Generative AI Risks Exacerbating Health Inequality and Cyberchondria in

Generative artificial intelligence (GenAI) is changing how people around the world find, understand, and use health information. This technology lets people ask questions and receive quick, personalized answers in a conversational style. It offers more than traditional websites and search engines, which mainly help people find existing information. Conversational AI combines information and creates answers that sound natural and are tailored to each person. These answers can also sound direct and trustworthy to people looking for explanations of symptoms or advice about what to do next. Although this technology could greatly improve health education and make information easier to access, especially in places where formal healthcare is hard to reach, it also creates serious public health concerns. These include false information, uncertainty, misplaced trust, and increased worry about health. This article explores how using GenAI to find health information may connect to cyberchondria, unequal access to digital health resources, and the weaknesses of healthcare systems. It uses Somalia as an important example of these challenges.

Cyberchondria means searching for health information online too often or repeatedly, in a way that can increase worry about health and make people feel more uncertain. GenAI creates a different kind of environment for finding information. These systems give quick, interactive answers that can seem final and certain. This may change the reasons people experience cyberchondria, such as having too much information to sort through or not knowing which information to trust. Instead of simply showing a list of sources, AI provides fast answers tailored to the user. The main idea explored here is that conversational AI may do more than extend existing habits of searching for symptoms; it may change those habits themselves. This way of interacting could make people more likely to keep exploring their symptoms and make AI-generated information more convincing. Careful research is needed to find out whether this actually happens.

An important limitation is that large language models (LLMs) are not always correct. They can produce information that sounds believable but has no reliable support, leave out important advice, give different answers to the same question, and sound very confident even when their answers are not reliable. There is still no clear evidence proving that using GenAI causes cyberchondria. However, one possible explanation is that people may continue to feel uncertain or become even more worried when they use systems that are fast and conversational but can also give incomplete, inconsistent, or contradictory answers. For people who are already likely to worry about their health, this could lead to a familiar cycle: noticing a symptom, checking it repeatedly online, feeling reassured for a short time, becoming doubtful again, and eventually feeling more distressed.

The use of these technologies in fragile healthcare systems with limited resources, such as Somalia's, may make these problems worse. Digital health services have grown quickly across low- and middle-income countries (LMICs), but their expansion is often held back by disconnected systems, poor coordination between technologies, and a shortage of trained workers, rather than a lack of interest in technology. Where access to doctors is uneven and referral systems are weak, AI-generated information may have a particularly strong influence because people often cannot get timely advice from a healthcare professional. The effects of AI also depend heavily on differences in digital skills and health knowledge. Limited internet access, lack of suitable devices, and language barriers may prevent some people from using these tools. At the same time, these barriers can deepen health inequalities if people with limited digital health knowledge struggle to judge the quality of information in an increasingly digital healthcare environment.

Somalia provides an important example of these structural problems. Although mobile money is widely used, internet access remains very unequal. Overall access has been reported at 9.3%, with much lower access among rural and nomadic populations. This gap, together with the difficulty of reaching healthcare facilities, shows how much people's access to care is limited by their circumstances. A reported 64.0% of respondents identified the physical distance to health facilities as a major problem. For this reason, the use of GenAI to find health information must be understood in the context of these limitations. Existing research does not provide a nationally representative estimate of how many people in Somalia use GenAI to seek health information. The discussion must therefore focus on how factors such as income inequality, digital skills, and the availability of health information checked for local use will shape its growing use. The central contradiction is that GenAI may make health information easier to find while also repeating or worsening existing health inequalities.

From both a technical and a clinical point of view, the risks of GenAI-generated answers are serious. Research has shown that LLMs can produce false or misleading medical information. An answer that sounds clear and convincing is not necessarily true. In specialized fields such as cancer care, reviews combining multiple studies have reported that these systems produce hallucinations, or invented or unsupported information, in about 23% of cases. Tests designed to challenge the systems with difficult clinical situations have found much higher rates, ranging from 50% to 82%. These figures come from specific evaluations and do not show how often hallucinations occur during ordinary consumer use. These systems can also be used to create convincing false information on a large scale. Both accidental technical errors and deliberate manipulation are therefore public health threats. The effects on fairness are also important. If training data do not adequately represent different populations, if systems do not work equally well across groups, or if they are introduced without adapting them to local needs, they may carry assumptions from high-income countries that do not fit fragile healthcare settings.

Moving forward requires a careful, step-by-step plan for managing these technologies, with safe access taking priority over restricting them too early. In the short term, practical steps include sharing affordable public guidance about the appropriate uses of GenAI for health and its limitations. Basic AI and digital health education should also be provided to the public and frontline healthcare workers. Simple rules requiring people to be told when health information has been generated by AI are essential. In the medium term, the focus should shift to checking tools locally and making these checks part of normal practice. This should include oversight groups with experts from different fields, evaluations of Somali-language systems and medical translations, and minimum safety standards for tools that provide information directly to the public. This stage should also include practical ways to monitor false information and shared reporting systems to help coordinate responses to the spread of harmful health information.

Long-term success depends on building lasting systems for responsible oversight. These should include independent audits, assessments of bias and fairness across different population groups, and the development of reliable Somali medical-language datasets. Public trust must be treated as an important part of responsible AI use. Building that trust requires not only improving technical performance but also making responsibility clear, communicating openly, and providing trustworthy ways to review problems and seek remedies. The policy plan should build step by step: first establish guidance and human oversight, then develop local testing and coordination, and only after that move toward more advanced and widespread use in fragile healthcare systems. This approach can help deliver the possible benefits of GenAI, such as improving health knowledge through personalized learning and reducing administrative work for healthcare staff, without weakening the basic principles of safety, fairness, and accountability.

Subject of Research: The public health effects of using generative AI to find health information in places with limited resources, such as Somalia.

Article Title: Generative AI, cyberchondria, and digital health inequality: a public health challenge for Somalia and other fragile health systems

Article References: Aadam, A. A., Daahiye, M. A., Mohamed, A. A., Ali, A. D., & Khalif, I. A. (2026). Generative AI, cyberchondria, and digital health inequality: a public health challenge for Somalia and other fragile health systems. Frontiers in Public Health, 14, 1942140. https://doi.org/10.3389/fpubh.2026.1942140

Image Credits: AI Generated

DOI: 10.3389/fpubh.2026.1942140

Keywords: Generative AI, Cyberchondria, Digital Health Inequality, Fragile Health Systems, Health Misinformation, Somalia, Public Health Policy, LLMs, Health Equity, Generative, digital, health