AI for medical practices: reception, appointments and reminders

The AI for Medical Studies lightens the secretariat on three requests that come back every day in an outpatient clinic: book the right visit, with the right specialist and preparation instructions; remember the appointment and collect the cancellations in time to give the place back to someone else; answer those who ask if the reports are ready. An AI assistant connected to the agenda and management software can deal with it in chat or on the phone, with rules written by the studio.

Diagnosis, therapies, explanation of results and certificates remain with the doctor. The Italian Artificial Intelligence Act also calls for systems used in healthcare to be "reliable, periodically verified and updated in order to minimize the risk of errors" (law 132/2025, art. 7): an assistant for patients does not set up only once, it is controlled over time.

Book a specialist visit: the card that the assistant must know

"I would like to book a visit" is not enough to choose a time. In a clinic, at least four pieces of information are needed: which specialist, if it is a first visit or a check, if the service is private or with the health service representative, in which clinic there is the equipment that is needed. An experienced secretary asks her by heart; an AI assistant asks her only if someone has written them.

True work, before any automation, is therefore a board for each performance. The assistant reads it, asks the questions that are missing and proposes only the times compatible with doctor, room and duration. Three examples of how to compile:

First Cardiological Visit

Who: the cardiologist, in his days of surgery. Where: the room with the electrocardiograph. Duration: the one set by the doctor for a first visit, longer than a check-up. To be carried: previous examinations and list of drugs in use. Regime: private with the price list, or in agreement with the binding.

Abdomen ultrasound

Who: the ultrasound doctor. Preparation: the text written by the doctor, for example the indications on fasting, sent word for word and never summarized by the AI. When to send it: to the reservation and again with the reminder. To be carried: any previous ultrasounds.

Control visit

Who can book it: only those who have already been visited by the same specialist. To be taken: the exams required at the previous visit. What the assistant asks: the date of the last visit, to propose the right doctor. What it does not ask: how is the patient, symptoms or outcomes.

The channel counts less than the card. On the phone the pattern is that of telephone reservations with AI, in chat that of the appointments set automatically. In both cases the card is the same and updates your voicemail, because you know when a doctor changes days or a room stays closed.

Reminder and cancellation: give back the place to those on the waiting list

The reminder isn't just a reminder. His most useful task is to get the cancellations to arrive when there is still time to reassign the place: a slot released at the last moment, to be reassigned one phone call at a time, risks to remain empty.

For Health Service services there is also a written rule. The user who does not show up, or does not announce the impossibility of enjoying the service booked, "is required, where not exempt, to pay the participation fee at the cost of the service" (D.Lgs. 124/1998, art. 3, paragraph 15). If your clinic also works in convention, the reminder can remember it in one line, without threatening tones. For private visits the cancellation policy written in the study regulations applies: the assistant applies it, does not invent it and does not make exceptions on his own initiative.

Think of a polyclinic with a cardiologist on Tuesday and an ultrasound on Thursday. The reminder starts with the advance you have chosen and asks for confirmation. If the patient replies that he cannot come, the assistant frees the slot and writes to the patients on the waiting list for that performance, in order of request; the first one who accepts takes the place and immediately receives the preparation. The secretariat finds the agenda already updated and intervenes only if no one answers.

Also be careful what you write in the text. The GDPR wants data that is "adequate, relevant and limited to what is necessary" (Article 5), and the preview of a message is also read on a blocked phone, perhaps by a family member. "Oncological visit tomorrow at 9 o'clock" says too much; "date tomorrow at 9 o'clock in via Roma 12" enough. Specialist and preparation the patient finds them in the booking confirmation or in the reserved area.

Messages almost always travel on WhatsApp or SMS. How to set automatic sending by study number is explained in Automation WhatsApp Business.

"Are my exams ready?": reports, channels and consent

It is one of the most frequently asked questions in the clinic, and the most delicate to automate, because the complete answer is a health document. Online Report Guidelines: the online delivery is an optional service, the patient must be able to continue to withdraw the report on paper and adheres only on the basis of "a specific information and a specific consent", being able to say no even for a single examination. In the clarifications of 7 March 2019 the Guarantor reiterated that for online reporting consent remains required by industry regulations, even if for the treatments necessary for care it is no longer needed.

In practice, the assistant does three things: checks in the management software whether the report is ready, tells the patient how to receive it through their chosen channel (collection at the counter, private area, email if consent exists), and records the request. It does not attach the report in chat if that channel was not chosen and never comments on a value.

An example: a patient writes in the evening that she has had her tests and asks whether the results are ready. The assistant sees in the management software that the report is signed, reminds her that she chose the secure area and sends her the access link, without attachments. If she then asks what a value means, the question goes into the doctor’s queue with her name and the reference to the test.

Clinical questions in chat: a queue for the doctor, not automatic answers

Each channel open to patients sooner or later receives clinical questions: a new symptom, a doubt about a drug, a result to be interpreted. The assistant does not answer on the merits, puts them in order. If the message contains emergency signals it immediately indicates to contact the emergency services; in other cases it prepares for the doctor a summary with name, question and related documents, and tells the patient by when he will receive an answer according to the rules of the study.

It is the same boundary drawn by the Medical Deontology Code in Article 78: the doctor who uses IT tools "guarantees the acquisition of consent, the protection of confidentiality, the relevance of the data collected", and when treating clinical data with these technologies makes his own decisions "ensuring the conscious participation of the assisted person". A linguistic model left free to answer a clinical question may err in a safe tone; the mechanism is explained in AI Hallucinations.

After the visit: dictations and drafts that the doctor rereads and signs

Part of the secretarial work comes after the visit: the letter for the attending physician, the report, the summary to be delivered to the patient. Here the AI acts as a typist: it transcribes the doctor's dictation, arranges the entries according to the model of the study and prepares the draft. The signature remains of the doctor, who rereads each line before the document comes out.

We need two cautions. If the visit is recorded to be transcribed, the patient must be informed first, as with any new processing of his/her data, and Law 132/2025 still grants him/her the right to be informed about the use of artificial intelligence technologies (Art. 7, paragraph 3). And a tool that goes from transcription to suggesting diagnosis or therapy is no longer a typist: the code of ethics asks the doctor who uses these technologies "for the purposes of prevention, diagnosis, treatment or clinical surveillance" to adhere to the "criteria of proportionality, appropriateness, effectiveness and safety" (Article 78). Before adopting it, have the manufacturer tell you in writing what it does and with what certifications.

Before turning on the assistant: log, contract and impact assessment

An outpatient clinic treats health data every day, and an AI assistant adds a new treatment. The GDPR exempts organisations with fewer than 250 employees from the treatment register, but not when the treatment includes special categories of data such as health data (Article 30(5)): the medical office keeps the register anyway, and the assistant enters it with a voice of his own. The provider hosting the conversations becomes responsible for the processing and must be bound by a contract (Article 28).

Then there is the impact assessment. The the Guarantor list of the treatments that require it name ‘artificial intelligence systems’ and ‘the use of on-line voice assistants’ when adding to at least one other of the criteria of the European guidelines, and in another entry the non-occasional treatment of patient data. For a polyclinic that puts an assistant in front of patients is a concrete reason to do it, with the privacy consultant; the method is in How to Assess the Risks of an AI System.

• A new entry in the processing register: purposes, data collected, retention times of conversations.
• The contract with the assistant provider and the indication of where the data are stored; the points to be checked are in GDPR and artificial intelligence in the company.
• The impact assessment made, or reasoned in writing if you conclude that it is not necessary.
• The updated information: those who write to the clinic know that an AI system responds, as required by Article 50 of the AI Act from August 2, 2026.
• A person in the study who sampled conversations and updated performance cards.
• A list of what the assistant never does, approved by the health director or the holder.

The framework for all private health is on the page AI for Private Health; if your is a dental practice, hygiene calls and laboratory work have its own rules, collected in AI for Dentists. If you want an assistant that you really book in the schedule of your clinic starting from the cards of your performance, Giallo Studio He designs it with you and tests it on only one specialty before extending it to the others.