Blog
Guides and insights on AI voice agents, call centers, and calling at scale.
Telephony & Dialing
Answering Machine Detection and Voicemail Drops: How They Work
Published
In outbound calling, a large share of calls are never answered by a person. Many reach voicemail, others ring out, and some are picked up by automated systems. For a business making calls at scale, that means a lot of time spent listening to voicemail greetings rather than talking to people.
Two technologies were developed to deal with this: answering machine detection, which tries to tell whether a person or a machine has answered, and voicemail drops, which leave a prerecorded message automatically when a machine answers. Both are widely used, both have real limitations, and both are subject to rules that businesses need to understand.
What Is Answering Machine Detection?
Answering machine detection (AMD) is a feature in many dialers and calling platforms that analyzes the first few moments of an answered call to decide whether a live person or a voicemail system has picked up.
Depending on the result, the system can connect the call to a caller if a person answered, or handle it automatically if a machine answered, for example by ending the call, scheduling a retry, or leaving a prerecorded message.
The aim is to make sure the people making calls spend their time in real conversations rather than listening to greetings.
How AMD Works
When a call is answered, the AMD system listens to the audio and looks for patterns that typically distinguish people from machines.
The shape of the greeting
People usually answer with a short greeting, such as "Hello?", followed by a pause while they wait for the caller to speak. Voicemail greetings tend to be longer and continuous: "Hi, you've reached Sarah. I can't come to the phone right now, so please leave a message after the tone." AMD measures how long the initial speech lasts and whether it's followed by silence.
Silence and timing
The system looks at gaps in speech. A short greeting followed by silence suggests a person. Long stretches of uninterrupted speech suggest a recording.
The beep
Many voicemail systems end with a tone before recording begins. Detecting that beep helps confirm a machine has answered and marks the moment to start a voicemail message, if one is being left.
Machine learning
Newer AMD systems use machine learning models trained on large numbers of recorded call openings. Rather than relying only on fixed rules about timing, they analyze the audio itself to make a decision, which can improve accuracy, particularly in difficult cases.
The Limitations of AMD
AMD is useful, but it's never perfectly accurate, and its mistakes affect real people.
Mistaking a person for a machine
When someone answers with a longer greeting, such as "Hello, this is James speaking, how can I help?", AMD may decide it's a voicemail. The system might then hang up on them or play a recorded message meant for a machine. From the person's point of view, they answered a call and were cut off or spoken at by a recording. These errors are often called false positives.
Mistaking a machine for a person
A short voicemail greeting, or one that simply says the person's name, can be mistaken for a live answer. The call is then connected to a caller who finds themselves listening to a voicemail. These are called false negatives. They waste the caller's time but don't directly affect the person being called.
The silent pause
AMD needs a moment to listen and decide. During that time, the person who answered may hear nothing at all. Even a delay of a second or two can feel strange, and many people hang up immediately, assuming it's a robocall or a scam. This short silence is one of the most noticeable side effects of AMD, and one reason some businesses choose not to use it.
Accuracy trade-offs
Most AMD systems can be adjusted to be more cautious or more aggressive. A cautious setting connects more calls to callers, reducing the chance of hanging up on real people, but lets more voicemails through. An aggressive setting filters out more voicemails but increases the risk of misjudging real people. There's no setting that eliminates both kinds of error.
AMD and the Rules on Abandoned Calls
Because AMD can cause people to answer to silence or to be disconnected, it's closely linked to the rules on abandoned calls.
In the United States, the Telemarketing Sales Rule requires that when a telemarketing call is answered by a person, they're connected to a live caller within two seconds of completing their greeting, or the call counts as abandoned. The delay AMD introduces can push calls past that limit.
In the United Kingdom, Ofcom's guidance treats calls where AMD wrongly identifies a person as a machine as abandoned calls, and expects businesses to include a reasonable estimate of these in their abandoned call rate, which must stay within regulatory limits.
Businesses using AMD need to understand how it affects their abandoned call figures and set it up accordingly.
What Is a Voicemail Drop?
A voicemail drop is a prerecorded message left automatically when a call reaches voicemail. Instead of the caller waiting for the beep and speaking, the system plays a recorded message and moves on to the next call.
Voicemail drops are usually triggered in one of two ways: automatically, when AMD detects a machine and waits for the beep, or manually, when a caller hears a voicemail greeting and clicks a button to leave a pre-recorded message while they move on to their next call.
Why businesses use them
Voicemail drops save time and keep messages consistent. Every message is clear, accurate, and delivered the same way, rather than varying with each caller's mood or energy. For high-volume calling, they can turn unanswered calls into a message the person hears later, rather than a missed call with no context.
Ringless Voicemail
A related technology is ringless voicemail, where a message is placed directly into someone's voicemail box without their phone ringing at all.
Ringless voicemail has been controversial. Businesses argued it wasn't really a call, since the phone never rings. In 2022, however, the US Federal Communications Commission ruled that ringless voicemail messages are calls under the Telephone Consumer Protection Act. That means the same consent rules apply as for other prerecorded calls.
The Legal Side of Voicemail Drops
Because a voicemail drop is a prerecorded message, it's generally treated the same way as any other prerecorded call under the law.
In the United States, the Telephone Consumer Protection Act requires prior express consent for many calls that use prerecorded or artificial voices, and prior express written consent for prerecorded telemarketing calls to residential lines and mobile phones. Prerecorded messages must identify the business at the start and include a phone number, and prerecorded telemarketing messages must offer an automated way for the person to opt out of future calls. State laws may add further requirements.
In the United Kingdom, automated marketing calls, including recorded messages, generally require the person's specific prior consent under privacy and electronic communications rules.
The details vary by country and by type of call, so businesses using voicemail drops should make sure their consent records and message content meet the requirements wherever they call.
Writing a Good Voicemail Message
A voicemail message has only a few seconds to be useful. Good messages tend to share a few qualities:
- They identify the caller clearly, with the person's name and the business name at the start.
- They state the reason for calling in a sentence.
- They're short, usually under 30 seconds.
- They give a clear way to respond, such as a callback number, said slowly and ideally repeated.
- They sound natural, recorded in a clear, friendly voice rather than read hurriedly from a page.
- They're honest, avoiding misleading urgency or vague statements designed to provoke a callback.
- They meet legal requirements, including any required identification and opt-out instructions.
Alternatives and Newer Approaches
Some businesses choose not to use AMD at all, connecting every answered call to a caller and letting them leave a voicemail personally when needed. This avoids silent pauses and misjudged calls, at the cost of some efficiency.
Others are turning to AI voice systems, which can respond to whoever answers in real time. Instead of making a split-second decision about whether a person or a machine has picked up, these systems can begin the conversation naturally, recognize a voicemail greeting as it plays, and adjust accordingly, either leaving a message after the tone or continuing the conversation if a person is on the line. The same rules on disclosure, consent, and prerecorded or artificial voices still apply.
Getting the Balance Right
Answering machine detection and voicemail drops exist to make outbound calling more efficient, and used carefully, they do. But both involve trade-offs that affect the people being called: silent pauses, misjudged greetings, and recorded messages they may not have agreed to receive. Businesses that use these tools responsibly configure them with care, keep a close eye on abandoned calls, make sure they have the consent they need, and remember that every call, answered or not, reaches a real person.