Key Takeaways
Transcribe the voice message and treat the text as the record. The problem with a spoken instruction isn’t the time it takes to listen, it’s that three weeks later you can’t search it, quote it or forward it to the person doing the work. Addresses, quantities, dates and name spellings are the details most often misheard and the most expensive to get wrong. Converting what a client sends is within your control. Changing how they send it usually isn’t, so plan around the habit rather than fighting it.
Some clients type. Some clients talk. If yours talk, you already know the shape of it: a forty-second note at seven in the morning with the job, the address and a change to last week’s order all in one breath, and no written trace of any of it anywhere.
You can listen to it. That part is fine. What’s not fine is what happens later, when someone asks which flat number you were given, or whether the client actually approved the second option, and the only answer you have is somewhere inside an audio file in a chat thread you’ve scrolled past two hundred times since.
The listening isn’t the expensive part
Forty seconds is forty seconds. Annoying when you’re driving, trivial over a week. The real cost shows up much later, and it shows up as absence: a voice message cannot be searched. Your phone will find the word “invoice” in a message someone typed eight months ago. It will not find the word “invoice” inside a recording, because as far as the search index is concerned there are no words in there at all, just a file with a duration.
That matters most at exactly the moment you least want extra work. Three weeks after a job, the question is never “roughly what did we discuss”. It’s specific. Which address. Which finish. Which date. Whether the extra hour was agreed or assumed. When there’s a disagreement about what was agreed, one line of text you can point at beats an audio file every time, and it beats it even when the audio file contains the identical information, because the audio file requires the other person to sit and listen and agree with your interpretation of what they heard.
There’s also the scrubbing. Finding a detail inside a voice note means playing it, guessing where the relevant bit is, dragging the slider back too far, playing it again. Do that across five notes from the same client and you’ve spent more time hunting than you would have spent doing the task. A written record turns that into a search box and a couple of seconds.
Voice is worst at the details that cost the most
Look at what actually arrives by voice. It’s rarely abstract. It’s a house number, a postcode, a quantity, a delivery window, someone’s surname, a model number read off a box. These are the details that get misheard, and they’re the ones where being wrong is expensive rather than embarrassing.
Spoken numbers are fragile. Digits blur together at speed, fifteen and fifty sound alike in a hurry, and a phone speaker in a noisy van doesn’t help. Street names get mangled by accent and by assumption, because your brain fills in the version it expects. Names are worse: you hear something plausible, write down the plausible thing, and the spelling on the invoice is wrong from that point forward. None of this is carelessness. It’s a property of listening once, at speed, while doing something else.
The rule worth adopting is simple: any detail that will end up written somewhere else, on a label, an invoice, a delivery note, a calendar entry, should exist as text before you copy it. Not because you’ll definitely get it wrong from audio, but because when you do get it wrong from audio there’s nothing to check against except the same recording that misled you the first time.
A message you can’t play is a job that doesn’t get started
Text is quiet. You can read it in a meeting, on a train, in a client’s kitchen while they’re standing next to you, in a workshop where the extractor is running. A voice note needs either silence or headphones, and often you have neither.
What happens then is predictable and it isn’t laziness. A spoken instruction that arrives somewhere you can’t listen gets postponed rather than acted on. You tell yourself you’ll play it in the car. The car happens, the phone is in your pocket, and by evening there are three of them stacked up and you’re not sure which one had the thing that was urgent. The instruction that would have taken twenty seconds to read has now taken a day to reach you, and the client’s mental model is that you were told at eight in the morning.
Text collapses that gap. You skim it wherever you are, decide whether it’s urgent, and either act or put it on the list with the actual content attached rather than a note to yourself that says “listen to Dave”.
Turning what they send into something you can keep
The practical move is to convert first and deal with the content second. A transcript makes the instruction searchable, quotable and forwardable, which covers most of what was missing.
- Searchable — the address from last month is findable by typing part of the street name, not by remembering roughly which week it arrived.
- Quotable — when there’s a dispute, you can paste the client’s own sentence back rather than paraphrasing it and having the paraphrase argued with.
- Forwardable — the person who actually has to do the job gets the words, not a file they have to find somewhere private to play.
If the messages arrive on WhatsApp and you work at a desk, the SozAI extension for WhatsApp Web puts a Transcribe button under each voice message and shows a one-line summary with the full text underneath, in the chat where the message already is. It runs in your own browser session and nothing is sent back into the conversation, so the sender isn’t notified and the chat isn’t marked in any way. That last part matters more than it sounds, because plenty of people would rather not signal to a client that they’re processing their messages.
For everything that isn’t a chat message, recordings you made yourself, files someone emailed you, longer audio you need with speaker labels or in another language, that’s the transcription app for iOS, Android and macOS, which handles audio and video files across a long list of languages. Same principle either way: get it into text, then work from the text.
Once you have the words, do the boring thing with them. Paste the address into the job sheet. Put the date in the calendar with the client’s phrasing next to it. If the instruction changes something previously agreed, reply in writing summarising what you now understand the job to be. A transcript plus a written confirmation is a solid record. A transcript sitting alone in a chat is only half of one.
Whether to ask clients to write it down
Ask. It costs nothing and it works with some people. “Can you text me the address rather than saying it” is a normal request and a fair number of clients will simply do it, especially if you frame it as protecting them from a delivery going to the wrong door.
Then accept that a good share of them won’t, and won’t for reasons that have nothing to do with you. Some find typing genuinely hard. Some are always driving. Some are older, or dyslexic, or just faster at talking, and for them voice is the reason they can send you anything at all at seven in the morning. Pushing harder tends to buy you a week of compliance and then a quiet return to voice notes, plus a slightly worse relationship.
The line worth drawing is by content, not by habit. Let the chat be the chat. Ask specifically for the things that hurt when they’re wrong: addresses, numbers, spellings, anything financial. Most people will type a postcode when they won’t type a paragraph. And treat the rest as your problem to convert, because converting what arrives is within your control in a way that changing how someone communicates is not.
What a transcript won’t do for you
It won’t make a vague instruction clear. If the client said “do it like the last one but a bit bigger”, you now have that sentence in writing, which is better than nothing and still not a specification. Text is a record of what was said, not an improvement on it, and reading it back is often the fastest way to notice that you need to ask a follow-up question.
Accuracy has limits too, and they cluster around the same details that made you want a transcript in the first place. Unusual surnames, street names, product codes and fast-spoken digits are hard for any system working from noisy phone audio, so check those against the original before you commit them to an invoice or a delivery label. A transcript that’s right about the sentence and wrong about the house number is worse than useless if you never look twice.
And a written record is not the same as agreement. Having the client’s words in text puts you in a much stronger position than having a file, but the thing that ends most disputes is a short reply from you confirming the job as you understand it, sent at the time, in the same thread. The transcript is what makes writing that reply take a minute instead of ten. If you want more of the mechanics of getting spoken material into usable text, the pages on converting voice memos to text and turning audio files into text cover the rest of the cases you’ll run into.
