AI legal transcription: a practical first-draft workflow
A useful draft saves typing without hiding the work that still needs a human ear. Here is how to evaluate one on your own recordings.
Read guide →Upload the recording and the template you already use. Review a finished draft instead of typing one. You certify; we only remove the typing.
Get my first draft freeYou review, correct and certify every draft, exactly as today. You just start from a document instead of a blank page.
Any common audio or video format — including courtroom FTR files. Attach one finished document from the client, Word or PDF — VerbaDraft learns the layout, speaker labels, verbatim level, even the certificate page.
Speech-to-text with speaker identification (including unexpected voices), then your client's conventions applied: Q./A. layout, false starts as dashes, fillers kept or cleaned per house style.
Your .docx draft is usually back in about a quarter of the recording's length — fifteen to twenty minutes for a one-hour recording — with timestamped review notes flagging every low-confidence word, so the reviewer goes straight to the spots that need ears on them.
okay so today is march 3rd and im here with detective alvarez uh can you state your name for the record yes detective maria alvarez badge 4471 thank you now on the night in question you responded to the call at approximately 11 pm is that right yes we arrived at i want to say 11:05
Same words, in the same order. We add the punctuation, the speaker labels, the layout and the review notes.
Every client wants their own format. Upload one finished sample — Word or PDF — and VerbaDraft reproduces it: title pages, speaker label styles, indenting, numbered lines, certificate pages. It remembers the format for every future job.
A hard technical guard means the AI can only touch punctuation and capitalization. Every spoken word — every um, false start, and repetition — survives exactly as said, or is cleaned only if the house style asks for it.
Each draft ships with a checklist of low-confidence words and their exact timestamps. Your reviewer checks those spots first instead of re-listening to everything.
Interviewer, witness, counsel — identified automatically from context, labeled per your template's convention. Handles surprise extra voices instead of breaking on them.
MP3, WAV, WMA, MP4 and other video, multi-hour recordings, phone-line audio — and courtroom FTR (.trm) recordings uploaded directly as the whole set of files, no export step. If it plays, we can draft it.
Drafts arrive as .docx. Review in Word alongside ExpressScribe or any player you already use. No new software to learn, nothing to install.
Going over never stops a job — extra minutes go on your next invoice. How billing works · Need a custom plan?
Usually within minutes. Processing runs at roughly a quarter of the recording's length, so a one-hour interview is typically back in fifteen to twenty minutes. Within 24 hours is our standard commitment for the rare case — a very long recording, or a busy moment on our side — where it takes longer.
Phone lines, echoing rooms, cross-talk — we process it all. The review notes flag exactly which words the transcription was unsure about, with timestamps, so rough patches cost your reviewer minutes instead of hours.
Yes. Files are used only to produce your transcript, never for AI training, never sent to third-party AI services, and auto-deleted 14 days after delivery. Details on our Security page — and you can put a signed NDA in place online in about a minute, before you upload anything.
Nothing. No card, no auto-enrollment. You get one real job done free; if the draft doesn't save your team serious time, walk away — the draft is yours either way.
All questions, including billing and confidentiality →
Transcription runs on systems we operate. Your recordings are never submitted to OpenAI, Google, or any external AI service.
Your audio and documents produce your transcript — nothing else. No model training, no "service improvement" fine print.
Files are auto-deleted from our active systems 14 days after your draft is delivered. Enforced by software, not by promise.
TLS on every transfer. Files stored on access-controlled servers in the United States. Payments handled entirely by Stripe.
Handling law-enforcement or court material? Read the full Security & Confidentiality page — and a signed NDA takes about a minute to put in place online.
No card. Judge the draft against a transcript typed from scratch.
From the guidebook
Practical help with first drafts, templates, and who said what.
A useful draft saves typing without hiding the work that still needs a human ear. Here is how to evaluate one on your own recordings.
Read guide →Your template is a formatting example. A clear, consistent one gives a draft a much better starting point.
Read guide →Correct words are not enough if they belong to the wrong person. Here is a practical way to review speakers in an AI transcript.
Read guide →Beyond 150 audio-hours a month, or with special retention, template or turnaround needs — describe it briefly and we'll come back with a quote, usually the same business day.
Send one real job — the audio and the template you'd normally type it into — and get the finished draft back in your format, usually within minutes. No card, no auto-enrolment, no strings. If it doesn't save your team real hours, walk away and keep the draft anyway.
Try it on a real job — free