Voicemail drop: pick immediate or delayed detection

ElevenLabs added voicemail detection for Twilio calls on September 7, 2026, with an immediate verdict mode and a delayed mode that leaves a message after the greeting. Pick delayed mode for drops, and record the message ahead of time with Sume's tts_create.

What ElevenLabs shipped

Per the Daily AI Pedia summary, voicemail detection for Twilio calls arrived on September 7, 2026. It also lists DTMF keypad input on August 31, with a # terminator and transcript redaction.

Which mode for which job

• Immediate: you need to know fast whether a human or a machine answered, so the agent can hang up or switch scripts.
• Delayed: you want to leave a message, so the system waits for the greeting to finish before speaking.
• For a voicemail drop, delayed mode avoids the message starting over the beep.

Record the message with Sume

A drop is a fixed message, so synthesize it once instead of generating speech during the call. Sume's hosted MCP exposes tts_create as a paid tool; it needs an idempotency_key, and dry_run=true previews cost. Read the schema with tools_schema and name: "tts_create" for the current voice and format options, then generate and download the file as an asset.

Keep the script short, say who is calling and why, give a callback number twice, and avoid anything that depends on the called person's name unless you generate one file per recipient.

Compliance note

Recorded or synthetic voice messages are regulated in many places. Check consent and disclosure rules for your market before you send them. This post is not legal advice.