Answering machine detection (AMD) is how an outbound call center tells a live person from a voicemail greeting. In VoiceOI, each bot checks what the called person says against your machine phrase list, turn by turn. When a machine answers, the bot ends the call and logs it as AM (answering machine), so bots and closers spend their time on people.
On an outbound floor, plenty of answered calls never reach a person. They reach a voicemail greeting, a carrier's mailbox prompt or a phone's built-in voicemail. Answering machine detection is the part of a dialer or voice bot that decides, call by call, whether a person or a machine picked up.
What happens next depends on the system: some hang up, some call the lead back later, and some leave a recorded message. VoiceOI hangs up and doesn't leave a message; see voicemail drop and what VoiceOI does instead.
AMD in a call center matters because every machine costs something:
Detectors work from two kinds of evidence.
Timing and sound.Classic AMD measures the audio at the start of the call. Asterisk's AMD application, for example, returns MACHINE when the silence before the greeting, the greeting itself or its word count runs past set limits, and HUMAN when the greeting is followed by enough silence. MightyCall's AMD page lists voice tone and pitch, the duration of pauses and background noise as its signals, and Call Center Studio's says it analyzes voice and silence durations.
Words.Speech-based detection transcribes the greeting and looks for voicemail wording, such as “leave a message” or “is not available”.
VoiceOI works from the words, and recorded-voice scripts can add a greeting-length check.
Sources: Asterisk documentation: the AMD application, MightyCall: AI answering machine detection and Call Center Studio: answering machine detection, all reviewed October 2026. Vendors change their products, so check their pages.
Where the machine and do-not-call checks sit on a VoiceOI call, from dial to disposition.
VICIdial passes an answered call to a bot working as a remote agent, or the built-in dialer places the call through your SIP carrier.
Speech recognition turns what the called person says into text. Before the bot replies, the text is checked against your machine and do-not-call lists, on the greeting and every turn after it.
Machine phrase: the bot hangs up, logged as AM.
An AM Detect step also treats a long greeting as a machine: people tend to answer with a word or two, while recordings keep going.
Greeting too long: the bot treats it as a machine and hangs up.
After your optional AI-voice disclosure, the script or AI agent asks its questions. A greeting that slipped past at the start can still be caught by a later phrase.
Do-not-call phrase: the call ends, logged as DNC.
Fronter and closer bots hand qualified live calls to a human closer over SIP, with AI-written notes on the Good Fronter tiers.
Each call keeps its disposition, recording and transcript; on VICIdial, the outcome is written back too.
More on each path: the VICIdial AI voice bot, the SIP trunk for AI voice agents and AI live transfer.
From an empty list to a tuned one. Machine and do-not-call lists are kept per organization.
Add phrases from the greetings your lead lists actually reach, such as "leave a message" or "mailbox is full".
List the ways people ask not to be called, such as "take me off your list".
In recorded-voice scripts, add the step where the bot listens to the first answer, so long greetings count as machines too.
Before any lead is dialed, try a short "Hello?" and a typical voicemail greeting, and check what the script does with each.
Assign bots to a VICIdial campaign as remote agents, or run the built-in dialer through your SIP carrier.
Check recordings and transcripts of machine-answered calls, fix phrases that caught people or missed machines, and export the tuned list.
Starting points, not a finished list. The risk column is a rule of thumb, not a measurement.
| Phrase | List | Where you hear it | Risk of ending a call with a person |
|---|---|---|---|
| "leave a message" | Machine | Personal voicemail greetings | Low |
| "after the tone" | Machine | Mailbox prompts from phones and carriers | Low |
| "mailbox is full" | Machine | Carrier prompts when a voicemail box is full | Low |
| "not available" | Machine | Carrier prompts and personal greetings | Medium: a relative may say "he's not available right now" |
| "can't come to the phone" | Machine | Personal voicemail greetings | Medium: people answering for someone else say it too |
| "take me off your list" | Do-not-call | A spoken opt-out | Low: when a person says it, ending the call is the point |
| "stop calling" | Do-not-call | A spoken opt-out | Low: add the other wordings you find in transcripts |
Illustrative examples. Calls are English-first, so phrase lists work best on English greetings.
The same checks run however the bot connects, and a tuned list can move between client organizations.
Bots join VICIdial campaigns as remote agents, provisioned in bulk. VICIdial keeps your lists and reports, and outcomes, AM included, are written back.
The built-in dialer calls through Twilio, Telnyx or your own SIP trunk, added by pasting its sip.conf or pjsip.conf block into the Carrier Manager.
A BPO running each client as its own organization can tune a list in one, export it and import it into another.
Every detector makes two kinds of mistake, and tightening one usually loosens the other.
False positive: a person treated as a machine.The call ends and the conversation is lost. A broad phrase such as “not available” can cause it, and so can a length check when someone answers with a full sentence.
Missed machine: a greeting that gets through. The bot talks to the recording until a later phrase matches or the call ends, costing bot time and carrier minutes.
VoiceOI doesn't publish an accuracy figure, because results depend on your lists and tuning. Measure it on your own calls: how many AM calls were really people, and how many short calls without an AM outcome were really machines. Test each change in the simulator; scripts, including the AM Detect step, are built in the no-code AI voice agent builder.
VoiceOI bills prepaid per bot with no per-minute AI fee, so calls that end on a machine add no AI charges. Fronter bots start at $25 per bot per month at list price (list prices run $25 to $155), with volume discounts up to 60%.
Two costs remain: your carrier bills the minutes these calls use, and a bot busy with a greeting that slipped past detection isn't talking to a lead. If your lists produce many short calls, ask your carrier how it treats them: the SIP trunk selection guide quotes carrier thresholds and billing increments for short calls.
Every tier is on the VoiceOI pricing page, and AI voice agent pricing compared sets per-bot pricing against published per-minute rates.
Detecting machines doesn't change the consent rules for the call itself. Since the FCC's February 2024 ruling, AI-generated voices count as "artificial" voices under the TCPA, so telemarketing calls that use them need prior express written consent. Load only leads who gave it, and keep the records.
VoiceOI's part is the controls: do-not-call phrases that end the call, an optional TCPA AI-voice disclosure at the start of the call, timezone calling windows, and a recording and transcript of every call. Compliance stays your responsibility, and none of this is legal advice. See TCPA controls for outbound calling and do-not-call (DNC) controls.
Source: FCC news release (February 8, 2024): AI-generated voices in robocalls are "artificial" under the TCPA, reviewed October 2026.
Common questions about AMD on outbound call center campaigns.
Answering machine detection (AMD) is how an outbound dialer or voice bot tells whether a call reached a person or a voicemail greeting. Call centers use it so fronters, bots and closers spend their time on people.
What the called person says is turned into text and checked against your machine phrase list on every turn, before the bot replies. Recorded-voice scripts can also add an AM Detect step that treats a long greeting as a machine. Either way, the bot hangs up.
Not today. When a bot detects a machine, it hangs up and logs the call as AM, without leaving a message. VoiceOI does not leave voicemails today.
VoiceOI doesn't publish an accuracy figure. Results depend on your phrase lists, the greetings your leads' phones play and how you tune, and no detector is perfect. Test in the simulator, then review recordings and transcripts of calls logged as AM and adjust the lists.
A call detected as a machine ends before any transfer. A greeting that slips past the checks is a missed machine: the bot may keep talking to it until a later phrase matches or the call ends. Review short calls and recordings, and add missed greetings to your machine list.
Yes. The same checks run when bots work as VICIdial remote agents, with the AM outcome written back to VICIdial, and when the built-in dialer calls through Twilio, Telnyx or any other SIP carrier.
Yes. Machine and do-not-call lists are kept per organization, and you can edit, export and import them, so a list tuned for one client can be imported into another client's organization and adjusted there.
Not in AI fees: VoiceOI bills prepaid per bot with no per-minute AI fee. Your carrier still bills the minutes those calls use.