In Switzerland, 42% of the people surveyed for the 2026 health study by the CSS health insurance group put their symptoms to an artificial intelligence (AI) chatbot over the past year. A year earlier, the figure was 20%. Conducted by the research institute Sotomo among 2,459 people, the survey was reported on 3 November 2026 by the specialist website Inside IT. It describes AI as one more source of health information, not as a substitute for the doctor.
How was the CSS health study conducted?
The CSS health study has been published every year since 2020. According to the CSS presentation page, this eighth edition is based on a survey by the Sotomo institute among 2,459 people in French-speaking Switzerland, German-speaking Switzerland and Ticino.
The question concerns self-reported use. Respondents state whether they turned to a chatbot to interpret their symptoms during the previous year. Our reading: a single use is enough to be counted among the 42%, and the figure says nothing about frequency or about how serious the symptoms were.
The CSS page does not specify the survey period, its collection method or its margin of error. Nor does it name the chatbots used. The full report is available as a PDF on the same page.
Who asks a chatbot about their health?
The increase affects all age groups, according to Inside IT. Young adults are the most numerous (62%). Among the over-65s, the share tripled in one year, from 10% to 30%.
| Group | 2026 study | Previous study |
|---|---|---|
| All respondents | 42% | 20% |
| Young adults | 62% | – |
| Over 65 | 30% | 10% |
Symptoms are not the only reason. According to Inside IT, 32% of respondents have used a chatbot at least once for lifestyle questions, such as diet, physical activity or mindfulness. 14% have sought support there for emotional or psychological difficulties.
This psychological support mainly concerns the youngest. It applies to 22% of those under 36 and 28% of people whose mental state is currently fragile.
Does using a chatbot mean trusting it?
No, and the study separates the two. More than a quarter of respondents (28%) place chatbots among their main sources of information when they have physical symptoms. But, according to Inside IT, they replace neither doctors nor family and friends. Respondents see them rather as a complementary source of information.
Feeling understood is not a measure of the quality of the answers. Among people who have used a chatbot privately, 60% felt well understood most of the time, and a quarter rather not. This result describes a perception. The study does not check whether the information received was accurate.
The figure on loneliness also calls for precision. The CSS page states that 27% of people sometimes felt less isolated thanks to a chatbot. According to Inside IT, this rate applies to people who had felt lonely more often recently and used AI privately. Across all respondents, the share falls to 15%.
The same study also observes a broader withdrawal. Only 50% of people turn to those close to them when they are struggling psychologically, the lowest share since 2022 (57%). Among young adults, 70% say they withdraw into themselves at such times.
What are a chatbot's limits when faced with symptoms?
A chatbot can produce a wrong answer that seems plausible. The CSS study does not test the tools' answers, but other measurements exist. A test published on 30 October 2026 by Intochat, a publisher of chatbots for businesses, illustrates this with language models tested in Swiss German.
According to this test, the five models from Anthropic, Google and OpenAI all produced answers containing invented information. The share ranged from 4% for the best to 20% wrong answers for the worst. One answer even made a diagnosis about a smell coming from the ventilation.
This test covered customer messages, not medical questions. It therefore cannot be used to quantify errors in health matters. It only shows that a plausible answer is not a verified answer.
Health data: what does Swiss law say?
Symptoms shared with a chatbot are health data. The Federal Act on Data Protection (FADP), in force since 1 September 2023, classifies them as sensitive personal data, in its Article 5. Several rules follow for whoever processes them:
- consent must be explicit for processing sensitive data (Art. 6 para. 7);
- disclosing sensitive data to third parties in principle constitutes a breach of privacy, unless there are grounds for justification (Art. 30 and 31);
- large-scale processing of sensitive data requires a prior impact assessment (Art. 22);
- if the data goes abroad, the person must be informed of the destination country (Art. 19 para. 4).
The law applies to matters that have effects in Switzerland, even if they occur abroad (Art. 3). It does, however, exclude processing by a natural person for exclusively personal use (Art. 2). Our reading: the user who asks a question privately is not covered, but the company operating the chatbot is.
The Federal Data Protection and Information Commissioner (FDPIC) pointed out on 24 September 2025 that this law applies directly to AI. Our page on AI and data protection details this position. No general AI law is in force. According to the SME portal of the State Secretariat for Economic Affairs (SECO), a preliminary draft is expected by early 2027, as summarised on our page on AI regulation in Switzerland.
What this changes for a patient in Switzerland
The sources offer a few pointers, which are not medical advice:
- treat a chatbot's answer as complementary information to be checked, the use the CSS study observes among respondents;
- share only strictly necessary data, in line with the FDPIC's general recommendation; the Federal Office for Cybersecurity (FOCS) warned on 16 October 2026 against entering personal information into an AI without precautions, as our page ChatGPT in business points out;
- ask the provider what data it keeps: Article 25 of the FADP grants a right of access, including to the retention period and, where applicable, to the existence of an automated decision under the FADP.
What we do not yet know
- Which chatbots respondents use: the CSS presentation page names none.
- Whether these exchanges led to a medical consultation or replaced one: neither the CSS presentation page nor Inside IT says.
- The content of the preliminary draft on AI expected by early 2027.
Sources
- CSS, « Etude sur la santé CSS 2026 », presentation page, and its German version « Die CSS Gesundheitsstudie 2026 »
- Inside IT, « Immer mehr Schweizer nutzen KI bei Gesundheitsfragen », 3 November 2026, with Keystone-ATS
- Fedlex, Federal Act on Data Protection (SR 235.1), status as of 7 July 2025
- 2clics.com, articles cited in the text, based on Intochat, the FDPIC, SECO and the FOCS
