"Is the answer correct?" is the wrong question to ask when it comes to AI in healthcare.

The debate over generative AI in healthcare revolves almost exclusively around one question: Are the answers provided by ChatGPT, Gemini, or Claude medically accurate? In a new Viewpoint article in the European Journal of Public Health, Christoph Dockweiler argues that this narrow focus is misleading. It reduces a structural shift in the healthcare information landscape to a technical quality issue—and overlooks the bigger picture.

When millions of people turn to AI systems for information on symptoms, prevention, or treatment, the conditions under which populations form, question, and revise their health knowledge shift—and this happens without public health oversight. AI content carries no authorship, no conflicts of interest, and no peer review; it blends evidence, approximations, and plausible-sounding conjecture into a fluid, personalized narrative whose sources cannot be identified or verified.

This gives rise to a new form of inequality—an “AI Divide.” It does not concern access to AI (which is rapidly diminishing), but rather the unequally distributed ability to recognize, contextualize, and question AI-generated health narratives. Since nearly half the population in several EU countries has low health literacy, conversational AI is increasingly replacing—rather than supplementing—institutional information for precisely these groups. It is a gap not in access, but in epistemic agency.

Meanwhile, regulation is focusing on the wrong issue: The EU AI Act classifies clinical decision support as high-risk—yet the consumer-facing chatbots through which millions obtain their health information remain largely unregulated. In short: Europe is erecting guardrails around the clinical setting, while the information landscape outside it is being rewritten.

What does this imply? Three concrete starting points: recognize consumer-facing health AI as a public-health-relevant information infrastructure and subject it to transparency requirements; Expand health literacy concepts to include the skills required by generative systems; and encourage public health institutions to establish their own presence in conversational health communication—rather than leaving this field to systems optimized for engagement, not for informed decision-making.

“The crucial question is not whether AI provides correct answers, but who is still able to question its answers. That is precisely where health equity will be determined in the future.” Prof. Dr. Christoph Dockweiler