Retell AI Voicemail Detection: How to Handle Answering Machines on Outbound Voice AI Calls (2026)
TL;DR: On most outbound voice AI campaigns a large share of dials reach voicemail, not a person. If your agent cannot tell the difference, it pitches a beep, burns minutes, and leaves rambling messages that make your client look bad. The fix has four parts: turn on voicemail detection in Retell, decide per campaign whether the agent hangs up or leaves a short scripted message, tag every call outcome so voicemails do not pollute your reporting, and cap retries so nobody gets called five times in a day. I build production voice agents for US clients on Retell, n8n, GoHighLevel and Twilio, and we run VoiceDash, the white-label client portal agencies hand their clients. This is the setup I use.
Inbound agents rarely think about machines. The caller dialed you, so a human is on the line. Outbound is the opposite. Reminder calls, lead reactivation, speed-to-lead callbacks and follow-ups all hit answering machines, carrier voicemail greetings, and the occasional phone tree. The broader rules for running those campaigns are in AI outbound calling campaigns. This post goes deep on the one piece most builds get wrong.
Why voicemail handling decides whether outbound works
Three things go wrong when an agent treats a voicemail greeting like a person.
It talks to the greeting. The agent hears "Hi, you've reached Mike, leave a message" and answers "Hi Mike, this is Sarah from the dental office, is now a good time?" Then it pitches an empty recording, and Mike gets a confusing half conversation in his inbox.
It burns minutes. Every second the agent spends talking to a machine is billed. On a list of a few thousand numbers, a sloppy voicemail flow can quietly eat a large part of the campaign budget. The per-minute math is in Retell AI pricing explained.
It wrecks the numbers. If voicemails are logged as answered calls, the client sees a terrible conversion rate and blames the agent, which barely talked to anyone.
All three are fixable, and the fix is mostly configuration and discipline, not clever prompting.
How voicemail detection works on Retell
Retell has a voicemail detection setting on the agent for outbound calls. When it is on, the platform listens to the start of the call and decides whether it reached a person or a machine. Once it decides it is a machine, you choose what happens: end the call, or have the agent leave a message. Option names and defaults change as Retell ships updates, so check the current agent settings in your dashboard rather than trusting a screenshot from last year.
A few things to understand before you rely on it.
Detection takes a few seconds. A machine is identified by how the audio behaves at the start of the call: a long uninterrupted greeting, then a beep. A person usually says a short "hello" and stops. The system needs a moment of audio to tell them apart, which is why your agent's opening line matters (more on that below).
It is not perfect. Some people answer with a long greeting of their own. Some voicemail greetings are a single word. Carrier greetings, call screening features and business phone trees all behave differently. Expect a small share of misclassifications in both directions and design for them instead of pretending they will not happen.
Phone trees are a separate problem. An IVR that says "press 1 for sales" is not a voicemail, and a voicemail setting will not navigate it. If your list is full of business main lines, decide whether the agent should try to reach a person or simply end the call and mark it. For most small business lists I end the call and flag the number.
Hang up or leave a message: decide per campaign
This is a client decision, not a technical one, and it should be in writing before launch. My defaults:
| Campaign type | Default on voicemail | Why |
|---|---|---|
| Appointment reminders | Leave a short message | The customer expects the business to call, and the reminder still does its job |
| Speed-to-lead callbacks | Leave a short message, then text | The lead just asked for contact, so a message plus SMS keeps the business first in line |
| Lead reactivation | Hang up on first attempt, message on the last | Cold-ish contacts tune out repeated messages |
| No-show and rebooking | Leave a short message | One clear callback number helps them rebook |
| Anything sensitive (medical, legal, financial) | Minimal message or none | Voicemail reaches whoever holds the phone |
For reminders specifically, the full flow is in AI appointment reminder calls. For the sensitive cases, the rules in HIPAA compliant AI receptionist apply to voicemail too: no clinical detail, no appointment type, just the business name and a callback number.
How to write the voicemail message
When the agent leaves a message, it should be a fixed script, not something the model improvises. Improvised voicemails ramble, and a rambling voicemail is the fastest way for a client to decide the agent sounds like a robot.
My rules for the message:
- Under fifteen seconds when spoken
- Business name in the first sentence
- One reason for the call, stated plainly
- One callback number, said slowly, ideally twice
- No pricing, no offers that expire, no personal details beyond a first name
- Approved by the client, word for word
A good example: "Hi, this is Ava calling from Brightside Dental with a reminder about your appointment tomorrow. If you need to reschedule, call us at 555 0142. Again, that's 555 0142. Thanks."
Pair the voicemail with a text
A voicemail on its own has a low response rate because many people never check voicemail at all. When the call ends on a machine, I trigger an SMS through n8n or GoHighLevel with the same information and a reply option, as long as the contact has given consent to texts. In practice the text often does more work than the voicemail. The wiring for that is in how to connect Retell AI to n8n and how to add voice AI to GoHighLevel.
Design the opening line for detection
The first thing your agent says has two jobs: earn a few seconds of attention from a person, and not trample a voicemail greeting while detection is still deciding.
Let the other side speak first. On outbound, configure the agent to wait for the callee to say something before it starts. A person says "hello." A machine starts its greeting. Either way, detection gets the audio it needs, and your agent does not talk over anyone.
Keep the first line short. "Hi, is this Mike?" works on both a person and a misclassified machine. A long opening pitch does not.
Tag every outcome
The most underrated part of voicemail handling is what happens after the call. Every outbound call should end with one clear outcome, so the client's reporting reflects reality.
The outcome labels I use:
- Connected, goal met (confirmed, booked, qualified)
- Connected, goal not met (declined, rescheduled later, not interested)
- Voicemail, message left
- Voicemail, no message
- No answer or busy
- Phone tree, ended
- Wrong number or disconnected
Retell gives you call data and disconnection details after every call, and its post-call analysis can extract a structured outcome field from the transcript. I define the outcome list in post-call analysis so every call lands in exactly one bucket. How I set that up is in Retell AI post-call analysis.
Then report the funnel honestly: dials, connects, voicemails, and conversions among the people actually reached. A campaign with a modest connect rate and a strong conversion among connects is a good campaign. Showing only "calls made" and "appointments booked" hides that. What to put in front of clients each month is in voice AI client reporting.
Retries, spacing and consent
Voicemail handling and retry logic are the same problem. If a number hits voicemail, the system has to decide whether and when to try again.
My defaults for most campaigns:
- Cap attempts. Two or three attempts per contact per campaign, not unlimited.
- Space them out. Different days and different times of day, because someone who never picks up at 10am may pick up at 5pm.
- One message per contact, not one per attempt. Leave a voicemail on the first or last attempt, not every time. Three identical voicemails from the same number is how a business gets reported as spam. The flip side of that problem is in voice AI spam calls.
- Stop on any opt-out. If the contact replies STOP to the text or tells the agent not to call, they leave the list immediately, across every campaign for that client.
- Respect calling hours. Call in the contact's local time zone, during normal daytime and early evening hours.
Outbound calling rules vary by country and US state, and AI-generated voices are a specific focus for regulators. Make sure your client has proper consent for the contacts on the list, and have them confirm it with their own counsel. I am an engineer, not a lawyer, and you should not take compliance advice from a blog post. The recording side of this is in voice AI call recording compliance.
Test it before the campaign, not during
Voicemail behavior is easy to test and almost nobody does it. Before any campaign goes live, I call a short list of test numbers:
- A phone with a normal personal voicemail greeting
- A phone with the default carrier greeting
- A phone with call screening turned on
- A business line with a phone tree
- A real person who answers with a long "Hello, this is Mike speaking, how can I help you?"
Listen to every recording. Did the agent wait? Did detection classify correctly? Did the voicemail message start after the beep, not during the greeting? Was the outcome tagged correctly? Fix what breaks, then run the list again. The rest of my pre-launch testing is in how to test a voice AI agent.
Let the client hear what actually happened
Outbound clients worry about one thing more than anything else: what is this robot saying to my customers? Voicemails make that worry worse, because the client never hears them unless someone shows them.
That is a big part of why we run VoiceDash. It connects to your Retell account and pulls calls, recordings, transcripts and usage into a portal with your logo, on your domain, scoped so each client only sees their own data and never touches Retell. A clinic owner can open a voicemail recording, hear the exact fifteen seconds the agent left, and stop worrying. Live in under 10 minutes, no code, on Starter, Growth or Ultimate plans from $19/mo with a 7-day free trial. VAPI and Bland support are coming soon.
The voicemail checklist
- Voicemail detection turned on for every outbound agent
- Hang up or leave a message decided per campaign, in writing
- Fixed voicemail script under fifteen seconds, client approved
- No sensitive details in any message
- SMS follow-up on voicemail, only with consent
- Agent waits for the callee to speak first
- Short identity check as the opening line
- Every call tagged with exactly one outcome
- Attempts capped and spaced, one voicemail per contact
- Opt-outs removed across all campaigns immediately
- Test calls to five greeting types before launch
- Recordings visible to the client in a branded portal
The bottom line
Voicemail is not an edge case on outbound. It is often the most common outcome. Turn on detection, script the message, tag the outcome, cap the retries, and test against real greetings before the list goes live. Do that and your outbound numbers mean something, your clients' customers get a clean fifteen-second message instead of a confused monologue, and your clients stop asking what the robot is saying.
Want your outbound clients to hear every call and voicemail in a portal on your own domain? Start free on VoiceDash or book a demo and I will walk you through the setup I give every client.