AI for medical practices · WhatsApp
WhatsApp for your practice, answered by an assistant that knows its limits.
IndunAI is a WhatsApp assistant for South African doctors and clinics. It tells every patient it is an AI before it asks them anything, books appointments into your diary, hands out your forms, tells patients when their script is ready, and never gives medical advice. Anything it should not handle goes to your team.
The demo runs in your browser. It is the same assistant a practice runs on WhatsApp, answering for a fictional practice, with that practice's diary and request lists updating beside the chat.
online, replies in seconds
Hi, is my repeat script ready?07:58
Before we start: you're chatting with the practice's AI assistant. I can answer questions and help you book. I can't give medical advice or decide on treatment; the practice's own clinicians do that, and you can ask to speak to a person at any time.07:58
Your request came in yesterday and the doctor has not marked it done yet. I will message you here the moment it is ready.07:58
An illustration. The first paragraph is the product's standard disclosure, with the practice's name where it says "the practice's". A practice may word its own, and a wording that never says it is an AI is not used.
The channel patients chose
Why patients already WhatsApp the practice.
Nobody decided that patients would WhatsApp their doctor. They just started, because it is the app already open on their phone, because a message does not need the rooms to pick up, and because 20:30 is when a parent remembers the school form. The trouble is the other end: the practice number sits on a phone at reception, and the messages that arrive during consultations, after hours and over the weekend wait for somebody to have a minute.
A WhatsApp assistant for a practice answers those in the same minute, on the number patients already have. It reads text and transcribes voice notes, it replies in the language the patient opened in, English or Afrikaans, and it can open with the practice's own menu as buttons, so a patient taps rather than types. It answers messages, not phone calls: it does not answer the practice's phone line, and WhatsApp voice calls are refused outright on a medical practice.
The first words
What the assistant says first: that it is an AI.
On a medical practice, the first message a patient receives says they are talking to an AI, what it can do, what it cannot, and that they can ask for a person, before anything asks them for information. The server puts it there, not the model, so it is on the first reply whichever way the conversation started: a greeting, a question, a tap on the menu, an ad.
In the patient's language
An Afrikaans patient reads the disclosure in Afrikaans. The language is taken from the message the patient opened with, so the notice above an Afrikaans reply is never English.
Above the practice's own notice
Where the practice has its own opening notice, the disclosure goes above it and the practice's words are untouched. Where the practice asks patients to accept its privacy and payment policies, the links are on that same message.
And again after a person replied
When somebody at the practice has answered a chat by hand and the assistant takes it back, it says again that it is the AI. That is the one moment a patient has good reason to think they have been talking to a person.
The front desk's messages
What it handles: bookings, hours, forms, scripts ready, directions.
Most of what arrives on a practice's WhatsApp is administration, and administration is what the assistant does. Everything it says about the practice comes from facts the practice gave it, and the things that matter are done by the server rather than promised by the model.
Bookings
New appointments, follow-ups, moves and cancellations, straight into the practice diary, with reminders after. The detail is on patient booking on WhatsApp.
Hours, fees and directions
Opening hours are said by the server from the practice's own hours. Fees, the address, parking and what to bring come from the practice's own knowledge and price list, and a question it has no fact for gets a check-with-the-team answer rather than a guess.
Forms
New-patient forms, and repeat prescription or sick note requests, are handed out as a button to a form the patient fills in on their phone. The details go on the form, not into the chat.
Your script is ready
A patient who sends a prescription or sick note request is thanked by name and told they will hear back. When the doctor marks it done, the practice picks the message: emailed to you, sent to your pharmacy, or ready to collect.
The requests land on a list beside the diary, one row per patient, with Mark done and Cancel on each. A practice can also put a fee on a request and ask for it before the script is made out. And the doctor can WhatsApp the assistant from her own phone, "find prescription today", and read the answer back without opening the dashboard.
Where it stops
What it never does: diagnose or advise.
An assistant in front of a practice's patients must never give medical advice. It books, reminds, collects and hands over. That is the whole job, and on a medical practice it is enforced twice.
The instruction
The assistant is told that clinical detail belongs to the practitioner: side effects, risks, suitability, dosages, recovery and expected results. Where the practice has not written the answer itself, it says the doctor goes through that at the consultation, and carries on with the booking.
The check after the model
An instruction is a request. So on a medical practice the server reads every reply before it is sent, looking for a sentence that gives clinical advice: a call on whether a treatment suits this person, a predicted result, a list of side effects, a medication call, or reassurance about a symptom. Such a sentence is taken out wherever the rest of the reply can stand without it.
It never refuses care
Where one of the practice's own screening answers means the assistant should not book, the patient is passed to the practice's team instead of the conversation simply ending. A patient can ask for a person at any time.
The first question, answered carefully
Meta's WhatsApp policy on health, explained.
Every practice that asks us about this has heard some version of it, usually second hand. Meta publishes eligibility rules for its own business AI agent, and the list of excluded verticals reads:
"all verticals are supported except Finance, Government, Health, Alcohol, Gambling, over-the-counter drugs, and matrimony services"
Meta, Business Agent eligibility documentation, updated 28 August 2026. Read the page yourself.
Health is on that list. That sentence is about Meta's own agent, the one Meta runs for a business inside WhatsApp. The same page scopes itself to accounts on the WhatsApp Business Platform, the Cloud API, and whether the same exclusions reach the free WhatsApp Business app is not documented anywhere we have been able to find. If somebody tells you it definitely does, or definitely does not, ask them where that is written.
IndunAI is a different product: software running on the practice's own WhatsApp Business number through the Cloud API, on the practice's own instance. Meta's general WhatsApp Business terms and messaging policies still apply to every business on the platform, a practice included, so read them and take your own advice on your own position. Nothing here is legal advice. What a practice owes its patients under South African law is a separate question, set out on POPIA, patient confidentiality and an AI receptionist.
The rule above every other rule
Crisis and self-harm messages go to a person, always.
A patient who writes that they want to hurt themselves is not a booking. On that message the assistant does not run the AI at all. It sends a fixed reply carrying the South African crisis lines, and it files one urgent alert to the practice, which skips quiet hours and mutes. It works on a voice note too, which is transcribed first so the words can be checked.
Nothing that normally holds a message back applies to a disclosure: not a block, not a rate limit, not a paused chat. And from then on, in that conversation, the assistant does not carry on talking as though it could help. It keeps pointing to the helplines and to the practice, because counselling somebody is a person's job.
Setting it up
Your practice number, or a new one.
Connected in Meta's own window
The number is connected in Meta's own sign-up window, which opens from the dashboard. Patients message the number they already know and see the practice's name.
Keep using the WhatsApp Business app
An existing number can stay on the WhatsApp Business app at the same time. When somebody replies from the app, the assistant steps back from that chat, and where the practice sets it, it comes back on its own a few minutes after the practice's last reply.
Or a separate number
A practice can put the assistant on a new number instead, or run more than one number on the same assistant. Meta's own message fees are separate from ours: see WhatsApp Business costs.
For the general version of this, outside medicine, see the WhatsApp AI chatbot.
FAQ
Fair questions.
Is WhatsApp allowed for patient communication?
We cannot answer that for your practice, and anybody who answers it in one sentence is guessing. What is published: Meta's eligibility documentation excludes Health from its own business agent platform; the National Health Act makes a patient's health information confidential and requires written consent to disclose it; and POPIA treats health as special personal information that may be processed only on a specific ground. Read those, then take your own advice on your own position.
What if a patient sends a medical question?
It is designed not to answer it. The assistant says the doctor goes through that at the consultation and offers to book, and on a medical practice the server checks every reply for clinical advice before it is sent. The patient can ask for a person at any time. A message about self-harm takes a separate path: a fixed reply with the South African crisis lines and an urgent alert to the practice.
Can it send prescription-ready messages?
Yes. A patient who sends a repeat prescription or sick note request is thanked by name and told they will hear back. When the doctor marks the request done, the practice chooses the message: it has been emailed, it has gone to your pharmacy, or it is ready to collect. It goes in English or Afrikaans, in the practice's own words where it writes them.
Can we reply ourselves in the same chat?
Yes. Anybody at the practice can reply from the dashboard, and a number connected alongside the WhatsApp Business app can be answered from the app as well. Either way the assistant steps back from that chat. Where the practice sets it, the assistant comes back on its own a few minutes after the practice's last reply and says again that it is the AI; a chat the practice wants to keep can be paused for good from the dashboard.
Message the demo practice.
Ask it what your patients ask you: whether there is a time on Thursday, whether your script is ready, what a consultation costs, and whether that rash is anything to worry about. Watch what it does with the last one. When you want it on your own number, book a demo.
The rest of the medical guides, and the general ones they build on.