A focus group asks the moderator to do two incompatible jobs at once: run a live, cross-talking discussion among eight people, and keep an accurate record of who said what. You can't do both — moderating well means being present in the conversation, not stenographing it. So the record gets deferred to after the session, when it's rebuilt from memory and a recording that, for a group of eight, is hours of untangling. And the stakes on that record are unusually sharp: a report quote attributed to the wrong participant — crediting the competitor's user with the loyalist's praise — isn't a typo, it's a finding that's wrong, and findings are what the client paid for.
Recording plus speaker separation does most of the untangling. But this is also the use case where the honest limits of that technology matter most, so this article states them up front.
Key takeaways
- The moderator can't moderate and attribute simultaneously; the record has to come from the recording, not live note-taking.
- Speaker-separated transcription splits the group by voice and timecodes every line, so quotes come with attribution and a place to verify them (product fact below).
- Renaming "Speaker 3" once to "mom, 34, competitor user" propagates the label through the whole transcript, turning voices into meaningful segments (product fact).
- Honest limit stated plainly: a fast, overlapping eight-person group is diarization's hardest case. Treat attributions as high-quality drafts to verify against the audio before they go in a report — not as automatically citable truth.
Why the record can't be kept live
Good moderation is a full-attention activity: reading the room, following a thread, drawing out the quiet participant, managing the dominant one. A moderator taking verbatim attributed notes is a moderator not doing that — which is why professional practice records the session and analyzes it afterward. The problem has always been the afterward: for eight speakers, manual transcription-with-attribution is hours per session, and it's the least enjoyable, most error-prone part of qualitative work, which is exactly why it gets rushed.
That rush is where attribution errors enter — and in a focus group, attribution is the data. "A young competitor-user praised the packaging" and "a loyal long-time customer praised the packaging" are different insights with different implications. Get the speaker wrong and the analysis is wrong in a way that looks perfectly confident.
What speaker separation gives you
Process the recording and it returns speaker-separated: the discussion split by voice, every line timecoded, participants distinguishable (product facts). Rename each once — "Speaker 3" becomes "mom, 34, competitor user" — and the label carries across the whole transcript, so the raw voices become the meaningful segments your analysis needs (product fact). Themes surface in the summary; quotes for the report come with attribution and a timecode to check them against.
That timecode is the workflow's backbone here. Because the quote isn't just "attributed by the system" — it's checkable: one click to the exact moment in the audio, so before a line goes in the deliverable, you confirm it's the right person saying the right thing. The mechanics of how diarization tells voices apart, and its failure modes, are covered in speaker diarization explained.
The honest limit — and why it's still worth it
State it directly, because a focus group is where diarization is hardest: eight people talking rapidly, interrupting, overlapping. Overlap muddies the audio; short interjections carry too little voice to fingerprint cleanly; similar voices on a shared room mic can blur. So attribution in a busy group is not guaranteed perfect, and the shortest lines are the most likely to be misassigned. Any tool claiming flawless eight-way attribution from a crosstalk recording is overselling.
Here's why it's still transformative despite that. The alternative isn't perfect manual attribution — it's hours of manual attribution that's also error-prone and rushed. The recording gives you a high-quality first pass in minutes, with every line timecoded for verification, so your effort moves from transcribing-from-scratch to checking-and-fixing. That's the same principle as journalistic quote work — the transcript nominates, the audio confirms — applied to research, and covered in how to transcribe interviews. You still verify report quotes against the audio. You just don't spend the weekend building the transcript first.
Across a series of groups
Most research runs several groups, and the archive turns them into a corpus: "which group praised the packaging?" and "where did price come up?" are questions to the archive, answered with quotes and dates across sessions (product facts) — segment comparison without hand-merging transcripts. The series-into-structure discipline from customer discovery interviews applies; a focus-group program is that method with group dynamics instead of one-on-ones.
Sources and method
Product facts (speaker-separated timecoded transcripts on the premium engine, speaker renaming that propagates, thematic summaries, full-text search and archive-wide questions with quoted answers) describe MeetResult as documented in the product catalog at the time of writing. Diarization accuracy degrades with overlapping multi-speaker audio; attribution in a large crosstalking group is a draft to verify against the recording, not guaranteed — no flawless multi-speaker attribution is claimed, and no accuracy figures are cited. Participant consent to recording and processing is the researcher's responsibility. No statistics are cited or invented.
Related: Speaker diarization explained · How to transcribe interviews · Customer discovery interviews: structure the series · Ninety minutes, eight voices, the transcript knows who said what (use case)
Process your last focus group in @meetresultbot, rename the speakers once, and see how much of the analysis is already done — then verify your quotes against the audio. New accounts include welcome turbo minutes; diarization runs on the premium engine.