Customers send WhatsApp voice notes: how to answer them without listening to every one
Voice notes are quicker for your customers and slower for you. How to read them instead of listening, what AI can answer on its own, where transcripts go wrong and the line your privacy notice needs.

Contents
Another voice note, and your hands are full. It waits with the others until evening, when someone plays them all, writes down addresses and calls back people who got in touch 8 hours earlier.
You can read them instead. Below are free fixes for your phone, a setup where every voice note arrives as text, the mistakes transcripts make, what it costs and the line your privacy notice needs.
Why voice notes slow a small team down
For the customer, talking was quicker than typing. For you, the message stays locked until you have a quiet minute and a free ear. That happens all day:
- At 10:12 a.m. a roofer up on a roof gets a voice note of 2 minutes 40 seconds about a leak.
- At 2:30 p.m. a hairdresser has color setting on a client's head and 3 voice notes from the same customer waiting in WhatsApp.
- At 7:45 p.m. a restaurant owner is at the pass when someone asks, by voice, whether a group of 9 with a dog fits on the terrace on Saturday.
Listening costs more than the minutes
Three written lines take seconds to read. A voice note of 3 minutes takes 3 minutes, or 90 seconds at double speed, and you need quiet and a free hand.
Then come the other costs:
- The useful part is buried. The address comes after the story about the neighbor, and the days that suit the customer come right at the end.
- Only the person who listened knows what was said. When a colleague takes over, they listen again.
- Audio can't be searched. Three weeks later, nobody finds the customer who mentioned the entrance round the back.
- Even the easy answers wait. "Yes, we're open on Saturday" goes out in the evening, because nobody knew that was the question.
You won't train customers out of it
WhatsApp said in 2022 that its users send 7 billion voice messages a day on average (WhatsApp blog).
Plenty of people find them too long. In a YouGov survey of British smartphone users from 2022, 65% of those with an opinion said a voice note of 1 minute is already too long.
They keep coming anyway, so the fix is on your side: read them.
What you can do today without new software
If you get a few voice notes a day and you are the only one answering, the tools on your phone may be enough.
Let WhatsApp transcribe on your phone
Since November 2024, WhatsApp can turn a voice message into text. It works one message at a time, on the phone in your hand:
- Switch it on under Settings > Chats > Voice message transcripts.
- Press and hold a voice message.
- Tap Transcribe.
WhatsApp says the transcript is created on your device, so not even WhatsApp can read it. The feature started with a few selected languages (WhatsApp blog).
Telegram has a similar button for Premium users. Without Premium you get a limited number of trial transcriptions a week (Telegram API docs).
Play them at 1.5x or 2x
WhatsApp plays voice messages at both speeds, forwarded ones included (WhatsApp blog).
Ask for the facts in writing, once
If your WhatsApp sends an automatic greeting or away message, add one line: "Voice messages are welcome. Please also type your address and the days that suit you."
Customers keep talking, and you get the part you need as text.
Write down what you heard
After listening, reply with the facts so the customer can confirm them: "So: 27 Linden Road, water under the kitchen sink, Tuesday or Thursday after 3?"
Mistakes get corrected before you drive out, and nobody has to listen twice.
Where the phone stops being enough
This stops working once several people share one number (see using WhatsApp Business with multiple employees), or once voice notes arrive faster than anyone can play them.
Skim the transcript instead of listening
Once a voice note is text, you jump straight to the parts that matter instead of playing the audio from the start. An example we made up, from a plumber's inbox:
"Hi, yes, hello, it's Maria Keller, we spoke last year about the bathroom, anyway, now it's the kitchen, under the sink, there's water in the cupboard every morning, not a lot, my husband put a bowl there, and I don't know if it's the pipe or the tap.
"We live at 27 Linden Road, the yellow house, the entrance is round the back. We're at work most days but Tuesday or Thursday after 3 would work, or maybe Saturday morning. Thanks, bye."
You read that standing at the van and take out what matters:
- a small leak under the kitchen sink
- 27 Linden Road, entrance at the back
- Tuesday or Thursday after 3, or Saturday morning
Where transcripts go wrong
Speech recognition struggles most with the details a small business needs:
- surnames and street names
- house numbers and phone numbers ("fifteen" and "fifty" sound alike)
- dialect
- the noise of a building site or a kitchen
Treat every name and number in a transcript as unconfirmed until the customer has typed it or confirmed it in writing.
Before you rely on it, send a voice note in your customers' language from your personal phone, with some background noise, and read what arrives.
The setup: every voice note arrives as text
In Deskwoot, our customer service software, voice notes and audio files on WhatsApp and Telegram turn into text as they arrive, for messages of up to 10 minutes. Nobody has to play the audio to find out what the customer wants.
Transcripts are on by default. You switch them off under Settings > Account > Transcribe voice messages.
That covers WhatsApp numbers connected through Meta's WhatsApp Business Platform or through Twilio, and your Telegram bot (setup guides for WhatsApp and Telegram). Audio on LINE and SMS stays audio.
What your team sees
- In the web dashboard, the transcript sits under the audio player, marked "Transcript (automatic)". Long ones are folded, and "Show full transcript" opens them.
- The conversation list shows the first words of the transcript instead of just "Voice message". So does the push on your phone, unless you set pushes to leave out message content. The full transcript shows in the web dashboard.
- When there is no text, a grey line says why: the message was too long, the audio format can't be read, the month's AI conversations are used up, or the transcription failed. The audio stays in the conversation.
Fynn answers the routine ones
On the paid plans, Deskwoot's AI agent Fynn reads the transcript like a typed message. It answers from your website, help center, documents and FAQ lists (how Fynn handles voice messages).
Say a salon customer asks by voice whether you're open on Saturday, what a cut and color roughly costs and where to park. Fynn writes back your hours, the price from your price list and the parking tip.
If your price list doesn't cover the question, Fynn says so instead of guessing and offers a person. Its instructions mark a transcript as speech that software turned into text, so it asks again instead of guessing a name or number that may have been misheard.
When a voice note has no transcript, Fynn asks the customer in one sentence to write it briefly. If it already asked once, it hands the conversation to your team.
Fynn is not part of the free Hacker plan (1 agent). There, the transcripts help you read, and you answer yourself.
When a person takes over
Fynn also hands over when the customer asks for a person, when your content runs out, or when only a teammate can help.
It leaves a private note for your team with the issue in one sentence, the customer's mood and a suggested next step. The Deskwoot app for iPhone and Android tells you that the conversation is now yours.
On a busy day, ask Copilot in the conversation for a summary of a long voice note. Copilot reads the transcript too, and it works on the web and in the app (Copilot in conversations).
What it costs
Transcripts work on every Deskwoot plan, the free Hacker plan included. They share the meter with Fynn and Copilot: a conversation with a transcribed voice note counts once per window as an AI conversation (a chat counts again after 12 hours without a message).
Five voice notes in one conversation count once. A conversation that Fynn or Copilot already worked on in the same window costs nothing more (how AI conversations are counted).
| Plan | Per agent a month (annual billing in brackets) | AI conversations included a month | Each further one |
|---|---|---|---|
| Hacker | $0 | 10 | Not available, upgrade needed |
| Startup | $4 ($3.20) | 50 | $0.07 |
| Business | $8 ($6.40) | 100 | $0.05 |
| Enterprise | $12 ($9.60) | 300 | $0.03 |
The included AI conversations count for the whole workspace. Further ones come out of a prepaid AI balance.
When the included ones are used up and the balance can't pay for another, voice notes in conversations that have not counted yet in their current window arrive as audio, with a line saying so.
An example
A salon with 2 people on Startup, billed annually, pays $6.40 a month.
If 80 conversations in a month get a transcript, a Fynn answer or Copilot help, 50 are included. The other 30 cost 30 × $0.07 = $2.10.
The line for your privacy notice
When a service provider turns customer audio into text, that provider receives personal data. In the EU and the EEA, the GDPR counts it as a recipient (Art. 4(9)).
Several rules follow from that:
- Your privacy notice has to name "the recipients or categories of recipients of the personal data" (Art. 13(1)(e)).
- Where data goes to a country outside the EU and the EEA, the notice says so and names the adequacy decision or other safeguards it relies on (Art. 13(1)(f)).
- A processor needs a contract with you (Art. 28(3)), and a subprocessor it brings in is bound by a contract with that processor (Art. 28(4)).
- The processing needs a legal basis (Art. 6), such as steps taken at the customer's request before a contract (Art. 6(1)(b)) or your legitimate interests (Art. 6(1)(f)). Which one fits is a question for your data protection advisor.
A sentence you can adapt
Voice messages: if you send us a voice message on WhatsApp or Telegram, it is turned into text automatically so we can read and answer it faster.
For this we use our customer service software [name] (a processor under Art. 28 GDPR) and its subprocessor [transcription provider and its country], which processes the audio on servers in [region]. The audio and the text are stored with your conversation.
[Add your legal basis, how long you keep conversations and any transfer safeguards your advisor asks for.]
If you use Deskwoot
Deskwoot is your processor, and AssemblyAI, Inc. is its subprocessor for the transcription. According to Deskwoot's DPA (Annex B), AssemblyAI is a US company that processes the audio on servers in the EU.
It does so under the EU-U.S. Data Privacy Framework or otherwise standard contractual clauses, and it does not use files processed on its EU servers to train its models. You sign the DPA yourself in the dashboard (how to sign it).
In the sentence above, you fill in Deskwoot, AssemblyAI, Inc. (USA) and the EU.
Outside the EU
In Brazil, the LGPD gives people the right to clear information about the shared use of their data by the controller and its purpose (Art. 9, V LGPD). So the same sentence belongs in your privacy notice there.
Other countries have their own rules.
This section is general information, not legal advice.
Start by counting last week's voice notes
Scroll back 7 days in your WhatsApp. Count the voice notes, and how long each one waited before someone listened.
- If it was 2, the transcript button on your phone and one line in your away message will do.
- If it was 20 and most waited until the evening, have them arrive as text.
Test it on your own number
Sign up, connect your WhatsApp number and send it a voice note from your personal phone. Then read what arrives.
The free Hacker plan (1 agent) transcribes voice notes in up to 10 conversations a month, enough for that test. Every new workspace also starts with 7 days of the Enterprise plan, Fynn included, without a card and for up to 30 seats including yours.
When the trial ends, you pick a paid plan or keep working on Hacker. Startup includes 50 AI conversations a month and Fynn (plans and prices).
Examples by line of work: trades, salons, restaurants and online stores.
Also available in:DeutschEspañolFrançaisPortuguês (Brasil)


