Skip to content

AI in medical practices: what assistants can do without patient data

By the KI können editorial team8 min read

Written with AI and checked against the sources by a second AI. As of November 16, 2026. More in the notes on AI.

In a doctor’s practice, the work with patients comes with a lot of writing: notices, replies to e-mails, templates for reminders. An AI can prepare some of it as long as nobody can be identified. This guide shows what works well, what doesn’t belong in a public AI chat (a freely available service with a personal account and no contract with the practice) and who decides in the practice.

At the reception desk of a medical practice, a medical assistant works on a laptop, a patient stands in front, a shelf of binders behind.

Key points

  • Under the GDPR, health data is specially protected. On top of that comes medical confidentiality under Section 203 of the German Criminal Code, which also covers the practice team.
  • A lot works without patient data: notices, general information sheets, reply templates, phone guides, checklists.
  • Test results, doctor’s letters, diagnoses, medication and appointment lists don’t belong in a public AI chat.
  • Whether AI is used in the practice, for example on the phone, is decided by the practice owner, with contracts and a commitment to confidentiality.
  • Every medical statement in a text is approved by the doctor.

In short: a lot works without patient data

Many texts in a practice are addressed to all patients or to the team. For these, an AI like ChatGPT, Claude, Gemini or Copilot needs no data about individual people. As soon as a particular person is involved, the text stays in the practice.

What may go into a public AI chat and what may not
Works without patient dataDoesn’t belong in it
A notice about the practice closing for vacation, with the covering practiceTomorrow’s appointment list
A reply template for prescription requestsA patient’s e-mail describing their symptoms
A general information sheet on blood testsA test result or lab value to explain
A phone guide for frequent questionsThe callback to a particular patient
A checklist for new colleaguesA doctor’s letter to summarize

The rule of thumb: if you could put the text up in the waiting room without anyone being identifiable in it, it fits in an AI chat. Otherwise it doesn’t.

In the trial lesson you put your first instruction together and get feedback. No account, in a few minutes. Start the trial lesson

Why patient data is so strictly protected

The EU General Data Protection Regulation (GDPR) counts health data among the special categories of personal data. It may only be processed if one of the exceptions listed in Art. 9 applies, for example for treatment by professionals who are bound by confidentiality. Health data also includes the fact that someone is being treated in a practice at all, and when.

Then there is professional secrecy. Under Section 203 of the German Criminal Code (Strafgesetzbuch), a doctor or dentist who discloses another person’s secret without authorization commits an offense if it was entrusted to them, or became known to them, in that role. Medical and dental assistants (in German Medizinische and Zahnmedizinische Fachangestellte, MFA and ZFA) may learn these secrets for their work. Anyone who works in the practice and passes them on without authorization also commits an offense under the same section.

Typing patient data into a public AI chat means passing it on to an outside company that the practice has no contract with and that nobody has bound to confidentiality. That applies even to a quick question with a name and date of birth.

Notices and general information sheets

Amira is a medical assistant in a family doctor’s practice. Before every vacation, she rewrites the notice about the covering practice. She gives the AI the facts:

Sample instruction
Write a notice for our family doctor’s practice, formal tone, 80 words at most, easy to read from two meters away.
Facts: practice closed from [date] to [date]. Covering practice: [name of covering practice], [address], [phone number], office hours [times].
In an emergency: 112. Outside office hours: out-of-hours medical service 116117.
Reopening on [date].
No apology, no holiday greetings, just the information.

In Germany, 112 is the emergency number and 116117 the number of the out-of-hours medical service. For general information sheets, such as “What to keep in mind before a blood test”, the rule is: the AI phrases, the medical content comes from the doctor. So Amira gives the AI the instructions the doctor has set and has it only put them into plain language. She submits the finished sheet for approval.

Reply templates and phone guide

On the phone and in the inbox, the same questions come up every day: reordering a prescription, picking up a referral, moving an appointment. Fixed replies save time and make sure everyone on the team says the same thing. The templates contain only placeholders:

Sample instruction
Write reply templates for e-mails to our family doctor’s practice, formal tone, 60 words at most each, with placeholders in square brackets.
1. Prescription request: from [day] your prescription [can be redeemed at the pharmacy as an e-prescription with your health insurance card / can be picked up at the practice].
2. Referral: can be picked up from [day] during office hours; please bring your health insurance card.
3. Rescheduling: new appointment [date, time], please confirm.
Each template ends with the sentence: “If you have acute symptoms, please call the practice; in an emergency, call 112.”
No medical advice.

Whether and how prescriptions and test results may be sent by e-mail is settled by the practice. The AI only drafts the text. You write the reply with real details in your practice software or inbox, not in the chat.

Reminders in a dental practice

Nora is a dental assistant and looks after preventive care. For the reminder about the next professional teeth cleaning, she has the AI write a friendly template, without pressure and without advertising. It is sent only to patients who have agreed to reminders, through the practice software.

Onboarding and team routines

New colleagues at the front desk have to learn a lot in a short time: how check-in works, where things are, who is responsible for what. An AI can turn your bullet points into a clear checklist:

Sample instruction
Turn my bullet points into a checklist for the first week of a new medical assistant at the front desk of our family doctor’s practice. Organize it by day, eight points per day at most.
Bullet points: get to know the practice software, check-in procedure, read the phone guide, show emergency numbers and the emergency kit, read and sign the hygiene plan, prescription ordering procedure, discuss data protection and confidentiality and sign the commitment, who on the team is responsible for what.
Don’t add any medical or hygiene content. Where a practice document is needed, write “as per practice document”.

Whatever is in the practice’s hygiene plan and work instructions, copy word for word. The AI sorts; it doesn’t rewrite any rules.

Notices in several languages

For notices about office hours or check-in, an AI can provide versions in other languages, such as English, Turkish or Arabic. Have them read by someone who speaks the language well. Only translate medical instructions if the doctor approves the translation.

What doesn’t belong in a public AI chat

  • Test results, lab values and doctor’s letters: not even “just to explain” or to summarize.
  • Diagnoses and medication: assessments are made by the doctor; doses are in the medication plan.
  • Appointment lists, billing data, health insurance cards: even the fact that someone is a patient is protected.
  • E-mails and messages from patients: don’t copy them into the chat to get a reply.
  • Photos: for example of skin, wounds, X-rays or documents, even without a face.

A code instead of the name doesn’t make a case anonymous. “Patient, 47, on sick leave since May, works for the local fire department” gives a person away in a small town even without a name. Leave such cases out altogether.

AI on the phone or in the practice software

Samir is a family doctor with his own practice. A provider has shown him an AI phone assistant that answers calls and takes prescription requests. The decision is Samir’s as the practice owner, not his team’s. He has several things to settle:

  1. Is there a data processing agreement (DPA; in German AVV), and where is the data processed?
  2. Is the provider bound to confidentiality? For data under professional secrecy, a DPA alone is usually not enough.
  3. Does the service meet the extra rules for cloud services in practices that bill health insurers?
  4. Does the assistant say right at the start that it is an AI? The EU AI Act has required this since August 2026.
  5. How do callers reach a human, and what happens in an emergency?
  6. How is the team briefed, and who checks what the assistant notes down?

If a practice uses AI, the EU AI Act also requires it to promote its team’s AI literacy. A short written rule and a briefing are a good start. More on this in the guide to the AI Act.

Check every text before it goes up

  1. Are dates, times, phone numbers and the covering practice right?
  2. Has the doctor approved every medical statement?
  3. Has the AI added advice that wasn’t given?
  4. Does the text promise anything the practice can’t deliver, such as a callback the same day?
  5. Is really nobody identifiable, neither a patient nor a colleague?

Delete anything that hasn’t been approved. Responsibility for texts to patients stays with the practice.

How to practice this

Writing instructions so that nobody is identifiable and the AI still delivers what you need takes practice. In the app “KI können” you practice this in short lessons, including examples from health care, and put the instructions to work in your own AI. The trial lesson needs no account.

Sources

Information as of November 16, 2026. Laws and offers can change; this article is not legal advice.