Verbatim vs clean-read — you choose, in the editor
The model produces a near-verbatim transcript: it keeps the words as spoken but punctuates and cases them properly, and it drops most pure stutters. From there the editor lets you go either direction. For qualitative research and oral history, leave it verbatim — false starts and hedges are data. For a published Q&A, clean-read it: click any answer, tighten the phrasing, and the edit autosaves straight into every export format.
Every segment carries a timestamp, which is the quote-verification workflow: search the transcript for the phrase you want to quote, click its timestamp, and the built-in player jumps the audio to that exact moment so you can confirm the words before they go to print.
Recording quality is the whole game
An interview is usually two voices, which is the easy case — far easier than a six-person meeting. What actually decides accuracy is the recording setup. In person: put the phone between you, not in your pocket. Phone calls: use your phone's call-recording feature or record the call speaker-out with a second device, and expect the callee's side to be slightly harder for the model than yours. Video calls: record locally rather than relying on the platform's compressed cloud audio when you can.
One thing this page won't do is legal advice, but know the basics: consent requirements for recording conversations vary by jurisdiction (one-party vs all-party consent), and professional codes usually expect you to tell your subject they're being recorded. Get consent on the recording itself — it's then in the transcript too.
Frequently asked questions
Is the transcript verbatim or cleaned up?
Near-verbatim by default: the words as spoken, with punctuation and casing added, and most stutter repetitions dropped. Filler words like "um" are sometimes kept, sometimes dropped — if strict verbatim matters for your methodology, do a listen-through pass with the segment player and restore them where they carry meaning.
How do I verify a quote before publishing it?
Search the transcript for the phrase, click the segment's timestamp, and the audio player jumps to that moment. You hear the exact delivery — including the sarcasm or hesitation a transcript can't carry — and can fix any misheard word inline before exporting. Faster and safer than scrubbing through a recording by hand.
Can it separate my questions from their answers?
With speaker labels (Pro or a credit pack), yes — two-voice separation is diarization's best case, and you can rename the labels to "Q" and "A" in one click each. On the free tiers the transcript runs together, but question marks and your own memory of the conversation make manual splitting quick for a one-hour interview.
Do phone-call recordings work?
Yes. Call recordings are narrowband (the classic "phone sound") and the model handles them well, though expect a few more errors on the far end of the line than on your side. AMR files from Android call recorders upload directly — no conversion needed.
Do I need consent to record an interview?
Rules vary by place: some jurisdictions require only one party's consent (yours), others require everyone's. Journalism and research ethics codes generally expect disclosure regardless of the legal floor. This isn't legal advice — check your jurisdiction — but the safe universal practice is simple: ask on tape, and keep the ask in the recording.
How much does a 90-minute interview cost to transcribe?
With a $5 credit pack (600 minutes), a 90-minute interview costs 90 credits — 75 cents, no subscription, and the rest of the pack never expires. A human transcription service charges $90–135 for the same audio. Interviews up to 30 minutes are free with no account; up to an hour with a free account.