Skip to main content

Transcribe interviews and research sessions

Interviews raise the bar on every part of transcription. Consent has to be documented, quotes have to be exact, and participants have to be protected. AI removes hours of mechanical labour per session — inside a disciplined process.

What changes

  • A first-pass transcript in minutes, not hours

    Manual transcription runs at roughly four to six times the length of the audio. An automatic first pass turns that into a verification task, which is a fraction of the effort.

  • Cross-interview retrieval

    Find every participant who mentioned a concept across dozens of sessions, including the ones who used different words than you would have searched for.

  • Negative case search

    Actively look for participants who contradict an emerging theme. This is the most valuable and least used capability in AI-assisted qualitative work, because it counteracts confirmation bias.

The workflow

  1. 1

    Get written consent covering AI processing, then restate it in the first seconds of the recording.

  2. 2

    Record close to the participant, with a backup device running.

  3. 3

    Take the automatic transcript as a draft, then verify every quote, number, name and attribution.

  4. 4

    Anonymise direct and indirect identifiers, then code and query the verified corpus.

Frequently asked questions

Is AI transcription rigorous enough for research?

As a drafting tool, yes. As a final artefact, no. The dangerous property of AI transcripts is fluency: a wrong word arrives in a grammatical sentence with no confidence signal attached. Every quote you publish should be checked against the audio.

Should research transcripts be full verbatim?

It depends on your analytic approach. Conversation and discourse analysis need pauses, overlaps and false starts preserved. Thematic analysis usually works with cleaned verbatim that removes filler while preserving exact wording. Decide before you start and apply one standard across the whole corpus.

How do I anonymise transcripts properly?

Replace names with consistent participant codes, then generalise indirect identifiers such as employer, job title, location and unusual details that could re-identify someone in combination. Keep the code mapping in a separate access-controlled place, and do a human pass, because find-and-replace misses nicknames and inflections.

Go deeper

Other ways people use it

Start recording this week

myvoicenotes.ai records anything in one tap, transcribes it automatically, and answers questions across everything you have recorded. Join the waitlist for early access.