Transcribing research interviews: consent, anonymity, and method

What to settle with the ethics committee first

Most universities and research institutes require an ethics review for any study with human participants. The committee will want to know how you plan to record, transcribe, store, and share your interviews. If you use automated transcription, say so in the application, name the tool, and state where the audio is processed. Every institution has its own rules, and the final call is theirs. What follows here is general guidance, not official approval or legal advice.

Research ethics typically rest on three principles, laid out concisely in the widely cited Belmont Report: respect for persons, a favorable balance of benefits and harms, and fair selection of participants. Day to day, this means making sure people understand what you are asking of them, keeping them from harm, and giving them a clear way to pull out of the study.1

Draft your data management plan before you record anything. Write down who touches the audio, where the files live, who can read the transcripts, when the recordings get deleted, and what happens to the anonymized text once the project wraps up. Putting this in place ahead of the first interview is much easier than piecing it together after the fact.

What the consent form should say about recording

Be explicit on the consent form that you will record the audio and transcribe it, and mention if software is involved. Spell out who can access the files, how you will obscure identities, how quotes or details will show up in print, and how long you will hold on to both the audio and the text. You also need to explain how someone can withdraw, and whether they can still do so once you have started analyzing the data.

Break the permissions down so participants have real choices. Ask separately for consent to record and consent to be quoted, even if the quotes are anonymous. Make it clear that you will pause or stop the recorder whenever they ask. People often say more on the record when they know they can go off it.

How to record for a usable transcript

Find a quiet room and put the microphone close to the participant, not in the middle of the table. Before you begin the interview, record a one-minute test and play it back. Save every interview as its own file, and name it with an anonymizing code right from the start instead of the participant's name.

Only ask the participant to say a first name or pseudonym on tape if your data plan allows it. If not, just read out the interview code yourself at the start. Try not to talk over the other person, and let silences hang. Silence counts as data too, and two voices talking at once are hard to untangle when you transcribe.

When an interview jumps between languages, like Hebrew, Arabic, and English, jot that down in your research log before you transcribe. Automated tools need the closest checking on multilingual audio.

Which transcription style to choose

Pick the transcription style that fits how you will analyze the data. True verbatim leaves in every hesitation, repetition, false start, and filler word, which makes it suitable for discourse and conversation analysis. Clean verbatim suits most thematic analysis because it keeps the exact words spoken while stripping out fillers and false starts. Summaries are only useful for a first pass. Whatever convention you choose, write it down in your methods section.

Set your rules early for how you will mark pauses, laughter, cross-talk, and mumbled bits you cannot make out, then stick to the same notation throughout. Automated transcription usually gives you something close to clean verbatim. If your method calls for finer detail, you will have to listen back and type those markers in by hand.

Leave the speaker labels and timestamps in place. Examiners and peer reviewers sometimes ask to check a quote against the raw data, and timestamps let you jump straight from a line in your draft to that exact second in the audio.

How to anonymize interview material

Swap out the names of people, places, workplaces, and institutions for consistent codes or pseudonyms, and keep the key that links them to real identities in a separate, secure file. Do not stop at proper names. An uncommon profession, a small town, or a distinct event can still give someone away after their name is gone. If you are studying a small community, think about whether certain descriptions are safe to publish at all.

A voice recording is biometric data, since the voice itself identifies the speaker. Lock audio files down more tightly than text transcripts, and delete them on the schedule you promised in your data plan. Check where your transcription software processes audio, and talk to your data protection officer before uploading sensitive material. KolWrite lists its setup on its security page: encrypted connections, AES-256 storage, servers in Germany, named AI providers, and a policy of never training AI models on customer files.

How to verify the transcript and prepare it for coding

Play the whole recording from start to finish while reading along, and fix errors as they come up. Automated transcripts always miss things, and in research a single misheard word can skew your findings. In KolWrite, you can upload the recording, get a draft with speaker labels and timestamps, fix text and speaker names right in the browser, and export to TXT, DOCX, or JSON for your coding software.

Look closely at negations, numbers, organization names, technical terms, and any sections where the speaker switches languages. If you plan to quote a passage directly, listen to it a second time and check the quote in its full context, not as an isolated fragment.

Once the text is checked, pull the clean file into your qualitative coding software or code it in a word processor. Keep your file versions straight so the transcript you code is the exact one you can hand over if someone asks to review your data.