A chatbot for a dental clinic: routing emergencies to the phone and booking the rest, without giving medical advice

A dental clinic gets two completely different kinds of messages: the patient in pain who can't sleep, and the patient who wants a cleaning in two weeks. A good chatbot has to treat them differently. Otherwise it either traps emergencies inside a booking form, or floods your phone with trivial questions. Everything hinges on this.

The mistake we see most often is a bot that starts answering medical questions: "my tooth hurts, what should I do?" A bot must never give advice like that, and it doesn't need to. Its job is administrative triage, not clinical: recognize the emergency signals, get them to a human fast, and quietly book the rest. Here's exactly how to build it.

Administrative triage, not diagnosis

The distinction is simple but critical: the bot doesn't tell you what's wrong, it tells you where to go. It never judges the medical severity of a symptom; it sorts it into an operational path.

In practice the bot has only two drawers:

• The fast lane (possible emergency): any message with alarm signals goes straight to a human, with the phone number shown immediately.
• The routine lane: consultation, cleaning, whitening, checkup, braces, second opinion. These book themselves.

The bot never says "it's not serious" or "wait until tomorrow." Even when unsure, it errs on caution: route to a human. A false positive (an emergency that wasn't) costs one phone call. A false negative (a missed emergency) costs the clinic's reputation and, potentially, the patient's health.

The trigger words that fire the emergency lane

Triage that works rests on a concrete list of signals, written with the dentist, not invented by the developer. These are the kinds of phrases that should go straight to a human:

• Trauma: "my tooth fell out," "broke my tooth," "got hit," "accident," "my child fell."
• Swelling and infection: "my cheek is swollen," "abscess," "pus," "fever," "can't swallow."
• Severe pain: "unbearable pain," "can't sleep," "painkillers stopped working."
• Bleeding that won't stop, especially after a recent extraction.

The key is covering how people actually write, not just clinical terms: a patient types "my whole face blew up," not "facial edema." That's why the list is tested against real messages from the clinic's history and expanded continuously.

What the bot actually does with an emergency

Recognition isn't enough. A well-built bot, at the first emergency signal, does three things at once:

• Instantly shows the phone number and a call button: "This looks urgent. Please call [number] now."
• Sends a notification to the clinic (WhatsApp or email) with what the patient wrote, so the human knows the situation before answering.
• Doesn't close the conversation or force the patient through a long form.

If it's outside working hours, the message changes honestly: where to go for overnight dental emergencies, not a false promise to call back in eight hours. A bot that hides the fact it's closed loses trust instantly. Being honest about hours is part of triage.

How it books the rest on its own

For routine messages, the bot can carry the task to the finish without bothering anyone. The flow that works best:

• Ask for the type of service (checkup, cleaning, whitening, bookable non-urgent issue).
• Offer real slots from a connected calendar, not made-up times.
• Ask for a name and a phone number, and that's it. No medical data in the chat.
• Send a confirmation and, ideally, a reminder before the appointment.

Integration is what matters here: if the bot writes into the same calendar the front desk uses, double-booking disappears. If not, you're left with a pretty form that creates extra work. A compliance note: minimize data. Collect only what a booking strictly needs, and state clearly in your privacy policy what happens to it.

Legal and common-sense limits you don't cross

A dental chatbot touches two sensitive areas: health and personal data. A few rules we always keep:

• Zero medical advice. No painkiller dosage, no "it's probably a cavity." The bot explicitly says it doesn't replace a consultation.
• A short, visible disclaimer, plus a clear channel for emergencies.
• GDPR: consent for contact data, a legal basis, and deletion on request. Health data is a special category, so you keep it out of the chat as much as you can.
• No outcome promises ("you'll be pain-free by tonight").

At MPO Web Studio we build bots like this as part of the site, with the triage written together with the dentist and a ready-made demo you see before paying anything. We deliver remotely, nationwide, with transparent pricing. If you'd like to see how it would look for your clinic, message us on WhatsApp and we'll prepare a version.