Clinic AI training: a step-by-step guide

A clinic’s inbox is always full of the same questions: "How much does a hair transplant cost?", "When do you have an opening for fillers?", "Should I come to the first examination on an empty stomach?", "How many sessions does it take?". Aesthetics, dental, hair transplant, physiotherapy, dietitian: whatever the specialty, the same twenty questions come in a hundred times a day. While reception tries to keep up with the phone and WhatsApp, dozens of "price?" comments pile up under a before/after post, and a patient writing in English from abroad waits for hours. Every appointment request answered late goes to another clinic.

inTusell takes over this load. But to do that, it first needs to learn your clinic, your services, and your operating routine. This article explains how a clinic trains the inTusell AI from scratch, step by step. This is the health chapter of our sector-based "how to train your AI" series, and it is the series’ main article.

Quick answer

Why does training matter?

inTusell is not an off-the-shelf chatbot; it is an AI assistant that behaves like an experienced receptionist and patient coordinator. No good coordinator starts on day one without knowing the clinic: they need to know which services you offer, your appointment routine, and which questions to leave to the physician. Training teaches the AI to do two things well. First, give the right information: your services, preparation instructions, working hours. Second, stop at the right boundary: leave medical matters to the physician, and never promise outcomes.

An untrained assistant either gives overly generic answers or guesses on a topic it doesn’t know. In healthcare, guessing is dangerous. Wrong preparation information wastes an appointment, and a boundary-crossing line like "don’t worry, it’s a sure fix" both misleads the patient and violates regulations. A well-trained assistant stops when it doesn’t know and directs the patient to the physician or the live team.

Who is it for?

This guide is for clinics whose inbox is dominated by appointment requests and who answer the same questions over and over again:

• Dental, oral health, and orthodontics clinics
• Aesthetics, medical aesthetics, and dermatology clinics
• Hair transplant centers (often with international patient traffic)
• Appointment-heavy specialties such as physiotherapy, dietitian, ophthalmology, and polyclinics
• Teams receiving dozens of appointment requests a day from Instagram and WhatsApp
• Centers that need to communicate in multiple languages for health tourism

You don’t need a technical team. All of the steps below are completed from sections in the panel, without writing code.

The structure of training: 5 core steps + 2 continuous steps

Splitting the training into two groups makes your job easier. The first 5 core steps get the assistant up and running; these steps are the one-time part of the setup. The 2 continuous steps that follow (reviewing and turning fixes into rules, and opening channels) are the ongoing part that you sharpen as you use the assistant. You’ll find all of them in order below.

1. Upload the knowledge base

The foundation of training is the knowledge base (RAG). Under My business › Knowledge › Knowledge Base, you upload your clinic’s information. Each time the AI generates a response, it automatically searches these sources and uses only the information written there.

Supported source types:

Every file you upload is automatically chunked and made searchable with pgvector. When you upload a "pre-implantation preparation" document, the AI finds the right paragraph and responds when a patient asks "Can I smoke before the procedure?".

What a clinic must definitely upload: service and procedure descriptions, pre/post-procedure preparation instructions, working hours, team and physician information, payment/communication policy, and the most frequently asked questions. An important limit: the knowledge base is not for diagnosis-containing text. The AI uses the information here to answer "what do you offer, how do you prepare" questions; it leaves medical questions like "I have this, what should I do?" to the physician.

2. Set up the appointment system and working hours

The knowledge base covers "what you offer"; the appointment system lets the assistant give the patient an actual time. In tourism this step was the price catalog; in a clinic the equivalent is the appointment infrastructure. In the panel you define three things:

Once these three are defined, the AI no longer invents times; it suggests appointments based on your working hours and full slots. The system blocks two appointments at the same time (double-booking). If Google Calendar is connected, each new appointment is also copied one-way to the team’s calendar.

We’ve gathered all the details of the appointment engine (durations, break times, conflict prevention, reminders, and calendar sync) in a separate article: clinic appointment automation, calendar, and reducing no-shows. You can also see the service and appointment capabilities on the solutions page.

3. Set the persona and tone

You taught the AI what to say in the first two steps; now you’ll define how it will say it. There are two settings in the panel:

• Assistant name: The name it will introduce itself with to the patient. Most clinics choose a reassuring name consistent with their coordination team.
• Tone: For health, the default tone is professional; warm but measured, calm and reassuring. Emoji use is off and sales pressure is kept low.

Health mode is deliberately more cautious than other sectors. Even the greeting message points the patient in the right direction. A typical opening reads like this: "Hello. How can I help you? For medical advice, please consult your doctor." That single line carries both the clinic’s tone and what the regulations require.

A multilingual note: the AI replies in whatever language the patient writes in. For messages arriving from abroad in English, Arabic, or German to hair transplant, dental, and aesthetics clinics, you do nothing extra. The assistant detects the language on its own and answers in it while keeping the health tone. Fixed system messages such as reminders go in Turkish to Turkish speakers and in English to everyone else.

4. Define the response rules and health limits

The persona defines "who it is," and the response rules define "how it will behave." In a clinic, this step does two jobs at once: conveying your sales discipline and drawing the health limits. With the AI Manager Chat, you add rules by writing in plain language, just as you would instruct a teammate; you see the rules in AI Rules and switch ready rule packs on or off in the Behavior Library.

Typical rules for clinics:

• "Before the appointment, ask about the patient’s request and suitability, but do not diagnose."
• "When asked about the treatment fee, do not quote a price; say the fee will be finalized after the examination and direct them to an appointment." (Can be changed according to your clinic policy.)
• "Do not guarantee outcomes; do not use phrases like 'sure fix/guaranteed.'"
• "In a medical complaint or emergency, direct the patient to the physician or, if necessary, the emergency service."

The most common mistake in healthcare is misleading the patient, and running afoul of regulations, with a single boundary-crossing sentence. In inTusell, this boundary is not left to the rules you write alone. First, the sector instructions of health mode tell the AI not to use such sentences. On top of that runs a built-in output filter: misleading health claims such as "guaranteed treatment," "certain cure," "cures cancer," "heals diabetes," or "Ministry of Health approved" are flagged and logged by a pattern scan and a second-model check. The filter is a flagging layer, not a guarantee that stops every sentence; the real behavioral control sits in the instructions and your rules.

To compare two AI tones, you can use the A/B test, which is in beta: for example, a more formal and a friendlier tone run side by side, each on half of the replies, and the experiment card shows which one leads to more wins. Today only the tone changes; the content and the rules stay the same.

5. Teach from past conversations

This is the step that moves training from "good" to "specific to your clinic." In the panel, you upload audio recordings of your past patient conversations (MP3, MP4, WAV, M4A) and mark each one as Won or Lost, for example, whether the patient came to the appointment and accepted the treatment, or was lost.

The system uses these recordings in two ways:

This way, the AI learns how your best coordinator put the patient at ease and moved someone who said "let me think about it" toward an appointment. Uploading requires explicit consent; phone numbers, emails, ID and card numbers are masked in the transcript before processing. Because health data is special-category, this matters even more in healthcare.

This step is not mandatory, but don’t skip it. The knowledge base teaches the AI "what" it knows; past conversations teach "how you coordinate." When the two are combined, the assistant truly resembles an experienced patient coordinator.

6. Review and turn fixes into rules (ongoing step)

The first five steps get the AI up and running. What follows is the continuous part that sharpens it over time. You check replies in two places: in AI Test, by asking questions the way a patient would, and in Conversations, by reading real chats.

Say the AI explained a preparation instruction incompletely, or referred to a service by the wrong name. You add the missing fact to the knowledge base or write a rule in AI Rules (for example, "when asked about preparation, pass on every item of the instruction"); the next reply works from that source. If you want to see each reply before it goes out, put the account in Full approval mode, or a single conversation in Approval mode.

The Label Manager is a quality record: customer satisfaction replies, the outcome of a handed-off conversation, won/lost classifications and labels your team adds by hand land there for your approval. Approvals and corrections are kept as a record; today they do not change how the AI writes replies on their own. In the first weeks, we recommend a short look every day.

7. Open the channels (continuous step)

Once training is ready, you put the assistant in front of patients. inTusell gathers all channels in a single inbox: Instagram DM, Instagram comments, WhatsApp, Telegram, and Web Chat.

Channel-opening methods:

• Instagram: connects through Meta login; you pick your Facebook Page and Instagram account. The account must be a professional account linked to a Facebook Page.
• WhatsApp: under My business › Channels › Messaging Channels you click "Connect with QR" and scan the code with your phone; your number stays the same and no Meta Business approval is required. The official WhatsApp line (Cloud API) opens after Meta approval; today you connect with a QR code or with an 8-character linking code for your phone number.

You control each channel separately. Once Instagram is connected, DM replies are already on. Comment replies have their own switch on the Instagram card: "Instagram comments", on by default. So you can switch it off, automate only DMs at first and keep the comments under before/after posts manual, then switch it back on when you’re ready. If you want the AI to stop answering a channel altogether, turn that channel off in the "Per-Channel AI Control" list in the AI settings.

The account mode decides how independently the AI works. A new healthcare account starts in Full approval mode (below). In Intern mode the AI sends replies itself and asks you before a cancellation or refund. In Senior mode it also handles cancellations and refunds itself. In Full approval mode every reply on WhatsApp, Instagram and Telegram, and any action the AI would take, waits for your approval in Decisions › Approval Queue; it is a good start for the first days. In every mode, conversations requiring a medical decision are handed to the team and land on the Decisions screen with the full history. For the details of Instagram and WhatsApp automation: Instagram and WhatsApp appointment automation for clinics.

How long does training take?

How long it takes depends mostly on whether your documents are ready. The steps of a working setup:

1. Uploading the first files to the knowledge base
2. Appointment types and working hours
3. Persona, tone, and first response rules
4. Connecting channels

But there is no moment when "training is done." In the first weeks, as you add what you notice in AI Test and Conversations to the knowledge base and the rules, the assistant gets closer to how your clinic works. Uploading past conversation recordings and trying tones with the A/B test are ongoing improvements too. Setup is done once; mastery is continuous.

What it isn’t

inTusell is an AI sales assistant and CRM that answers messages on WhatsApp, Instagram and web chat, books appointments and opens orders, and follows each customer through the sales pipeline; it is not a simple chatbot. Putting it in the right category matters, because the wrong expectation leads to the wrong setup.

• It is not a physician or a diagnostic tool. It does not diagnose, recommend treatment, or give medication dose/recommendations. It directs medical questions to the physician and the clinic.
• It is not a hospital information system (HIS) or an electronic medical record (EMR). It does not keep medical records; it is an appointment, communication, and patient coordination layer.
• It is not a marketing tool that promises outcomes. Phrases like "certain outcome/guaranteed" are forbidden; the sector instructions tell the AI not to use them, and the output filter flags such phrases.
• It is not an intermediary that replaces the clinic. The clinic always provides the service and issues the invoice; inTusell is not a party to payment.

In short: not a bot that gives canned answers, but a patient assistant that represents your clinic to the extent you train it and is set up to stay on the medical boundary.

Frequently asked questions

How long does clinic AI training take?

A working setup comes down to these steps: uploading the knowledge base, appointment types and working hours, persona and tone settings, and a few response rules. How long it takes depends mostly on how ready your documents are; the real refinement comes in the first weeks, as you add what you notice in AI Test and in real conversations to the knowledge base and the AI rules. Training is not a one-time job; it is continuous.

Does the AI give patients medical advice or a diagnosis?

No. In health mode the AI does not diagnose, recommend treatment, or promise outcomes; it directs medical questions to the physician and the clinic. The primary safeguard is the "do not give diagnosis/treatment advice" instruction that health mode gives the AI; in addition, an output filter (a pattern scan plus a second-model check) flags misleading phrases such as "guaranteed treatment" or "certain cure."

Do I need technical knowledge to train the AI?

No. All training is done from the panel; no code, API key, or developer is required. You upload PDF/Excel files to the knowledge base, enter appointment types and working hours, and write response rules in plain language. You connect Instagram through Meta login and WhatsApp by scanning a QR code.

How is patient data protected under KVKK?

Health data is special-category personal data under KVKK. Data is stored encrypted and each business’s data is kept separate; the AI never saves high-sensitivity health fields without confirmation; consent is recorded per channel and purpose, with its time. "Legal notices" sends your patients a KVKK notice with your own privacy link.

Does the AI quote an appointment price or treatment fee?

That depends on your clinic policy and health regulations. It does not market treatment fees on public channels; in a typical setup it defers the fee until after the examination/consultation or directs the patient to the live team. When it is unsure, it does not invent a price.

Does the AI create the appointment on its own?

Yes. The AI recommends the appropriate service and creates the appointment while accounting for working hours and full slots; double-booking the same time is prevented by the system. In cases requiring a medical decision or complex situations, it hands the conversation off to the live team.

Next step

You’ve trained your assistant; next comes using it in daily operations. The next article in the series explains how to run the inTusell you trained on a real workday: incoming appointment requests, calendar, reminders, reducing no-shows, and handing off to the team. Continue straight from there: how a clinic uses inTusell.

If you’d like to see it live before starting the setup, use Request a demo and we’ll open your panel together in a short session, or write to hi@intusell.com. You can see the ready-made clinic setup on the clinic solution page; for package and quota details, you can check the pricing page, and for other guides, take a look at the all articles list.