Public Health in the Age of Digitalization: From Observer to Petitioner

For two decades, digital public health rested on an unspoken assumption: that what people search for, share, and discuss online regarding health leaves a trail, and that this trail is visible to others as well as to the systems that record it. A new article by Prof. Dr. Christoph Dockweiler in the journal Policy & Internet argues that this assumption is currently breaking down.

The reason is a shift in location. Health-related questions are moving from publicly accessible websites and platforms to conversations with AI assistants. There, the exchange takes place in a private one-on-one interaction between a person and a system and is visible only to the provider. ChatGPT alone had around 700 million users per week by mid-2025, and according to the company, Microsoft’s AI products process over 50 million health-related questions daily. In the U.S., about one-third of adults used AI chatbots for health information last year—a proportion on par with that of social media.

This article categorizes this development into three regimes of observability. Under “open observability,” traces on the web were public, and public health served as an observer. Under “mediated observability,” access was routed through interfaces that platforms owned, priced, and could close at will, thereby turning public health into a licensee. The end of this phase can be dated, for example, to the shutdown of CrowdTangle in August 2024 or to X’s decision to discontinue its academic access program and set the price for its enterprise API at $42,000 per month. In the third regime—closed observability—public health is left with nothing more than the role of a supplicant.

What reaches the public from the outside are reports from the providers themselves. Microsoft analyzed over half a million de-identified health-related conversations with Copilot from January 2026. Just under one in five concerned the user’s own symptoms or medical conditions, and personal health inquiries peak in the evening and at night—precisely when regular care is hardest to access. Such findings are clinically significant but cannot be verified. The conversation data cannot be shared, the analysis code is proprietary, and all authors are employed by the company under study. OpenAI, in turn, reported in October 2025 that approximately 0.15 percent of weekly active users engage in conversations with explicit references to possible suicide planning or intent—an estimated one million people per week. Healthcare systems can neither break down this figure by region nor track it over time because the necessary access does not exist.

This article therefore argues that research access to conversational health data—while ensuring data protection—should be treated as public health infrastructure. Technically, this is not a pipe dream, as providers have long been analyzing their data in an aggregated and privacy-preserving manner. What is missing is independence and accountability. Health law has long recognized models for this, such as mandatory reporting of infectious diseases or pharmacovigilance. In those cases as well, certain privately held observations are so significant that their aggregated reporting becomes a condition of business operations. The Digital Services Act provides a framework with the research access provisions under Article 40, but has not yet captured the dialogic core of these systems.

The first generation of digital public health has learned to interpret the open web. The next generations will have to negotiate and regulate their right to access.

(Author: Christoph Dockweiler)