AI Chatbot for Clinics: Booking, Doctors and FAQs

What a clinic bot's job is, and is not

A chatbot project for a clinic should begin with one sentence: this bot does not give medical advice. It does not assess symptoms, suggest a diagnosis, name a medication, or say 'that does not sound serious' or 'you should see a doctor urgently'. That sounds like a restriction and it is in fact the product definition.

Once the boundary is set, a real and substantial job remains. Most questions arriving at a clinic are not medical: which doctor works when, what a service costs, when test results are ready, which documents to bring, where the branch is, whether children are seen. These consume the reception desk's day and none of them requires a medical judgement.

The practical consequence is to design the clinic bot as a reception desk that works around the clock rather than as a virtual doctor. The success criteria should match: appointments booked, and calls removed from reception.

The booking flow

1. 1Establish the service first, not the doctorMost patients do not know which specialist to see, and asking pushes them into describing symptoms. 'Which service is this for?' is safer than 'which doctor?' and produces the same result.
2. 2Then branch and time preferenceThe branch and the part of the day — morning, afternoon, evening. Those two fields narrow availability sharply and require no medical information from the patient.
3. 3Offer slots from the live scheduleThree concrete times rather than a month grid. The slots must genuinely be free, coming from the system that holds the schedule rather than from an indexed page. A patient booked into a slot already taken is the most damaging error in this sector.
4. 4Collect only what is neededName, phone and whether this is a first visit. Date of birth and other personal details belong at reception or in an authenticated system, not in an open chat.
5. 5Confirm concretelyDoctor, date, time, branch, what to bring and the lateness policy. Vagueness here is the main cause of no-shows.

Doctor and service selection, safely

Patients frequently ask which doctor is right for them. That question requires a medical judgement, so the answer must be information rather than direction.

• Give the list of doctors by service — specialty, days, branch. That is factual and firmly inside the bot's remit.
• Explain neutrally what each specialty deals with. 'A cardiologist deals with the heart and vascular system' is a fact; 'in your case see a cardiologist' is advice.
• If the patient cannot decide, transfer to reception. That is not a failure — it is the correct outcome and should be recorded as one.
• Do not give opinions, comparisons or 'best doctor' statements.
• For paediatric, pregnancy and age-restricted services, ask about age — that is an admission rule rather than a medical judgement.

Questions that can safely be answered

These make up most of a clinic bot's volume, and all of them are factual — not one requires a medical judgement.

• Service prices and what is included.
• Branch addresses, opening hours, weekend schedules, public holidays.
• Which insurers are accepted and which documents are required.
• When test and examination results are ready and how they are collected — the process, never the result itself.
• Preparation rules such as fasting, drinking water or pausing a medication — only as the clinic's own written instruction, quoted rather than paraphrased.
• Cancellation and rescheduling rules.
• Practical questions: parking, accessibility, bringing a child.

Where the bot must stop immediately

This is the most important part of a clinic bot and it belongs in the material as written text, not in a configuration screen.

• Any description of a symptom. The response is that symptoms are not discussed, and an appointment is offered or reception takes over.
• Signs of an emergency. The reply must be concrete and written in advance: state the official emergency number and transfer to a person immediately. This reply must exist in every language.
• Medication, dosage and interaction questions — without exception.
• Interpretation of a test result. 'Is this number normal?' is a medical judgement.
• Anything touching pregnancy, child health or mental health should carry an earlier escalation threshold.
• Disclosure of personal medical information — results, diagnoses, visit history — because identity cannot be verified in chat.

What to measure

1. 1Appointments bookedAppointments originating in a bot conversation. In a clinic this is the headline number, not conversation count.
2. 2Call volume removed from receptionCalls received before and after the bot went live. This is where the benefit is most visible.
3. 3Out-of-hours bookingsAppointments made while the clinic is closed. Usually a larger share of the total than expected.
4. 4Escalation reasonsParticularly the symptom and emergency triggers. A stable figure here shows the boundaries are working; a falling one should be investigated.
5. 5No-show rateDirectly tied to how concrete the confirmation is and whether a reminder was sent, so it is sensitive to bot changes.

How Vexvon works in a clinic

The bot answers from the knowledge base — your site crawled to as many as 4,500 pages, PDFs up to 25 MB and bulk entries at up to 200 per request. Entries are typed into thirteen categories including branch, pricing, opening hours and policy, which is exactly where clinic questions fall, and each can be activated or deactivated individually.

The schedule should not be answered from an index and does not have to be: your own system can be registered as an API tool the agent calls mid-conversation, so free slots come from the system that holds them. When the lookup fails, the defined behaviour is to say it could not be checked rather than to guess.

The stopping mechanisms already exist: there is a defined fallback when no context is found, a request for an operator raises a notification, an agent's stop character pauses the AI on that conversation for thirty minutes, and the AI can be switched off for a conversation entirely. The symptom and emergency triggers are built on top of these.

Channels share one customer record, so a patient who wrote on Instagram, messaged on WhatsApp and then called is one record rather than three. Name, phone and a one-sentence summary are extracted from the conversation, and the lead reaches the reception group as a Telegram notification with a Start button.

Frequently asked questions

1. 1Can a clinic chatbot answer medical questions?No. Symptom assessment, diagnosis, medication and interpretation of results must be entirely out of scope. The bot handles booking, schedules, prices, documents and branch questions — which are most of the volume and require no medical judgement.
2. 2What should the booking flow ask?Service, branch, time preference, name, phone and whether this is a first visit. It must not ask 'what is the problem?', which invites the patient to describe a symptom.
3. 3Can the bot recommend a doctor?It can give information, not direction: the list of doctors by service, their specialty and their days. 'In your case, see this doctor' is a medical judgement and belongs with reception.
4. 4What should happen in an emergency?A pre-written reply available in every language: state the official emergency number and transfer to a person immediately. This reply must never be improvised.
5. 5How should available slots be shown?From a live lookup against the system holding the schedule, offered as three concrete times rather than a month grid. Booking a patient into a taken slot is the most damaging error in this sector.
6. 6How should success be measured?Appointments booked, call volume removed from reception, out-of-hours bookings and no-show rate. Conversation count replaces none of these.

Start with one day at reception

Write down, verbatim, the questions reception receives in a single day, and split them into two columns: those requiring a medical judgement and those not. The second column is almost always far longer than the first, and it is your bot's specification — written in your own clinic's words.