Nine Minutes, Three Questions: Turning a Spoken Brief Into Something You Can Sign
Photo by Photo by Chase Chappell on Unsplash on Unsplash
It lands at 7:42pm on a Thursday. It begins with "right, so, quick one" and ends nine minutes later with a cough, a car door and one crucial sentence about the deadline that you will never find again. Voice notes have quietly become the default brief format for a lot of UK creative work, and honestly, you can see why — they're faster to send than a document, they carry tone, and nobody has to open a laptop.
The trouble is that a spoken brief is not a brief. It's a stream of consciousness with three decisions hidden inside it, and if you start working from your memory of it, you'll be rebuilding the same conversation in six weeks' time with a client who genuinely believes they told you.
Here's a method that takes about twenty minutes and has saved me from more than one awkward scope conversation.
Why the voice note keeps winning
A written brief asks the client to think structurally before they've had the thought. A voice note lets them think out loud, which is often where the real constraints surface — "we can't touch the header because the developer's away until the 3rd" is exactly the kind of thing that never makes it into a form but absolutely shapes the job.
So don't fight the format. Treat the audio as the raw material and the scope as the deliverable. Your job is translation, not transcription.
First: get it into text
You need words on a screen you can highlight. The phone's built-in transcription (both iOS and Android do this now) is fine for a first pass, and tools like Descript, Otter or a locally run Whisper model will give you a cleaner version. For a nine-minute note, most of these will take under a minute and mangle perhaps a dozen words.
Two practical notes. First, listen through once at 1.5x while reading the transcript — your ears catch what the machine drops, particularly names of files, brand terms and numbers. Second, timecode the awkward bits. If the client said something ambiguous about "the two versions", note the timestamp so you can quote it back verbatim rather than paraphrasing.
If the audio is a mess of overlapping speakers — a client and their business partner both chiming in — split it into speaker turns before you start extracting anything. You want to know who said the thing about the extra page.
Second: sort noise from decisions
Go through the transcript with three colours, digital or otherwise.
Decisions are the things that have been settled: the audience, the pages, the deadline, the platform, the fact that the old logo is staying. These go straight into your scope without modification.
Preferences are the things that have been floated but not agreed — "maybe a warmer palette", "I quite like that rounded typeface". These are not scope. They're options, and they belong in a short list the client picks from, with a price attached if it changes the shape of the work.
Noise is everything else: the apology for the ramble, the comment about the weather, the story about their cousin's website. Delete it. It cost them nothing to say and it'll cost you plenty to interpret.
Most nine-minute notes contain somewhere between three and six genuine decisions. If you're finding fifteen, you're treating preferences as decisions and you'll be building something nobody signed off on.
Third: the assumption list
This is the part that quietly does the heavy lifting. Alongside the scope, send a short list headed "Things I've assumed — tell me if I've got them wrong." Maybe four or five lines.
Assumed the newsletter sign-up is staying where it is. Assumed the copy is coming from you, not me. Assumed we're not doing the mobile nav this round.
Assumptions written down are cheap. Assumptions left in the air are where the unbilled extra week lives. The wording matters — you're not asking permission, you're giving them a chance to correct you, which is a much easier thing for a client to do in a reply than in a meeting.
Fourth: the one-pager and the single reply
Now write the scope on one page. Not two. If it doesn't fit, your scope is doing the job of a proposal and you've got two documents fighting each other.
Put the deliverables at the top as a plain list, the deadline underneath, then the assumptions, then one line on what happens if the shape changes. Finish with something like:
If that's right, just reply "approved" and I'll get started. If anything's off, tell me which line and I'll adjust.
That single reply is worth more than a signature for day-to-day work, because it's a dated, written record of what you both understood — and it references the voice note, so the original audio becomes evidence rather than folklore. It's not a substitute for your terms and conditions, which should still be attached or linked; it's the scope that hangs off them.
Consent and retention, briefly
Here's the bit most freelancers skip. A client's voice note is personal data. It's their voice, it often includes names of colleagues, and it may wander into things that aren't your business. Under UK GDPR you need a lawful basis to process it (usually legitimate interests or the contract itself), and you need to not keep it forever.
Practically, that means three small habits. Mention that you transcribe briefs so the process isn't a surprise — a one-line note in your onboarding email is plenty. Keep the audio and transcript in the client's project folder, not your downloads, and not synced to a personal account. And set a deletion point: once the project's signed off and the invoice's paid, bin the audio and keep only the extracted decisions and the approved scope.
WhatsApp's end-to-end encryption protects the message in transit, not the copy sitting on your desktop. And if you're using a cloud transcription tool, check where it processes and stores the audio — most have a data processing agreement, and for client work you want that in place.
The ten-minute version
If you want to try this tomorrow: transcribe the last voice note brief you got, colour the decisions, write five assumptions, and send a one-pager asking for a single word back. The whole thing is quicker than the note itself was to record — and it means the next time somebody says "quick one" at 7:42pm, you'll have something to work from that isn't a memory.