Focus Group Transcription: How to Record, Transcribe and Label Speakers
Dzianis Kukareka, Founder · Updated

Focus group transcription means turning a recorded group discussion into a written, speaker-attributed text you can code and quote. The fastest reliable way: record with a central microphone, run the file through AI transcription to get the words and draft speaker labels, then spend your own time checking who said what in the passages you will analyse. AI gets the words of a focus group right far more often than it gets the speakers right, so the whole workflow is built around verifying attribution.
This guide covers how to record a group so it transcribes well, a step-by-step workflow, what a finished focus group transcript looks like, and what human services and software cost.
How do you transcribe a focus group?
The short version, in order:
Record for transcription, not just for memory. One good table microphone in the middle of the group, a quiet room, a voice roll call at the start.
Get a machine draft. Upload the recording to a transcription tool with speaker labels and let it produce the words and a first guess at speakers.
Map labels to participants. Use the roll call and the moderator's name-checks to match "Speaker 3" to a real seat and participant code.
Verify the quotes that matter. Listen back to every passage you will code, quote or report, and fix attribution and mishearings there.
Mark what cannot be resolved, such as overlapping voices, instead of guessing.
Anonymise and export to your analysis tool with participant IDs.
Typing it all by hand is still an option, but it is slow. A qualitative methods textbook from Rutgers notes that transcribing a 60-minute focus group discussion can take up to 10 hours. With a machine draft, those hours shift from typing to checking.
Why focus group transcription is harder than an interview
A one-to-one interview transcript has two voices that rarely overlap. A focus group has six to ten, and the same Rutgers chapter gives 6–10 participants as the usual target size. That changes the problem in three ways.
Speaker labelling gets harder with every voice. Speaker diarization, the step that decides who spoke when, clusters voices by their acoustic features. Similar voices (same age, gender and accent, which is common when you recruit a homogeneous group) get merged into one speaker or split into two.
Overlap is normal. Agreement noises, interruptions and shared laughter put two or three voices in the same second. A transcript is a single line of text, so something has to give.
The microphone is far from most people. Participants at the far end of the table reach the mic through room echo, and the same person can sound different when they lean back.
The distance effect shows up in published benchmarks. The open-source pyannote diarization model reports a diarization error rate of 18.8% on the AMI meeting corpus when using headset audio and 22.4% on the same meetings recorded with a single far-field array channel, with overlapping speech counted. Those are small, four-person meetings. A larger, livelier focus group recorded on one room mic sits at the harder end of that range, while the word layer usually holds up much better. If you want the background on word accuracy, see why AI transcription accuracy varies.
How to record a focus group for a clean transcript
Most transcript quality is decided before anyone presses upload. These habits cost nothing during the session and save hours afterwards.
Use a central boundary or conference mic. Put it in the middle of the table, not in front of the moderator. A phone at one end of a long table captures the near half of the room and the echo of the rest. If you can, add a second recorder at the other end as a backup track.
Choose a soft room. Carpet, curtains and upholstered chairs absorb echo. Glass-walled meeting rooms and empty halls make every voice blur.
Open with a roll call. "Let's go round the table: first name and the last product you returned." Twenty or thirty seconds of each participant speaking alone gives you a reference sample for mapping speaker labels later.
Have the moderator say names. "Thanks, Maria. David, you were shaking your head?" Each spoken name is an anchor in the transcript for fixing labels around it. It is good moderation anyway, because it draws in quiet participants.
Set one ground rule about overlap. "One voice at a time, please, we're recording." Repeat it gently when the discussion heats up.
Keep a seating map and a note-taker. An assistant moderator who notes who spoke at key moments ("P04 strong objection on price, about 40 min") turns ambiguous passages into solved ones during checking.
Get consent on the record. Announce the recording and its purpose at the start and capture each participant's agreement per your ethics protocol. Rules differ by country and setting; our guide to recording consent in the US and EU covers the basics.
If the method allows it, smaller groups transcribe better. Each extra participant adds another voice to separate and more chances of overlap.
Transcribing focus group data step by step
Upload the best track. If you recorded two devices, listen to a minute of each and use the cleaner one. Long recordings are fine for most tools; check the file size limit before a 2-hour session.
Generate the transcript with speaker labels on. Expect a solid word layer and a speaker layer that needs work.
Calibrate labels against the roll call. Play the opening round, note which label each participant got, and rename the labels to participant codes (P01, P02…). Then skim for places where one label suddenly covers two opinions in a row, a sign that two voices were merged.
Verify where it matters analytically. Not every "mm-hm" needs an owner. Every passage you will code, quote or count does: click to the timestamp, listen, confirm the speaker and the words. Budget this time explicitly; for a lively group it is the largest part of the job.
Mark the unresolvable. Use consistent tags such as [overlapping], [inaudible 00:41:12] and [unidentified participant]. A marked unknown is honest data; a guessed name corrupts your analysis.
Add moderator and context notes. Insert notes from the observer, like [group laughs] or [P03 points to the prototype], where they change the meaning.
Anonymise, then export. Replace names and identifying details, keep a log of what you changed, and export to DOCX for reading or CSV/XLSX if your coding workflow uses spreadsheets.
For online groups run on Zoom, Google Meet or Teams, the same workflow applies to the platform recording; this guide to transcribing meeting recordings explains where each platform saves its files. Zoom's desktop client also has a local recording option to save a separate audio file per participant. When it works, each file contains one voice, which removes the attribution problem entirely, but test it on a short call first because users report it does not always produce the separate files.
Focus group transcript example
There is no single required format, but a usable focus group transcript has a header with session metadata, participant codes instead of names, timestamps at each speaker turn, and consistent tags for non-verbal events. Here is a short sample in that format:
Session: FG-03, online banking app, 8 recruited, 7 attended Date: 2026-10-14 | Moderator: MOD | Observer notes: AM Participants: P01–P07 (codes only; key held separately) [00:12:04] MOD: Let's talk about the transfer screen. P05, you mentioned it earlier? [00:12:09] P05: Yeah, I never know if the money's gone or not. There's no confirmation. [00:12:15] P02: Same. I screenshot it every time, just in case. [00:12:18] [overlapping, several participants agreeing] [00:12:21] P07: I actually like that it's fast, though. [laughs] [00:12:25] MOD: Fast but not reassuring. Does anyone feel differently? [00:12:29] [unidentified participant]: It depends on the amount. [00:12:31] P01: Right, for rent I check three times.
The conventions behind that sample, as a quick template:
| Element | Convention | Why it matters |
|---|---|---|
| Speaker IDs | MOD, AM, P01–P10; key stored separately | Keeps the transcript pseudonymised and codable |
| Timestamps | At each speaker turn, hh:mm:ss | Lets anyone jump back to the audio to check a quote |
| Overlap | [overlapping] with a short description | Records that agreement happened without inventing who said it |
| Unclear speech | [inaudible hh:mm:ss] | Flags the gap and where to listen again |
| Unknown speaker | [unidentified participant] | Prevents false attribution in coding |
| Non-verbal events | [laughs], [group laughs], [pause] | Preserves tone where it changes meaning |
| Redactions | [employer], [town], [section removed] | Shows readers that something was removed on purpose |
Verbatim or clean transcript for a focus group?
Decide this before transcribing, because it changes how much checking you need. A clean (intelligent verbatim) transcript drops fillers and false starts and is enough for most thematic analysis, where you code what people said. A strict verbatim transcript keeps every "um", restart and overlap; conversation analysis and discourse analysis usually need it. Automated tools produce something close to clean verbatim by default, and group crosstalk is exactly where strict verbatim is hardest to automate. If your method needs it, plan for substantial manual work on the overlapping passages.
Focus group transcription services vs software: cost and time
You have three realistic options, and they differ mostly in where the hours go.
| Option | Cost for a 90-minute group | Your time | Speaker attribution |
|---|---|---|---|
| Transcribe it yourself | Free | Many hours of typing (up to about 10 hours per recorded hour, per the Rutgers estimate) | Yours, so as good as your notes and memory |
| Human transcription service | About $179 at Rev's $1.99 per minute; other human services also charge per audio minute | Checking only | Done by a transcriber who was not in the room; still worth spot-checking |
| AI transcription plus your verification | Much less than a human service; about $8 from BlaBlaScribe's 200-minute pack, for example | Checking and fixing labels, usually a fraction of typing time | Draft labels from diarization; you verify the passages you use |
Rev's current price is listed on its pricing page ($1.99 per minute for human transcription as of November 2026). Human services make most sense when you need strict verbatim on difficult audio or have no time at all to check. For a fuller comparison, see when human transcription isn't worth it and our roundup of the best AI transcription software.
As one example of the software route, BlaBlaScribe transcribes uploaded recordings in 120+ languages with speaker labels and exports DOCX, CSV, XLSX and other formats. Accuracy on the words is up to 99.8% on clear audio; on a noisy group recording it will be lower, and like every diarization system it cannot reliably split two people talking at once. It is not a meeting bot and does not join live sessions; you upload the recording or paste a link. The free plan gives 30 minutes once, and a 200-minute pack costs $17.50 without a subscription. Files are processed on servers in Germany and are not used to train AI, which matters for ethics review.
How to anonymise a focus group transcript
Participants' voices, names and stories are personal data, so plan anonymisation as part of transcription, not as an afterthought. The UK Data Service guidance on anonymising text data recommends a few practices that fit focus groups well:
Use descriptive replacements, not blanks. "[colleague 1]" or "[home town]" keeps the discussion readable; a string of X's makes it useless for analysis.
Apply replacements consistently across every transcript in the project, and remember that people refer to each other by nicknames that a search will miss.
Keep an anonymisation log of every change, stored separately from the transcripts.
Group discussions add one risk interviews do not have: participants name each other and mention shared workplaces or neighbourhoods. Read for those cross-references, not just the speaker labels. Note too that under the EU GDPR (Recital 26), pseudonymised data that can still be linked back to a person through a key remains personal data, so the transcript with participant codes still needs secure storage while you hold the key.
Frequently asked questions
How long does it take to transcribe a focus group?
Typing it manually can take up to about 10 hours per hour of discussion. With an AI draft, the job becomes checking speaker labels and key quotes, which usually takes a fraction of that, depending on group size and audio quality.
How much does focus group transcription cost?
Human services charge per audio minute; Rev, for example, lists $1.99 per minute, about $179 for a 90-minute group. AI transcription costs far less (about $8 for 90 minutes from a $17.50, 200-minute pack, for example), but you spend your own time verifying who said what.
Can AI tell apart every speaker in a focus group?
Not reliably. Speaker labelling works best with a few distinct voices taking turns and degrades with more participants, similar voices, overlapping speech and a distant microphone, so plan to verify attribution in the passages you analyse.
What should a focus group transcript include?
A header with session details, participant codes instead of names, a timestamp at each speaker turn, and consistent tags for overlap, inaudible speech, unidentified speakers and redactions. That makes it codable and lets anyone check a quote against the audio.
Should focus group transcripts be verbatim?
Clean (intelligent) verbatim is enough for most thematic analysis. Strict verbatim with every filler and overlap is needed for conversation or discourse analysis and requires much more manual work on group crosstalk.
Is it better to record focus groups online or in person for transcription?
Online groups usually transcribe better because each participant is close to their own microphone, and some platforms can save separate audio per participant. In person, a central table mic, a roll call and a moderator who says names recover most of the difference.
Turn your recordings into text you can use
Upload audio or video, paste a YouTube link, or record in the browser — and get a speaker-labeled transcript in minutes, in 120+ languages.
Try BlaBlaScribe for freeNo credit card required · Stored in the EU · Never used for AI training





