The short answer
You can use AI transcription for patient interviews when your ethics approval and your organisation allow it, patients have given informed consent to recording and transcription, and the service processes data in the EU under a data processing agreement without training on it. Because health data is involved, expect your organisation to ask for more: often a data protection impact assessment and a named, approved tool.
Why patient recordings need extra care
Data concerning health is a special category of personal data under GDPR Article 9. Processing it is prohibited unless an exception applies, such as explicit consent (Article 9(2)(a)) or scientific research with appropriate safeguards (Article 9(2)(j) together with Article 89). National law adds its own rules, and clinical settings add professional confidentiality on top.
A patient interview often contains far more than the topic you asked about: diagnoses, medication, family situations, other people’s names. Assume the whole recording is health data, not only the parts you plan to analyse.
Processing health data at scale or with new technology often requires a data protection impact assessment (Article 35). Ask your data protection officer early, since this can take weeks.
Consent comes first, every time
Never record a patient without clear, documented consent. Patients must know that the conversation is recorded, that an automated service will transcribe it, where the data is processed, how long it is kept and that they can withdraw without any effect on their care.
- Make it clear that taking part is voluntary and separate from treatment.
- Get consent from everyone present, including relatives or interpreters.
- For children or adults who cannot consent themselves, follow your approved procedure for guardians or legal representatives.
- Repeat the key points on the recording itself at the start: "Are you still happy for this to be recorded?"
- Record consent separately from the interview audio so you can delete the interview without losing the consent record.
Research interviews vs clinical documentation
This guide is about research: interviews with patients as study participants. Recording clinical consultations to write medical records ("ambient scribe" tools) falls under healthcare rules, your organisation’s medical record system and often medical device regulation. Do not use a research tool to create clinical documentation, and do not reuse clinical recordings for research without the right approval.
A safe workflow for patient interviews
- 1. Approval: name the transcription tool in your ethics application and data management plan. Check if a DPIA is needed.
- 2. Recording: use an encrypted recorder or an approved device. Avoid personal phones that sync to consumer clouds.
- 3. Upload: transcribe with an EU-processed service that does not train on your data, with redaction of names and identifiers switched on if available.
- 4. Verify: check the transcript against the audio. Medical terms, drug names and numbers need extra attention.
- 5. Pseudonymise: replace remaining identifiers, including clinic names, rare diagnoses and dates that could identify someone.
- 6. Restrict access: share pseudonymised transcripts only with named team members.
- 7. Delete: remove recordings on schedule, and keep a log of the deletion.
How Kahubi supports patient interview studies
Kahubi transcribes with European providers and never trains on your recordings or transcripts. You can switch on redaction so names, health details and other identifiers are masked in the transcript, and you can label speakers as Interviewer and Patient straight away.
Analysis happens in the same project, so transcripts do not need to be copied into other tools: framework or thematic analysis with suggested codes you confirm, and a findings draft with quotes linked to the transcript. Redaction is a help, not a guarantee, so always read the transcript before sharing it.
What AI transcription changes for clinical researchers
Manual transcription takes around four to six hours per hour of audio. For a typical study of 20 patient interviews that is several weeks of work, often done by the busiest people on the team. AI transcription turns that into minutes of processing and a few hours of checking, so analysis can start while the interviews are still fresh, and themes can feed back into the next interviews.
Sources
Last updated 2026-10-09.