An interview produces two assets of wildly different sizes: an hour of audio, and the three quotes that will actually appear in the piece. Everything between those two points is overhead — traditionally, hours of manual transcription per hour of tape, or a budget line for someone else to do it. The overhead is why interviews pile up untranscribed, why quotes get reconstructed "close enough" from notes, and why the tenth interview of a project is harder to use than the first: nobody remembers which conversation contained what.
Machine transcription collapses the overhead. But for interview work specifically — where a misattributed or misheard sentence costs reputation — the workflow around it matters more than the transcription itself.
Key takeaways
- The transcript's job in interview work is navigation, not publication: it tells you where the quote lives; the audio remains the authority.
- Speaker separation is non-negotiable — an interview transcript that doesn't distinguish your questions from their answers is half-unusable (product fact: diarization runs on the premium engine).
- The iron rule: no quote goes to print unverified. Timecodes make verification a one-click habit instead of a scrubbing session.
- The compounding asset is the archive: a project's interviews, searchable as one corpus, answer questions that no single transcript can.
Record like the transcript matters
Transcription quality is decided mostly before processing. For interviews the essentials are old-fashioned: get the recorder close to the subject (your questions matter less than their answers), prefer a quiet room to a picturesque café, and for remote interviews record locally rather than through the call's compressed audio when you can.
Consent, in this genre, is craft rather than legal box-ticking: "I'm recording this for accuracy — I'd rather quote you right than quote you from memory" is a sentence that builds trust with a source. Respect off-record moments by actually pausing; the recorder's credibility is yours.
Process: separation first, words second
Drop the file into MeetResult — an hour-long interview is fine; processing is billed per minute of audio (product facts; the economics of irregular, project-based workloads are covered in per-minute vs. per-seat pricing). What comes back is a transcript split by speaker with timecodes on every line. Rename the speakers once — "Q" and the subject's initials — and the whole document inherits the labels (product fact).
That separation is what makes an interview transcript readable as an interview: your questions become scannable structure, and the answers — the actual material — stand out from them. How the separation works, and where it can stumble, is unpacked in speaker diarization explained; for interviews the practical takeaway is to glance over short interjections, the segments most often misattributed.
The verification habit
Here's the rule that keeps machine transcription compatible with journalism: the transcript nominates, the audio confirms. Names, numbers, technical terms, and anything quotable get checked against the recording before publication — and timecodes turn that from a chore into a click: the disputed sentence is at [23:41], not "somewhere in the second half."
This isn't distrust of the tool; it's the same standard you'd apply to your own typed notes. A transcript is a draft of a quote. The difference machine processing makes is that the draft exists at all, for every interview, the same day.
The archive: where the tenth interview pays off
A project of twenty interviews traditionally degrades into a folder of files nobody can query. Processed into one account, it becomes a corpus: full-text search finds which conversations mentioned the topic; questions to the archive — "who talked about pricing pressure?" — come back with quotes and timecodes across interviews (product facts: full-text search and archive-wide Q&A are built in).
For researchers this changes the analysis stage: themes get traced across subjects in minutes. For journalists it changes the follow-up: the second piece on a beat starts from a searchable base of everything sources already said.
Sources and method
Product facts (per-minute billing, speaker-separated transcripts with timecodes, speaker renaming, full-text search, archive-wide questions with quoted timecodes) describe MeetResult as documented in the product catalog at the time of writing. Recording and consent practices are editorial craft guidance, not legal advice — recording laws vary by jurisdiction. The estimate of manual transcription overhead reflects common practice of the genre; no precise benchmark is claimed. Machine transcripts may mishear names and terms — the verification rule above is part of the method, not a disclaimer.
Related: Speaker diarization explained · Per-minute vs. per-seat pricing · How to transcribe a Zoom recording · One hour of interview, three quotes (use case)
Take the interview that's been waiting longest: drop the file into @meetresultbot, rename the speakers, and find your three quotes tonight. New accounts include free processing minutes.