LinkedIn Voice Messages in B2B Outreach and When They Work

What is a LinkedIn voice message, and is it worth sending to prospects?

A LinkedIn voice message is a short audio note you record and send inside a LinkedIn conversation, in the same thread as your written messages. In B2B outreach it works as one warm touch to someone who already knows your name from earlier text. It never replaces the opening text. For that first message, see what to write in a LinkedIn cold message.

Open every conversation in text, short and human.
Send a voice note only inside a thread where you have already sent a few texts, as one touch in the middle of a sequence.
Record it yourself. A cloned voice brings an honesty problem that text never has.

I have no voice-note campaign numbers, so I will not invent any. Everything I say from my own campaigns below comes from text and email sequences.

A voice note is also not a phone call. A call rings and demands attention the moment it arrives. A voice note sits in a message thread, and the prospect decides when to play it, or whether to play it at all.

So it is a written-channel touch: easy to ignore, and it never replaces a conversation with someone who has not agreed to talk.

Do voice notes really get more replies than text, as vendors claim?

I have not seen it shown in a form you can check. If a vendor or a guru quotes a big reply-rate lift for voice notes, ask for a sample size, a time window, a definition of a reply and a text group to compare against.

Before you trust any voice-note statistic, put it through these questions:

Who sent the notes: one strong rep, or a whole team?
How many notes went out, to how many people?
Over which dates was it measured?
What counted as a reply? "Not now, thanks" and "send me some times" are very different answers.
What did a comparable text sequence get, sent to similar people in the same weeks?
Did the notes go to strangers or to people already talking to the sender?
Does the publisher sell the voice feature it is measuring?

A claim that cannot answer questions 1 to 5 is an anecdote. One that answers them but fails question 7 is still vendor data.

A reply rate alone says nothing about meetings, shows or signed deals. Judge any format on the whole chain: replies, interested, booked, showed, closed. If a voice note lifts replies but the extra replies are polite brush-offs, your pipeline has not moved an inch.

There is a second reason a borrowed reply rate tells you little. The rate follows the sender at least as much as the medium, because the prospect checks who you are before answering. In our campaigns, one message went out from two senders. One company got 2% replies. The other got 7.5%. Nothing else changed. Copy cannot fix a weak sender, and a voice note will not either.

What does independent research say about hearing a voice versus reading text?

The serious evidence comes from psychology labs, and it points one way: hearing someone makes them seem more like a thinking person than reading the same words. Three studies carry most of it.

Schroeder and Epley (2015) had evaluators judge job candidates' pitches by hearing them or by reading them. Listeners rated the candidates more competent, thoughtful and intelligent, and showed more interest in hiring them. In one experiment the evaluators were professional recruiters, and adding visual cues to the audio did not change the ratings (PubMed abstract).

Epley later described those recruiters as Fortune 500 recruiters who preferred the audio pitches to the written ones, with video adding little (Chicago Booth Review).

Schroeder, Kardas and Epley (2017) found that hearing a person explain their beliefs made them seem more mentally capable than reading the same content, and the authors expected the gap to matter most when the listener disagreed with the speaker. The study was preregistered, and a fourth experiment pointed to cues in the sound of speech itself, beyond the words (Psychological Science).

Kumar and Epley (2021) had people reconnect with an old friend by phone or email, and chat with strangers over video, voice or text. Voice built stronger bonds than text with no extra awkwardness, while people's wrong expectations about awkwardness pushed them toward text (Journal of Experimental Psychology: General).

People also underrate their own voice. In the Chicago Booth Review essay, about 1,000 survey respondents were asked how to come across as most intelligent, and 70% said they would choose writing.

Now the limits, study by study:

Study	What was tested	What it does not prove for LinkedIn outreach
Schroeder and Epley, 2015	Rehearsed job pitches, heard or read, judged by employers and recruiters	Replies or meetings. Evaluators were asked to judge; nobody had to decide to answer an unsolicited note
Schroeder, Kardas and Epley, 2017	People explaining views on divisive topics, heard or read	Anything about buying. It measured how thoughtful the speaker seemed, and nobody was asked to act
Kumar and Epley, 2021	Phone calls versus email between old friends, and live video, voice or text chats between strangers	Recorded notes. The friends were reconnecting by choice and the strangers chatted live

So the research supports one narrow claim: a voice note can make you sound like a real person with a working mind. None of these studies counted replies, meetings or deals, and none involved a busy buyer receiving a recording from someone they barely know. Any line drawn from them to LinkedIn is inference, mine included.

Text, voice note, video or call: a decision table by relationship stage

Choose the format by where the relationship stands. This is the table I would decide from:

Where the relationship stands	Text	Voice note	Video	Call
Stranger, no conversation yet	Yes: a short note with no pitch	No: nothing to refer back to	No	No: that would be a cold call
New connection, no reply yet	Yes: your first real message	Not yet	Not yet	No
Replied once, then went quiet	Yes: answer what they said	Yes, if tone helps more than words	Only to show something	No
Silent after several text touches	One more short text is fine	Yes: where I would place it	Rarely	No
Asked you a question	Yes, if the answer fits in two lines	Yes, if the answer needs nuance	Yes, if the answer is visual	Offer one
Replied and asked to talk	Confirm a time	Not needed	Not needed	Yes: book it

The decision rule underneath it:

The first message is always text. Short, human messages beat anything that sounds like a template. I never pitch in a first message, because a pitch is the quickest way to get ignored. What that first text should say is covered in what to send after a connection is accepted.
Voice is a later touch to someone already in the thread, when hearing you adds something the words alone cannot.
Video is for showing, such as a screen, a site or a drawing. If there is nothing to show, voice does the job with less effort on both sides.
A live call is earned. Asking for 15 minutes inside a written message is fine once you are past the connection request, but pick up the phone only for someone who replied, asked a question or asked to talk.

Where the voice note sits in a seven-touch sequence

One or two messages never start a real conversation. I plan for at least 5 to 6 touches and spread them across channels whenever the market allows. Here is a seven-touch version with one voice note in it:

Connection request. No pitch, no link.
First message, text. One specific reason you are writing, one question.
Follow-up one, text. A new angle or a useful observation about their work.
Follow-up two, text. One piece of proof, kept to a sentence or two.
Follow-up three, voice note. The 30-second script in the next section.
Warm email. It picks up the LinkedIn thread for anyone still silent.
Final short text. A polite close that leaves the door open.

Leave days between touches. In our text and email campaigns, the third follow-up, with real time before it, is usually what works best.

So the voice note sits in slot 5: the most human touch goes where replies already tend to arrive. That is a placement choice from text and email experience, not a voice-note result.

The warm email comes after the LinkedIn touches on purpose. The email never goes first, so the name is already familiar when it lands. The full logic of spacing and channels is in the follow-up sequence guide.

Before touch 1, check the profile the note will come from. Prospects look you up after they hear you, so build it as a page for clients, not a CV. If that profile looks empty, even a well-made voice note sounds like a stranger with a microphone.

What should you say in a LinkedIn voice message?

Say four things: their name, one specific reason tied to their work, one useful observation, and a small ask for a short call.

A made-up example. Tom runs a small firm that does site safety audits. Laura heads operations at a civil engineering contractor that just posted about winning a bridge contract:

Hi Laura, Tom again, a couple of my earlier messages were about site safety audits for builders. You posted last week that the bridge job starts in March, congratulations. When contractors reach mobilization, the safety plan is often the item still waiting for sign-off, and it slows the first weeks. I have a few notes on how teams get it approved earlier. Would 15 minutes next Tuesday or Wednesday be useful to go through them?

That is about 75 words, roughly 30 seconds at a calm pace. Line by line:

Name and sender in the first 5 seconds. Then move on. Nobody needs your history.
The reason is about her. Her post, her project, her month.
The observation is the value. Use something you actually see in her market. If you cannot say one true thing about her situation, send text instead.
The ask is small and concrete: 15 minutes, two days. In any single message, my team sells the next conversation and leaves the service for the call.

What goes wrong most often:

Reading a template aloud. It sounds read, and read sounds like a robot. Jot three points and talk.
Opening with 15 seconds about yourself. The prospect decides whether to keep listening before you reach the point.
Pitching in the first breath. Price, features and a list of case studies belong on the call.
Rambling past the ask. Once you have asked the question, stop recording.

Time every note before you send it. If it runs past 30 seconds, cut a sentence and record again.

Is it honest to send an AI cloned voice note?

Tools now clone your voice from a short sample and turn one script into hundreds of "personal" notes. How short? ElevenLabs' voice cloning docs ask for at least 10 seconds of audio in total, with each file at least 5 seconds long. Inworld's documentation says its clones work from as little as 3 seconds, and sound more like you with longer samples.

The honesty question is simple: does the prospect believe you recorded it?

	Real recorded voice	AI cloned voice	Text
Sounds like you	Yes, because it is you	Close enough to pass, by design	Your words, no sound
Prospect knows a machine made it	Nothing to know	Only if you tell them	Not an audio question
Disclosure duty under EU AI Act Article 50(4)	None	Possibly, if it counts as a deep fake; get legal advice	Outside the deep-fake audio rule
Risk if the prospect finds out	None	They start doubting everything else you sent	Low
Where it fits in a sequence	One later touch to a warm thread	Hard to justify without disclosure	Every touch, including the first

The legal side is not settled. Under the EU AI Act, a deep fake is picture, sound or video made or altered by AI that looks or sounds like real people, places or events and would mislead someone into thinking it is genuine.

Article 50(4) requires deployers, meaning businesses and professionals using an AI system that generates deep-fake audio, to disclose that it was artificially generated or manipulated, and Article 50(5) says that disclosure must be clear and given no later than the first exposure. The regulation applies from 2 August 2026 (EU AI Act on EUR-Lex).

Whether a cloned note in a sales sequence meets that definition is a question for a lawyer. This is not legal advice.

The trust side is clearer. The US Federal Trade Commission ran a public voice cloning challenge to find ways of protecting people from cloning harms such as fraud, and it warned that families and small businesses can be targeted with extortion scams, and that policymakers cannot count on industry self-regulation alone (FTC Voice Cloning Challenge).

To a buyer, a synthetic voice that turns up unannounced looks like a scam attempt before it looks like a personal gesture.

The lab work says voice lets a listener hear a real mind. A clone pretending to be you delivers the sound without the mind, which removes the one thing voice was supposed to add. Scale the text, where nothing pretends to be your voice, and save your real voice for threads where the prospect already knows your name.

Scale the text, record the voice yourself

Conversations come first. Calls come out of conversations. Clients come out of calls. A voice note is one way to restart a quiet conversation, later in the thread, in your own voice. Write the text sequence so it can run at volume. Record the voice note yourself, and send it only into a thread where the prospect has already seen your name a few times. Scale the text. Never scale the voice.

Wondering if a voice note belongs in your LinkedIn sequence? Book a call and tell me who you are messaging. I will say where in the thread I would put one.