How does it know it reached a machine?

It listens for what answering machines actually say — the standard "leave a message after the tone" phrasings, and the tone itself when the greeting was silent or too short to classify. The decision is made from the transcript by fixed rules rather than by asking a model to guess, so the same greeting always gets the same answer.

Is the message a synthetic voice reading my script?

No. It is the audio file you recorded, played back as recorded. That is the point of the feature: the person hears you, not an approximation of you.

Can a family member or colleague record it instead?

Yes. It is an audio recording, so whoever should be the voice on the message records it. For a wellness check that is often a relative rather than the business.

Do I have to rerecord for each campaign?

No. Recordings are kept in a reusable library and attached to whichever campaign needs them.

Can I send voicemails without an AI conversation at all?

Yes, that is a separate campaign type: a voicemail drop, where landing the recording is the whole job rather than a fallback when a conversation does not happen.

Will people know it is automated?

The AI disclosure is on by default on the conversational side, and the wording is yours. Where you have recorded a real voice, that recording plays verbatim — what you recorded is exactly what is heard.

It reaches the machine. Your voice plays.

Voicemail handling means an AI call recognises when it has reached an answering machine rather than a person, and plays a recording you made yourself — in your own voice, exactly as you recorded it. The call is marked as voicemail so you can tell those apart from the ones a human answered.

Record a message What it costs

Voicemail calls draw on the same per-minute voice rate, and they are short. Recordings are reusable, so the effort is once rather than per campaign.

+1 425 ··· ···· 0m 24s

Octo

Hello, this is Harbour Dental calling about your appointment…

Caller

"Hi, you’ve reached Tom. Leave a message after the tone." [beep]

Octo

▶ Playing your recording — 11 seconds, in the voice you recorded it in.

After the call

Detected: Answering machine
Action: Recording played
Message: spring-checkup-v2.wav
Outcome: Voicemail

The greeting is matched by fixed rules rather than a model judgement, so the same greeting always produces the same decision. What plays is your file, not a synthetic reading of a script.

Or just ask for it

Your voice on the machine, arranged in a message.

Record it, point at it, and check where it landed.

Gmailfrom:octo voicemail

Octo96 voicemails left — your recording — Played verbatim, no text-to-speech4:20pm

OctoNew recording saved: “autumn-offer” — 18 secondsTue

OctoSplit: 318 answered / 96 machine — Classified by fixed rules, not a guessTue

Inbox96 voicemails left — your recording

Dme9:40 AM

Use the autumn-offer recording when it hits an answerphone.

OOcto4:20 PM

Done — 96 machines, your recording played back exactly as recorded, no speech synthesis in between. Those 96 are marked voicemail rather than answered, so they sort apart from the 318 real conversations.

Octoonline

how does it know it’s a machine?4:31 PM ✓✓

Fixed rules over the transcript, not a model judgement — so the same greeting always gets the same decision.4:31 PM

can I hear the one it left for tom whitfield?4:33 PM ✓✓

It’s the same 18-second file every time. Here it is, plus the call it was left on.4:33 PM

record a new one for december4:35 PM ✓✓

Ready when you are — call the number and I’ll save what you record as “december”.4:35 PM

Message➤

#omniocto4 members

MRMarcus1:12 PM

@Octo which recordings do we have?

OOctoAPP1:12 PM

On it 🔧(runninglist_voicemail_recordings…)

Three: autumn-offer (18s), missed-you (12s), and one from March nothing is using. (edited)

✅ 1

1 reply · last reply 33m ago

DWDana1:40 PM

bin the March one

Message #omniocto

Email octo@agent.omniocto.com · WhatsApp 1-MAN-ASK-OCTO · Slack #omniocto

Three parts

Recognise the machine, play the human.

Most systems either hang up on voicemail or read a script in a synthetic voice. Neither is what you would have done.

It recognises the greeting

Answering machines announce themselves in a small number of predictable ways, so the detection is a fixed set of rules over the transcript rather than a judgement call — and the tone itself counts when the greeting was silent or too short to read.

Your recording plays

The audio you recorded is played back as recorded. Not paraphrased, not re-synthesised, not read aloud from a script — the person hears the voice you chose.

Recordings are reusable

A recording is kept and attached to whichever campaign needs it, so a message you make once serves every list that follows.

A voicemail is a real outcome, not a failed call. It is classified as one so your results tell you which is which.

How you'd hand it over

Record, attach, read the results.

Record the message

In the voice it should be in.

You, a colleague, or a family member — whoever the person on the other end should hear. It is an audio recording, so it sounds like whoever made it.

"Use the message I recorded if it goes to voicemail — do not try to have a conversation with the machine."

Attach it to the campaign

Then run the list as normal.

Calls that reach a person have the conversation. Calls that reach a machine get the recording. You do not build two campaigns.

See which was which

Answered and voicemail are different outcomes.

The result for each contact says whether a human picked up or a machine did, so the follow-up list is the people worth ringing again.

✓ 318 answered · 96 voicemail left · 61 no answer

What is actually happening

Why it sounds like you and not like software.

Detection is rules, not a guess

The classifier matches the standard machine greetings and the standalone tone, over the transcript, with fixed rules. The same greeting always produces the same decision — which matters when the alternative is talking to an answering machine for a minute and billing you for it.

fixed rules · repeatable · tone-only fallback

Verbatim playback of a real recording

What plays is the file you recorded. There is no text-to-speech step between your voice and the person hearing it, which is the whole reason to use a recording rather than a script.

your audio · played as recorded

Or make the voicemail the entire job

A voicemail drop is its own campaign type, for when landing the recording is the objective rather than the fallback — the same list building, scheduling and results as any other campaign.

voicemail-drop campaign type

Voicemail is a first-class outcome

It is one of the thirteen call outcomes, distinct from no answer and from busy. Your retry policy can treat it differently — most people do not ring back someone who already has the message.

distinct outcome · retry policy aware

Disclosure still applies to the conversation

On the calls a person answers, the AI disclosure is on by default with wording you control, and every call records what the person was told.

default-ON · editable · per-call audit

Questions

The things people actually ask.

How does it know it reached a machine?: It listens for what answering machines actually say — the standard "leave a message after the tone" phrasings, and the tone itself when the greeting was silent or too short to classify. The decision is made from the transcript by fixed rules rather than by asking a model to guess, so the same greeting always gets the same answer.
Is the message a synthetic voice reading my script?: No. It is the audio file you recorded, played back as recorded. That is the point of the feature: the person hears you, not an approximation of you.
Can a family member or colleague record it instead?: Yes. It is an audio recording, so whoever should be the voice on the message records it. For a wellness check that is often a relative rather than the business.
Do I have to rerecord for each campaign?: No. Recordings are kept in a reusable library and attached to whichever campaign needs them.
Can I send voicemails without an AI conversation at all?: Yes, that is a separate campaign type: a voicemail drop, where landing the recording is the whole job rather than a fallback when a conversation does not happen.
Will people know it is automated?: The AI disclosure is on by default on the conversational side, and the wording is yours. Where you have recorded a real voice, that recording plays verbatim — what you recorded is exactly what is heard.

Keep reading

Related jobs you can hand over.

Start

Record it once. Use it for every list.

Record a message See pricing

Voicemail calls draw on the same per-minute voice rate, and they are short. Recordings are reusable, so the effort is once rather than per campaign.