Building AI Phone Agents That Don't Sound Like Robots

We built Mercury our AI phone avatar and learned a lot about what makes callers actually enjoy talking to an AI.

The Voice Matters Most

We tested 15+ voices. The winner was not the most professional one it was the one with slight vocal fry and a casual cadence. Calls went from 30 seconds to 3 minutes.

Hand Gestures Matter

Video-based AI avatars with natural hand movements hold attention 2x longer than static talking heads. Mercury uses a landscape full-body setup with casual hand gestures.

Fallback to Human

Every AI phone system needs a clear, fast handoff to a human. If a caller asks are you a robot three times, a warm transfer triggers immediately. This eliminated 90% of negative feedback.