Why most AI voice agent prompts fail in the real world, and three practical fixes that make them hold up on actual customer calls.

Andy Norman from Antek Automation shares practical tips sourced from the blog post: How to Write Prompts for AI Voice Agents That Actually Work.

Write a tightly defined system role with explicit constraints: what the agent is, what it can and cannot do, response length capped at ten seconds per turn, and named failure modes like angry callers, pricing questions, and repeat callers.

Design prompts around the data you need, not a rigid question sequence. Track what the caller has already given you and fill the gaps in whatever order they arrive, because real callers will give you their postcode, problem, and availability all in one breath.

Treat your first prompt as a draft. Review call transcripts, find where it breaks, and iterate. Build a library of edge cases to test against every update, and keep refining until it holds up under real-world noise, interruptions, and accents.

Produced by Antek Automation — AI automation for UK small businesses.