AI note taker for doctors: a practical guide for Swiss practices
By AllesHealth GmbH ·
The last patient left at six, and you still have eleven consultations to write up. You have heard colleagues talk about AI note takers that "write the report for you", and vendors promise exactly that. But what does such a tool actually do, may you use it in a Swiss practice, and how do you make sure the notes are right? This guide answers those questions in practical terms. If you want a deeper look at the research on AI scribes, read our evidence review on AI scribes.
What an AI note taker for doctors does
An AI note taker (also called an AI scribe or, if it listens in the background, an ambient scribe) turns spoken words into a draft clinical note. There are two common modes:
- Dictation: the doctor speaks a summary after or during the visit, and the software turns it into structured text.
- Ambient listening: the tool records the conversation between doctor and patient, transcribes it and uses a language model to draft a note with history, findings, assessment and plan.
In both cases the result is a draft. The doctor reads it, corrects it and only then saves it to the record. That review step is not optional; it is what makes the tool safe to use.
Why it matters: what the studies show so far
Documentation time is a real burden, and the first randomised trials of ambient AI scribes have now been published. They are encouraging, but the effects are modest and come from US settings.
- In a randomised trial with 238 outpatient physicians from 14 specialties, doctors were assigned to one of two ambient AI scribes or to usual care. Users of one scribe saw a 9.5% decrease in time spent in notes compared with the control group; for the other scribe the difference was not significant. Burnout-related scores suggested improvement for both scribe groups. Physicians reported that the AI notes occasionally contained clinically significant inaccuracies; when they described them, omissions were mentioned most often, followed by structural problems (such as unwanted formatting or too much detail) and pronoun errors [5].
- In a second, 24-week trial with 66 practitioners in ambulatory clinics in two US states, ambient AI reduced work exhaustion and interpersonal disengagement, but did not significantly increase professional fulfilment. Time spent on notes fell by about 0.36 hours per day (a secondary outcome); according to the authors, this came without compromising diagnosis, billing compliance or note quality [6].
The honest summary: an AI note taker can save some time and may ease the feeling of being overloaded, but it does not remove documentation work, results differ between products, and every note still needs checking. None of these trials was carried out in a Swiss practice or in Swiss German, French or Italian consultations.
The Swiss legal frame: FADP, FDPIC and recording consent
The revised FADP applies to AI today
The Federal Data Protection and Information Commissioner (FDPIC) states that the revised Federal Act on Data Protection, in force since 1 September 2023, is directly applicable to AI-supported data processing. The law is technology-neutral. The FDPIC expects manufacturers, providers and users of AI systems to make the purpose, functioning and data sources of the processing transparent [1]. High-risk processing is allowed in principle if appropriate protective measures are taken [1].
For a practice, this means in concrete terms:
- Health data is sensitive personal data under Art. 5 let. c FADP [2]. Consultation audio and notes are exactly that.
- Duty to inform (Art. 19 FADP): patients must be informed appropriately when their data is collected [2]. Mention the AI note taker in your privacy notice and explain it in plain words.
- Data protection impact assessment (Art. 22 FADP): if planned processing may result in a high risk, an assessment must be carried out in advance [2]. Clarify with your data protection adviser whether this applies to the tool you choose.
- Data security (Art. 8 FADP) and processors (Art. 9 FADP): you remain responsible for appropriate security, also when a provider processes the data for you [2].
Ask before you record
Ambient listening means recording a conversation. Under Art. 179ter of the Swiss Criminal Code, a participant who records a non-public conversation without the consent of the other participants can be prosecuted on complaint [3]. A consultation is a non-public conversation. So the rule for your practice is simple: no recording without the patient's clear agreement, and an easy way to say no. Dictation after the visit, in which only the doctor speaks, avoids recording the patient at all.
The doctor stays responsible
The FMH calls for promoting the use of artificial intelligence while raising awareness of its risks [4]. In practice, that means the doctor decides what goes into the record and signs it, whatever tool produced the draft.
This article is general information, not legal advice. For your own setup, consult a data protection or legal adviser.
How to introduce an AI note taker in your practice: step by step
- Start with suitable visits. Routine follow-ups, chronic-disease check-ups and structured consultations are good candidates. Be cautious with sensitive conversations (mental health, domestic violence, sexual health), with visits involving interpreters, and with patients who seem unsure. Here, dictation afterwards is often the better choice.
- Inform patients before the visit. A short notice at reception, on your website and in your privacy notice explains what the tool does, what is recorded and how long audio is kept.
- Ask at the start of each consultation. "I use a tool that helps me write my notes. It records our conversation. Is that all right with you?" Respect a "no" without discussion, and note the decision.
- Check where data is processed and stored. Ask the vendor where audio is transcribed, where the language model runs, which subcontractors are involved and in which countries, and whether patient data is used to train models.
- Keep audio short-lived. Prefer tools that delete the recording once the note is approved.
- Review every draft in full. Pay special attention to medication names and doses, laterality (left/right), negative findings, allergies and follow-up plans. Look for what is missing, not only for what is wrong; in the larger trial, omissions were among the most common inaccuracies [5].
- Do not shorten the review to save time. The time gained comes from not typing, not from reading less carefully.
- Agree the team workflow. Decide who handles referral letters, prescriptions and follow-up appointments that come out of the note, so the MPA can act on a finished record rather than chase the doctor.
- Collect errors and review them. Keep a simple log of wrong or missing content and discuss it with the vendor every few weeks.
- Evaluate after a trial period. After four to eight weeks, ask doctors, MPAs and a few patients how it went, and compare documentation backlogs before and after.
What to look for in AI documentation software (2026)
When you compare AI note takers or practice software with AI documentation features, check these points:
- Draft-and-approve design: nothing enters the record without the clinician's approval, and AI-generated text is clearly marked.
- Integration with the record: the note lands in the right patient file and fields, not in a separate app you copy from.
- Languages: reliable results in the languages your patients speak, including dialect and mixed-language consultations.
- Hosting and subcontractors: clear, written answers about data location and processors.
- Consent support: easy start, pause and stop, and a way to document the patient's choice.
- Transparency: the vendor can explain purpose, functioning and data sources, as the FDPIC expects [1].
- Error reporting: a simple way to flag a wrong note.
How AllesHealth Connect helps
AllesHealth Connect is practice software for GP and multi-doctor outpatient practices, designed in Switzerland. It combines the electronic health record (EHR) with appointments, e-prescriptions, video consultations and billing. On the Premium plan, it adds AI support for documentation:
- AI-assisted EHR updates, including voice dictation: the doctor dictates, and the AI drafts the update to the patient's record.
- AI drafts for documentation: AI agents prepare documentation drafts, alongside AI support for booking and reminders.
- The clinician reviews every result: AI drafts and suggests; the doctor checks, corrects and stays responsible for the record.
- Swiss hosting: data of Swiss practices is hosted in Switzerland (Zurich).
For doctors: less typing after hours, notes that go directly into the right patient record, and structured data (the system is built on HL7 FHIR and supports SNOMED CT coding) instead of loose text files.
For nurses and MPAs: a finished record sooner, so prescriptions, follow-up appointments and patient messages can be handled the same day, in the same system, without waiting for a dictation to be typed up.
For patients: more attention during the visit, and follow-up information that reaches them through the Patient Companion app, in your practice's branding.
AI documentation works best when the rest of the practice runs in one place. Read more in our article on too many separate systems in the practice, and see our guide on how to stop charting after hours and on the revised FADP for medical practices.
Try AllesHealth Connect free for 60 days – no setup needed, no credit card: start free or book a demo. See the plans on our pricing page.
#AINoteTaker #AIScribe #ClinicalDocumentation #DigitalHealth #PracticeManagement #GPPractice #DataProtection #Switzerland
Sources
- Update – Current data protection legislation is directly applicable to AI, Federal Data Protection and Information Commissioner (FDPIC) (2025)
- Federal Act on Data Protection (FADP), SR 235.1, Fedlex – Swiss federal law (2023)
- Swiss Criminal Code, SR 311.0 (Art. 179ter), Fedlex – Swiss federal law (2026)
- Positionspapier: Wandel des ärztlichen Berufsbildes durch digitale Technologien, FMH Swiss Medical Association (2024)
- Ambient AI Scribes in Clinical Practice: A Randomized Trial, NEJM AI (PubMed record) (2025)
- A Pragmatic Randomized Controlled Trial of Ambient Artificial Intelligence to Improve Health Practitioner Well-Being, NEJM AI (PMC full text) (2025)