Interview transcript example: a worked research interview

This is the first two minutes of a semi-structured interview, transcribed clean verbatim with timestamps at each turn. The participant has been given a pseudonym. Read it first; the reasoning for each choice is underneath.

[00:00:08] INTERVIEWER: Before we start properly — you are happy for this to be recorded, and you can stop at any point?

[00:00:13] P04: Yes, that is fine.

[00:00:15] INTERVIEWER: Thank you. So, tell me about the first week after the ward reopened.

[00:00:22] P04: The first week was strange. Everyone had been told it was going to be the same as before, and it was not the same as before. The rota was different, the agency staff were different, and nobody had actually said that out loud. [pause] So you had a lot of people being quietly annoyed about something they had not been told had changed.

[00:00:51] INTERVIEWER: When you say quietly annoyed — what did that look like?

[00:00:56] P04: People not staying for handover. Which sounds small. It is not small. Handover is where you find out the things that are not written down.

[00:01:11] INTERVIEWER: Did anyone raise it?

[00:01:13] P04: [inaudible 00:01:14] raised it with the ward manager, I think in the second week. I was not in that conversation so I only know what I was told afterwards.

The consent exchange is in the transcript, not cut from it. If your ethics approval says consent was recorded, the recording has to show it. Cutting the first thirteen seconds because they are administrative removes the evidence that the rest is usable.

The participant is P04, not a name and not Participant 4. A short code is easier to scan across a set of twenty transcripts and harder to accidentally leave in a quotation. Keep the mapping from code to person in a separate file, stored separately.

The interviewer is INTERVIEWER, not their name. If several researchers ran interviews and it matters who, use INTERVIEWER-A and record the mapping elsewhere, for the same reason.

The [pause] at 00:00:22 is kept because the sentence after it changes direction. A pause that is just someone drinking tea is not marked. Marking every pause makes the marked ones meaningless.

The [inaudible 00:01:14] is a name that could not be made out. It is marked with its timestamp rather than guessed, because a guessed name in a transcript about a workplace complaint is the kind of error that follows people around.

For thematic analysis, clean verbatim is almost always right. You are coding what people said, and the ums get in the way of seeing it. Removing fillers is not editorialising as long as no sentence is rewritten.

For conversation analysis, discourse analysis or anything where delivery is the object of study, you need full verbatim and probably more — overlap markers, timed pauses in tenths of a second, stress and intonation. That is a different transcription system, and you should decide you need it before you start, not after.

For journalism, clean verbatim with timestamps. When you quote, quote what the transcript says. If you tidy a quote for print, the transcript is still the record of what was actually said, which is the point of having one.

Whatever you pick, use it for every interview in the set. A study where half the transcripts keep fillers and half do not cannot be compared line for line.

Names are the obvious one and the easy one. Replace them consistently: the same colleague should be [COLLEAGUE-1] everywhere, so the reader can follow who is who without knowing who they are.

The ones people miss are the identifying details that are not names: a job title only one person holds, a ward or department, an unusual illness, the date someone started, a town small enough to narrow it to a handful of people. In a small organisation, the woman who came back from secondment in March is a name.

Replace in square brackets with a description, not with blanks: [PARTICIPANT'S MANAGER] reads better than [REDACTED] and keeps the sentence usable as a quotation.

Do the anonymising pass on the transcript, then delete the audio according to whatever your approval says — and remember the automatic transcription service you used may still hold a copy. Check its retention settings before you upload anything sensitive, not after.

Machine transcription first, then a checking pass against the audio, then the anonymising pass. Doing it in that order means you are never retyping and never anonymising text you are about to delete.

Budget roughly a third of the recording length for the checking pass on clear two-person audio, and considerably more for group interviews or a strong accent the model has not heard much of. The video below covers the mechanical part.