AI Receptionists for Clinics: What Has to Work (and What to Refuse)
Most “AI receptionist” demos answer a FAQ and stop. A real clinic website gets GLP-1 curiosity at 11pm, after-hours appointment requests, and questions no chatbot should ever answer (dose, eligibility, predicted weight loss). That gap is where generic website chat dies — and where a vertical agent has to be configured, not prompted once. What actually has to work Three jobs, in order: (1) explain the programme (intake, labs, cadence) from pages you already published; (2) capture a callback with a stated SLA; (3) refuse clinical judgements out loud, with a next step instead of a dead end. If the agent invents coverage or says “you would qualify,” you have a compliance incident, not a lead. We tested this pattern on AgentLocalAI’s weight-loss and dental builds. The product pages are public: how the agent is grounded on your site , weight-loss clinic configuration , and pricing . Where a directory still wins Chat does not replace discovery. Patients in British Columbia still com...