When Hilary Hodge needs help making sense of her health, the doctor’s office isn’t always her first port of call. Often, she turns to ChatGPT.

Hodge, who lives in France’s Loire Valley and has severe asthma and adrenal insufficiency, calls the chatbot a “thought partner” that helps her prepare for appointments and navigate a condition that many clinicians rarely encounter.

“Many doctors know little about adrenal insufficiency,” said Hodge, patient co-chair of the severe asthma arm of the European Respiratory Society. “I have found ChatGPT to be fantastically helpful when it comes to brainstorming how to address [my health issues] and how to talk about them with my endocrinologist.”

Hodge is clear that she does not rely on AI blindly — she cross-checks its responses against medical literature and treats it as one source among many. But not every patient uses AI as a prompt for clinical care. As chatbots become a routine first stop for health questions, clinicians and regulators are asking when AI helps patients navigate care — and when it may delay, redirect, or replace clinical advice.

The Scale of the Shift

The 2026 Edelman Trust Barometer, drawing on data from more than 16,000 respondents across 16 countries including France, Germany, and the UK, found that 35% of people now consult an AI platform for health questions. Among those users, 78% look for treatment suggestions and 74% use AI to get a second opinion on a diagnosis.

When healthcare feels slow or difficult to access, some patients may look elsewhere for answers. That behavior doesn’t necessarily mean patients lack trust in their doctors, said Charlotte Blease, PhD, associate professor of digital health at Uppsala Universitet in Uppsala, Sweden.

“People are hungry for information, reassurance, and explanations, and generative AI has made those things available instantly and conversationally in a way we have never seen before,” she said. “It is now possible to ask a question about a symptom or test result at 2 AM and receive an immediate, highly personalized response in plain language.”

The use of AI could help drive increased demand for care in European health systems ill equipped to meet those needs. In 2022, the OECD estimated a shortfall of roughly 1.2 million doctors, nurses, and midwives across the EU, a gap expected to widen as the workforce ages.

However, expanded access to health information is not without risk, particularly in mental health, where AI can feel not only informative but also emotionally validating.

“Patients often feel more understood by AI than therapists,” said Julian Schwarz, MD, consultant in psychiatry and psychotherapy at the University Hospital of the Brandenburg Medical School in Germany.

Long waiting lists and complex diagnostic pathways may drive some patients to seek answers independently, and AI can give them a vocabulary and a degree of confidence to act on those answers.

From Information to Medicines

AI did not create self-diagnosis or self-medication. A 2024 analysis of European Health Interview Survey data from more than 255,000 people estimated that 34% of people in the EU already self-medicate. On antibiotics alone, the World Health Organization (WHO)/Europe found that across 14 countries in the WHO European Region, 1 in 3 respondents had either used leftover antibiotics or obtained them without a prescription.

The concern is that AI could make that behavior easier to act on by giving patients a more personalized and authoritative-seeming explanation. Schwarz said colleagues have reported cases of patients using generative AI to self-diagnose conditions such as attention-deficit/hyperactivity disorder, then obtaining drugs like Ritalin through illicit sources.

“Ritalin, for example, can affect blood pressure,” Schwarz said. “While AI may tell patients about potential side effects, they may not be able to manage them alone, potentially putting them in harm’s way.” Accessing some prescription drugs outside traditional brick-and-mortar pharmacies has become more common. A 2024 cross-sectional study of online pharmacies conducted from the UK found that many sites offered prescription-only medicines while only a minority required a prescription. Controlled medicines, including benzodiazepines, were also available, mostly through sites assessed as rogue.

For clinicians, the concern is that AI may give patients a name for their symptoms — and enough confidence to seek treatment — before anyone has checked the diagnosis, contraindications, interactions, or monitoring needs.

Not Always Riskier

But patients’ use of AI may have positive aspects as well. The picture is not uniformly bleak. Blease cautions against assuming that patients who turn to AI are behaving irrationally or that clinicians are always in the best place to determine what is an appropriate use.

“People have always sought health information wherever they could find it — from family members, magazines, libraries, television, internet forums, and now AI,” she said. While generative AI can hallucinate, she added, it may still be preferable to social media, where misinformation from influencers is widespread.

Hodge’s experience reflects what that safer model can look like: AI as a prompt for care, not a replacement for it. The question is whether the healthcare system can encourage that model while managing the risks of those who use AI to bypass clinical oversight altogether.

What Happens Next

If patients are already using chatbots for health questions, how can clinicians and regulators make that use safer? Stephen Gilbert, PhD, professor of medical device regulatory science at TUD Dresden University of Technology in Dresden, Germany, points to data showing that every week, 1.2 million people discuss suicide with ChatGPT, and 560,000 show signs of psychosis or mania.

Gilbert cautions, however, against overly rigid regulation. The EU AI Act imposes strict requirements on AI tools classified as medical devices, but general-purpose chatbots like ChatGPT sit outside that framework. When purpose-built health AI is difficult to get approved, he argues, patients simply default to unregulated alternatives.

“I don’t think the European approach — which is ultrastrict regulation — is the solution. It just results in a situation where it’s very, very difficult for any technology in the space to get approval. People just find a way around it, which is to use the general ChatGPT, Gemini, or Anthropic models.”

Gilbert favors a tiered approach — leaving general wellness advice unregulated at market access but subject to oversight responsibilities for the advice given by tools, while applying stricter approval and oversight to AI that guides patients on specific medications, mental health, or the management of chronic conditions.

Schwarz recommends that clinicians not dismiss the technology but rather engage with it directly — asking patients about their AI use rather than assuming it is not happening. The task for medicine is to ensure that AI helps patients and clinicians make better decisions together, rather than leaving patients to navigate risk alone.

Hodge, Blease, and Schwarz reported having no relevant financial relationships.

Gilbert reported being an advisory group member of the Ernst & Young-coordinated “Study on Regulatory Governance and Innovation in the field of Medical Devices” conducted on behalf of the Directorate-General for Health and Food Safety of the European Commission. He reported having or having had consulting relationships with Una Health GmbH; Lindus Health Ltd.; Flo Ltd; thymia; FORUM Institut für Management GmbH; High-Tech Gründerfonds Management GmbH; and Ada Health GmbH, the Saudi Arabia Food and Drug Authority funded through the United Nations Development Programme, Deutsche Gesellschaft für Internationale Zusammenarbeit GmbH. He also reported holding share options in Ada Health GmbH.

Annie Lennon is a medical journalist. Her writing appears on Medscape, WebMD, and Medical News Today, among other outlets.

This article is part of an ongoing, year-long Medscape series examining the decline of public trust in medicine. More content is available here.