Hi all, We're running an outbound calling campaign The problem: Carrier-based voicemail systems that read the phone number aloud (e.g., "4 5 6 5 3 1 8 8 7 is not available") are not being detected as voicemail. The transcriber picks this up as user speech, and our assistant starts conversing with the voicemail system. What we've tried: All voicemail detection providers: Vapi, Twilio, Google, OpenAI — none catch this Twilio voicemailDetectionTypes settings — caused worse issues (random hangups) V oicemail tool with prompt instructions to detect number sequences — LLM can't process fast enough before responding Structured system prompts with explicit voicemail indicators Current waitSeconds is 2.5s, but increasing further creates awkward pauses for real humans Is there a way to improve detection for this specific pattern (phone numbers being read aloud by carrier voicemail)? Or is this a known gap in the detection models?

This is a known gap — carrier voicemails that read numbers sound like humans, so most detectors miss them. The issue is the assistant responding before there’s enough confidence in the transcript. What’s worked for me is a short pre-response gate that checks early text for high digit density and cadence, then exits if matched. Do you control the first-response logic, or is Vapi/Twilio handling that entirely?

We wait for the customer to speak first and then we dont have a specific first message, but instead allow the LLM from Vapi to draft something for us