He called me on a Tuesday morning to talk about a chatbot for his clinical practice. The first thing he said before I could pitch anything was, "I hate chatbots. The fake-friendly ones especially. I am not paying for software that lies to my patients."
Most chatbot pitches start with "it'll feel just like talking to a real person." That promise is exactly what he was warning me against. So we threw it out and built the opposite.
The bot opens every conversation with one line: "Hi — I'm an automated assistant for Dr. R's practice. I can answer scheduling questions and intake forms, but for anything clinical I'll connect you to a human in under 90 seconds." No emoji. No "hey there!" No pretending.
From there, the bot does three things. It checks insurance acceptance against a small database. It runs the routine intake form. And it watches every message for clinical-sounding keywords — pain, urgent, bleeding, dizzy, anxious — and the second one shows up, it pages the on-call coordinator and hands off the conversation with full context.
We launched it on a Wednesday afternoon. By Friday the practice had handled 142 conversations, completed 38 intake forms, and routed 11 urgent clinical inquiries to a human within their 90-second target. Patient feedback was the surprise: "It told me right away it was a bot — that actually made me more comfortable, not less."
The lesson is the inverse of what the chatbot industry has been selling for five years. The honest, narrow-purpose bot that admits it's a bot consistently outperforms the friendly, broad-purpose bot that pretends to be human. Patients aren't fooled by the pretending. They are, however, deeply relieved by the honesty.
