Photo by Vitaly Gariev on Unsplash
There is no official interview transcript format. What exists is a small set of parts every usable transcript has, plus a house style you write down and stick to. Get those right and the file needs few changes for a dissertation, a client or a data archive. Below is the format I'd start from, with a template you can copy.
I'm writing this mainly for people learning to transcribe for pay. If you're a researcher, the template works for you too, though your supervisor or ethics board may add rules of their own. And when a client hands you a style guide, follow it over anything here.
The interview transcript format in one template
Interview ID: INT-007
Date: 2026-03-14
Setting: Video call
Speakers: I (interviewer), P7 (participant)
Recording length: 00:46:12
Conventions: Clean verbatim. [inaudible hh:mm:ss], [crosstalk], [laughs]
[00:00:04] I: Thanks for joining. Can you start by telling me about your role?
[00:00:09] P7: Sure. I've been a team lead for about three years, mostly on the support side.
[00:00:17] I: What does a normal week look like?
[00:00:20] P7: Busy. [laughs] Mondays are the worst, because [inaudible 00:00:24] comes in over the weekend.That's a fictional example, and it's everything most projects ask for: a header, labelled turns, time codes, and a way to mark what isn't plain speech.
Start with a header that lets the file stand alone
A transcript gets separated from its folder sooner or later. The UK Data Service's guidance on documenting data says each text file should carry enough context to stand on its own, and lists a minimum: a unique identifier, the date of collection, and the setting where relevant. Its model transcription template is worth a look if you want a second example.
Keep the header at the top, separate from the conversation, and leave real names out of it. Use the same ID in the file name, something like INT-007_clean_v1.docx, so nobody has to open a file to know what it is.
Label speakers the same way all the way through
Every turn starts on a new line with a speaker label. Interviewer and Participant 1 are fine. So are I and P1. What matters is that you pick one and never drift. The UK Data Service asks for speaker tags to be "applied consistently throughout the file", and ATLAS.ti's formatting guide gives the same advice: each speaker must be clearly identified. If "Interviewer" turns into "Int" on page nine, a search for the label misses half the file.
Researchers often want numbered lines too, so a quote can be cited exactly. The guide on transcribing interviews at Sustainability Methods recommends them for that reason. Add them last in Word, and share a PDF if people will cite them, because the numbers shift when the layout changes.
Time codes: how many do you need?
A time code is there so someone can get back to the audio. One at each speaker turn, in hh:mm:ss, is the safe default and the one I'd use unless told otherwise. Long answers benefit from an extra one every minute or so. Fewer than that and checking a quote turns into scrubbing through the recording, which is exactly what the time code was meant to prevent.
Mark what isn't plain speech
Square brackets tell the reader you stepped outside the speaker's words. Keep the set small, and list it in the header so the reader knows what each one means:
- [inaudible 00:12:40] when you can't make it out. Always add the time.
- [crosstalk] when people talk over each other.
- [laughs] or [pause 4 sec] when it matters to the meaning.
- [company name] in place of anything you've been told to remove.
Never guess at an unclear word and leave it unmarked. A marked gap costs a reviewer a few seconds. A confident wrong word can end up quoted in someone's findings.
Full verbatim or clean verbatim?
This is the decision that changes your typing most, so here's the same sentence both ways:
Full: So, um, I, I think we, like, we kind of stopped using it, you know, after the, the update.
Clean: I think we kind of stopped using it after the update.Clean verbatim removes fillers, repeats and false starts and keeps the speaker's own words. "Kind of" stays, because it softens what they said. Full verbatim keeps everything, and researchers studying how people talk usually need it. Our page on clean verbatim transcription has more examples and a practice worksheet, and this guide to verbatim transcription covers when each one fits a research project.
Save it in a format people can use
Word is the safe delivery format, because clients can comment on it and most analysis software can import it. Plain text works when someone will process the file by script. Subtitle files like SRT are built around timing blocks, so they're the wrong choice for a transcript someone needs to read and quote, unless that's what the client asked for.
If you're starting from the recording, not the format, our step-by-step guide on how to transcribe an interview covers the rest of the job.
A last check before you deliver
Listen to the whole file once with the transcript open, and read for sense, not spelling. Then search for your label variations, confirm every [inaudible] has a time, and check names of people and places against whatever notes you were given.
If you enjoy getting this kind of detail right, it's a skill people pay for. Create a free Spirelight contributor profile on our AI training jobs page to be considered for transcription, recording, annotation and checking projects. They open in waves, so joining doesn't guarantee immediate work, but you'll be on the list when one fits.