All notes

Testing an AI receptionist: check the whole path, not the reply

How I tested a clinic's AI receptionist in GoHighLevel: three scenarios end to end and the five things I checked in each, from the AI answer to the reminders.

When people demo an AI receptionist, they usually show the conversation: a patient writes, the AI answers clearly and politely. But the reply is only the first step. For a clinic, what matters is what happens after it.

I built the system for a fictional private dental and aesthetics clinic in the UK. It is a demo account in GoHighLevel for the client's internal tests and sales presentations: an AI receptionist in chat and SMS, three calendars, a seven-stage patient pipeline and six automations.

Three scenarios, because the paths differ. I ran an emergency, an Invisalign enquiry and a routine check-up end to end. Each takes its own route: emergencies get the earliest slot, Invisalign goes to the specialist's consultation calendar and check-ups go to the next free general slot. So a successful run of one scenario says nothing about the others.

Five things in every scenario:

  1. The AI answer: from the clinic's data, with no invented prices or services.
  2. The contact in the CRM.
  3. The deal in the right pipeline stage.
  4. The appointment in the right calendar.
  5. The reminders: 24 hours and 1 hour before the visit.

The AI answer can be flawless while the appointment lands in the wrong calendar, and the patient finds out at the clinic. That is why the conversation is only the first of my five checks.

What I build in before testing. Some mistakes are easier to prevent than to find.

For Telegram and a web chat I built a second agent, in n8n, on the same GoHighLevel account. Its design follows the same principles:

The end-to-end tests above cover the GoHighLevel path.

Handover. The client received a video walkthrough and documentation for copying the account to a new clinic.

Full caseAI receptionist and patient journey for a dental clinic
Discuss your project