Five tasks that take time in a clinic
This article is part of the series "AI for business by industry". It answers the same five questions as the others, but in a clinic the answer to the fourth question of the frame, "what not to do", is longer than all the rest put together. The tasks are:
- Bookings and appointment reminders. A patient does not come, and the doctor's hour stands empty.
- Standard questions to the front desk. Address, opening hours, consultation price, whether a referral is needed, how to get an appointment.
- Website and social texts. Service descriptions, answers to "what is…", posts.
- Replies to reviews on maps. Thanks or an explanation with no treatment details.
- Internal documents. Instructions for administrators, shift schedules, letters to suppliers.
A sixth task is offered most often: "transcribe the doctor's voice and write the notes in the patient file". It is the only one that touches medical information itself. The next section shows why we do not treat it as simple.
Three classes of data: which one a piece of information falls into
In a clinic it is easier to decide by what goes into the assistant than by the task. We split the data into three classes.
✓ — allowed; ≈ — allowed on a condition; ✗ — do not give to a chat assistant. The classification is ours, built on the provisions cited below; it is not legal advice.
The line between the second and third class is thinner than it looks. A name with a phone number is ordinary personal data. But a name with the date of an appointment with an oncologist or a venereologist already reveals that a person sought help of a particular kind. Article 40 of the Fundamentals of Healthcare Legislation, which sets out medical confidentiality, protects information "about illness, medical examination, inspection and their results", and such a combination may fall under it. This is our cautious reading, not settled practice; so the rule for a clinic is simple: patient data never goes into a chat assistant, and if it cannot be avoided it is anonymised so that nobody can recognise the person from the text.
Why "doctors work with data too" is not a basis
The general rule is strict. Article 7 of the Personal Data Protection Law prohibits processing data concerning health. The exceptions are listed in part two. The only one relevant to a clinic's work is item 6: processing is needed to establish a diagnosis, provide care or treatment or deliver medical services, provided the data is processed by a medical worker, a rehabilitation specialist or another person of a healthcare institution, or a sole trader holding a licence for medical practice and its employees, who are obliged to protect personal data and are subject to the legislation on medical confidentiality.
That list names people of the clinic. A chat-assistant vendor is not named in it. From this we conclude that the working approach is: patient data stays in systems where people from the list process it, and only what is not health information goes into the chat assistant. This is our reading of the provision, not the opinion of a specialist lawyer; if a clinic wants to go further, the decision should be taken with a lawyer and a separate contract.
Two further limits apply if a service does process patients' personal data, for example contacts for reminders:
- Under Article 4 of the Law, if the clinic entrusts processing to another party, a written contract is required, and the processor may process data only for the purpose and to the extent set out in the contract. The standard terms of a public service do not replace it.
- Under Article 29, transfer of data to foreign parties is allowed only if that state ensures adequate protection; the list of states is set by the Cabinet of Ministers. Where the service's servers are and whether that state is on the list must be established before connecting.
Medical confidentiality works separately. Article 40 of the Fundamentals of Healthcare Legislation forbids medical workers and other persons who learned the information through their professional or official duties to disclose it, except in cases provided by law. By the letter of the article, that includes an administrator who sees a schedule with names. So pasting "patient Ivanenko, tooth 36, pulpitis" into a chat is handing information about illness to a third party, not a question of model-training settings.
Can I paste this text into a chat assistant
Before every paste the employee answers four questions. If the answer to any of them is "yes", the text does not go into the chat.
- Does the text contain a surname, phone number, date of birth or any other sign that identifies the patient?Even without a diagnosis this is personal data, and with a diagnosis it becomes health information. Yes → do not paste; anonymise or use a template
- Does it contain a diagnosis, complaints, a drug, a test or scan result, or a procedure prescribed to a specific person?This is information protected by medical confidentiality. Yes → do not paste under any conditions
- Does the text advertise a method of treatment or diagnostics?Then it falls under Article 21 of the Advertising Law, and a doctor must check every statement. Yes → the assistant drafts, the doctor approves
- Will the reply reach a patient with no human in between?A bot may not give advice about symptoms, drugs or doses. Yes → only schedule, address, price, booking and hand-off to a person
Website texts: what Article 21 of the Advertising Law forbids
Texts about services are the part of a clinic's work where an assistant really helps. But the draft it produces by default is easy to write in exactly the way the law forbids. Article 21 of the Advertising Law (edition of 31.07.2026) regulates advertising of methods of prevention, diagnosis, treatment and rehabilitation. Its parts six and seven forbid, among other things:
| What the assistant writes by default | What Article 21 says |
|---|---|
| "The treatment is safe and gives a guaranteed result" | Statements that the therapeutic effect is guaranteed are forbidden, as are references to a method as the most effective or the safest |
| "Patient Olga: after the course the pain went away" | References to specific cases of successful use are forbidden, as are letters and thanks from individuals describing the result |
| "Our doctor recommends this method" | Recommendations by medical workers, medical institutions and organisations regarding the advertised service are forbidden |
| A photo of a doctor in a white coat next to the ad for a method | Participation of doctors and other medical professionals, and of persons made to look like doctors, is forbidden |
| "Do not delay, or it will get worse" | Statements that provoke fear of falling ill or of worsening one's health because of refusing the advertised service are forbidden |
| Tips on how to determine a diagnosis yourself from symptoms | Statements that encourage self-diagnosis and self-treatment with the advertised goods are forbidden |
The practical conclusion: AI-written testimonials and "patient stories" do not suit a clinic's website even as an "illustration". The same goes for a doctor's signature under an advertising text for a method. An informational section such as "What is caries and how is it treated" is a different thing: it does not advertise the clinic's specific method, but it too must be checked by a doctor. Where the line between advertising and information runs in each text is decided by the clinic's lawyer; we do not draw it.
Using an assistant for medical texts stays possible, on the condition that a doctor checks every medical statement against the clinical protocol and not against the assistant's answer.
What it costs
For a clinic this is the smallest item of spending: the cost depends on how many people need an assistant, not on how many patients there are. According to the ChatGPT pricing page on 16 October 2026, the personal Plus plan costs $20 a month, and the Business plan is $25 per person billed monthly and $20 billed annually, from two people; on it, data is not used for model training by default. The National Bank of Ukraine rate on 16 October 2026 is UAH 44.8871 per dollar.
| Plan | 1 person a month | 3 people a month |
|---|---|---|
| ChatGPT Plus (personal) | $20 ≈ UAH 898 | $60 ≈ UAH 2,693 |
| ChatGPT Business, monthly billing | $25 ≈ UAH 1,122 (from two people) | $75 ≈ UAH 3,367 |
| ChatGPT Business, annual billing | $20 ≈ UAH 898 (from two people) | $60 ≈ UAH 2,693 |
Prices of other assistants and of automation are in the series hub. We do not give prices of booking systems and medical information systems: they were not checked against a primary source for this article. The payback of a booking system with reminders is calculated as in a salon: divide the subscription by the income from one appointment the reminders brought back. The calculation is in the article on AI for a beauty salon.
What not to do
- Do not paste entries from patient files into the assistant, even "to shorten them". Shortening a doctor's notes is work for the medical information system, not for a general chat.
- Do not switch on a voice bot that "advises on symptoms". Answers about drugs and doses are a medical consultation, and the bot bears no responsibility while the clinic does.
- Do not publish generated reviews, treatment stories or before-and-after photos. The first two are named directly in Article 21; the third goes beyond it but undermines trust in the clinic.
- Do not send patient data to a service without a written data-processing contract and without checking where the data is stored.
- Do not rely on "we switched training off". Switching training off does not lift the ban on disclosing information.
The general principle for every industry is in the article on data for an AI assistant.
First step for one week: a "Questions to the front desk" page
This task belongs to the first class of data: there is not a single patient in it, and the gain is visible within a week.
- Collect the twenty questions the front desk hears most often. From the call log or the administrators' notes, without names. Typical ones: "how much is a consultation", "is a referral needed", "can I come with a child".
- Give the assistant only the answers from the price list and the clinic's rules and ask it to format them as a questions-and-answers page.
- Give the text to a doctor. The doctor strikes out everything that sounds like treatment advice, a promised result or a recommendation, and leaves addresses, hours, prices and booking rules.
- Count the corrections. How many of the twenty answers the doctor changed or struck out. If more than five, the assistant is writing about medicine when it should have written about organisation, and the request needs narrowing.
- Measure a week before and after. How many times a day the front desk answers the same questions. If the questions have disappeared from conversations, the page works.
Our position
A clinic needs an assistant for administrative work and for drafts of texts, but not for work with patients. It never touches class 3, class 2 stays in the booking system, and class 1 is the field where AI saves hours with no legal risk. Medical statements on the website are always signed off by a doctor, not by the assistant.
The position does not apply if the clinic has a written data-processing contract with the vendor, a lawyer's opinion on the basis under Article 7 and a service built specifically for medical data. Then the decision rests on those documents, not on the rule of this article. For a small clinic with no lawyer and an ordinary chat assistant the rule "do not pass class 3" remains the only safe one.
Provisions checked on 16 October 2026 against the texts of the laws on zakon.rada.gov.ua: the Personal Data Protection Law (edition of 14.06.2025), the Fundamentals of Healthcare Legislation (Article 40), the Advertising Law (edition of 31.07.2026); ChatGPT prices — on the pricing page; exchange rate — National Bank of Ukraine on 16 October 2026. This article is not legal advice; the question of the basis for processing in a particular clinic is decided by a lawyer.