Intel / article

Voicemail Detection

Source: elevenlabs.io
Author: ElevenLabs editorial sitemap
Date: Published Sep 22, 2026

Context

When a voice places outbound calls, every call that lands in a voicemail box still costs credits: the agent waits through the greeting, may get stuck in recording loops, or delivers its whole pitch to nobody. ElevenLabs' August 14, 2025 post announced voicemail detection for its Conversational AI agents (the product ElevenLabs now markets as ElevenAgents) to stop spending credits on them once a machine is detected.

Detection is handled by the same already running the conversation, which watches for cues such as automated greeting structures, phrases like an invitation to leave a message after the beep, standard voicemail wording and the absence of interactive replies. When it decides it has reached a machine, it calls a built-in voicemail detection tool, and the agent either plays a custom message or ends the call at once. The current documentation adds that the tool call records a reason, that the custom message can include per-call values such as a name or appointment time, that the call ends automatically afterward, and that detections appear in conversation history and batch-call results.

ElevenLabs says detection happens within seconds and adds minimal latency because it reuses the model already processing the call. The post's thousand-call campaign with 30% voicemail is an illustration, not a measured result, and neither the post nor the documentation reports how often the tool mistakes a live person for a machine or misses a voicemail.

Key takeaways · AI-distilled

ElevenLabs says detection runs on the agent's existing LLM, which watches the call for voicemail cues: automated greeting structures, phrases like "please leave a message after the beep", standard voicemail-system language, and a lack of interactive responses.
Per the post, detection happens within seconds of a voicemail greeting, and because it reuses the LLM already processing the conversation, ElevenLabs describes the added latency overhead as minimal.
The post illustrates the cost with a hypothetical: a thousand-call outbound campaign where 30% of calls reach voicemail means hundreds of conversations spent on greetings, recording loops or full pitches to empty inboxes.

Terms in this piece · Glossary

AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.

Read the source elevenlabs.io