Why "sounding confident" is the actual risk
Generative AI chatbots are fluent by default. Ask one a question it doesn't have a real answer for, and left unchecked it will still produce a confident, well-formed sentence — because generating plausible text is what the underlying model is built to do, not verifying the text is true. A human receptionist who doesn't know something usually says so. A badly-designed chatbot often won't, and a visitor reading a chat window has no way to tell the difference between an answer the bot knows and one it made up on the spot.
For most businesses that's an embarrassing screenshot. For a dental or health clinic it's worse, because the questions people actually ask a clinic chatbot — what will this cost, does my insurance cover it, is this serious — are exactly the ones where a wrong answer causes real harm: a price nobody agreed to honour, an insurance claim that turns out to be declined, or something that reads as medical advice from a system that was never licensed to give any.
The grounding model: three tiers, not one big prompt
The fix isn't a longer list of rules bolted onto a chatbot after the fact — it's deciding, before the bot ever answers a message, which category each possible answer falls into. A clinic AI built this way only ever has three moves available to it:
What a treatment like a cleaning, filling, root canal, crown or implant generally involves, in plain terms. Never framed as a recommendation or diagnosis for the specific person asking.
Services offered, the booking link, WhatsApp number, and whatever the clinic itself has explicitly confirmed — hours, location, insurers accepted, languages, payment plans, emergency policy. If the clinic hasn't given it, the bot doesn't state it.
Exact price, insurance confirmation, a diagnosis, a doctor's name or rating, a guaranteed appointment slot. Deferred to the team every time, and used as the moment to capture the patient's number instead of losing the conversation.
The list: what a clinic chatbot should refuse to say
Put into practice, that grounding model produces a short, specific list of the questions patients actually ask — and the safe way to handle each one without shutting the conversation down.
| Patient asks | Must NOT do | Says instead |
|---|---|---|
| "How much will this cost?" | Quote an exact price or firm range | Explains pricing depends on the visit, offers to book so the team can quote it precisely |
| "Does my insurance cover it?" | Confirm or estimate coverage | Flags it for the front desk, captures contact details to follow up |
| "I'm in a lot of pain — is this an emergency?" | Diagnose or judge severity | Points straight to the clinic's emergency contact, still offers the next available slot |
| "What's wrong with me?" | Offer a diagnosis or treatment recommendation | Explains what a treatment generally involves, never framed as advice for this person |
| "Is there a lady dentist available?" | Guess if it wasn't explicitly confirmed | States it only if the clinic gave that fact; otherwise says the team will confirm |
| "Are your hours different during Ramadan?" | Assume standard Ramadan hours | Defers to whatever hours the clinic actually confirmed, or flags it for the team |
| "Can you guarantee me a 6pm slot today?" | Confirm a booking the system hasn't verified | Captures the preferred time, hands it to the team or booking link — doesn't confirm on the spot |
A chatbot that says "I can't help with that" at the first hard question loses the patient to whichever clinic answers next. The point isn't to refuse and stop — it's to turn every deferred question into a captured contact: "I can't confirm insurance, but I'll get the team to check — what's the best number to reach you on?" The patient still gets forward motion, and the clinic gets a lead instead of a hallucinated promise.
The legal case that makes this concrete
This isn't a theoretical risk. In February 2024, Canada's Civil Resolution Tribunal ruled against Air Canada after its website chatbot told a customer he could apply for a bereavement discount retroactively — advice that contradicted the airline's actual policy. Air Canada argued the chatbot was "a separate legal entity responsible for its own actions." The tribunal rejected that outright: a company is responsible for all the information on its website, whether it comes from a static page or a chatbot — Moffatt v. Air Canada, 2024 BCCRT 149.
That case was about a refund policy, not a medical claim, but the principle transfers directly: whatever a clinic's chatbot tells a patient, the clinic is on the hook for — which is precisely why an exact price, an insurance confirmation, or anything that reads as clinical advice should never come from a system that's guessing.
It's also why the no-medical-advice rule matters legally, not just for UX. Remote clinical consultation is governed by telehealth licensing rules aimed at licensed practitioners giving diagnoses — a booking-and-enquiry chatbot that stays strictly out of medical advice sits outside that entirely. The moment a chatbot starts diagnosing or recommending treatment, that exemption disappears. For the fuller UAE-specific picture — MOHAP/DHA advertising rules and PDPL consent requirements — see our legal guide.
Frequently asked questions
Can a dental clinic chatbot quote an exact price?
Can an AI chatbot confirm my insurance is accepted?
What happens if a clinic chatbot gives wrong medical information?
Is a clinic chatbot allowed to give medical advice in the UAE?
How does CareLayer stop its AI from hallucinating?
What should a clinic chatbot do with a genuine emergency message?
See the guardrails in a live chat
CareLayer's founding-clinic pricing: a 30-day pilot at AED 2,500 up front, then AED 800/mo (first 3 clinics). Standard rate after: AED 6,000–9,000 up front, AED 1,500/mo. Full pricing and a live demo on the homepage.
See pricing →Published 14 August 2026 · Last updated 14 August 2026 by Ahsan Ishaq, Ahsomatic. The Air Canada tribunal decision is a matter of public record (Moffatt v. Air Canada, 2024 BCCRT 149); this page is general guidance, not legal advice — confirm clinic-specific compliance questions with a UAE-qualified lawyer.