (2024) found AI-led interviews comparable in data quality to human-led ones in a student sample. Tirumala et al. (2025) conclude that suitability depends on the context. Pilot with your own population.
Sources, checked 2026-10-12: Schober et al. (2015), Precision and Disclosure in Text and Voice Interviews on Smartphones, PLOS ONE 10(6); Lenzner and Höhne (2022), Who Is Willing to Use Audio and Voice Inputs in Smartphone Surveys, and Why?, International Journal of Market Research 64(5); Tirumala, Jain, Leybzon and Buskirk (2025), Mic Drop or Data Flop? Evaluating the Fitness for Purpose of AI Voice Interviewers, arXiv:2509.01814; Wuttke et al. (2024), AI Conversational Interviewing: Transforming Surveys with LLMs as Adaptive Interviewers, arXiv:2410.01824.