Generative AI Risks Exacerbating Health Inequality and Cyberchondria in Fragile Systems

Generative artificial intelligence (GenAI) is fundamentally reshaping the landscape of global health information by transforming how individuals access, interpret, and act upon medical data. This technology introduces rapid, personalized, and conversational interfaces that move beyond the limitations of traditional static websites or conventional search engines. Unlike previous digital health tools that merely retrieve existing sources, conversational AI synthesizes probabilistic information into fluent, tailored responses, often appearing direct and authoritative to users seeking explanations for symptoms or guidance on next steps. While this shift offers immense potential for health education and reducing information barriers, particularly in regions where formal healthcare access is constrained, it simultaneously introduces profound public health challenges related to misinformation, uncertainty, misplaced trust, and escalating health anxiety. This perspective examines the potential nexus between GenAI-mediated health-information seeking, the phenomenon of cyberchondria, systemic digital health inequality, and the inherent vulnerability of fragile health systems, using the structural conditions of Somalia as a critical case study.

Cyberchondria, defined as excessive or repetitive online health-information seeking, is a well-documented behavioral pattern that can intensify health anxiety and uncertainty. However, the emergence of GenAI introduces a qualitatively different informational environment. Because these systems provide immediate, interactive, and seemingly definitive answers, they may alter the established drivers of cyberchondria—such as information overload and uncertainty—by delivering fast, personalized responses rather than merely ranking a list of sources. The core hypothesis explored is that conversational AI may not just extend existing patterns of symptom searching, but actively modify them. This interaction style could potentially lower the threshold for repeated symptom exploration and increase the persuasive force of the information returned, requiring rigorous empirical investigation to confirm.

A critical caveat to this potential interaction is the inherent nature of large language models (LLMs). These systems are not infallible; they can generate plausible yet unsupported content, omit relevant guidance, produce variable answers even from identical prompts, and display a high degree of overconfidence despite limited reliability. Direct evidence establishing a causal link between GenAI use and cyberchondria remains elusive. However, the mechanistic hypothesis suggests that when users engage with systems that are conversational and fast, yet probabilistic, incomplete, and sometimes contradictory, unresolved uncertainty may persist or even intensify. For individuals predisposed to health anxiety, this scenario plausibly contributes to the familiar cyberchondria pathway: moving from initial symptom concern to repeated checking, transient reassurance, renewed doubt, and ultimately, escalating distress.

The application of these technologies in fragile and resource-constrained settings, such as Somalia, amplifies these concerns. While digital health adoption has accelerated across low- and middle-income countries (LMICs), its scale-up is frequently hindered by fragmented ecosystems, weak interoperability, and scarce skilled labor, rather than a simple lack of technological interest. In contexts where physical access to clinicians is uneven and referral pathways are weak, AI-generated information can carry disproportionately significant practical weight because timely professional interpretation is often unavailable. Furthermore, the consequences of AI deployment are heavily dependent on the uneven distribution of digital and health literacy. Limited connectivity, device access, and language fit can either hinder participation or, conversely, deepen health inequities by allowing individuals with low digital health literacy to struggle to judge source quality in an increasingly digitized care environment.

Somalia serves as a vital context for understanding these structural dynamics. Despite high mobile-money usage, internet access remains highly unequal, with overall access reported at 9.3%, significantly lower in rural and nomadic populations. This disparity, coupled with the persistent physical barrier of distance to health facilities—cited as a major problem by 64.0% of respondents—means that GenAI-mediated information seeking must be viewed through a lens of structural constraint. The current literature lacks a nationally representative estimate of GenAI-mediated health-information seeking in Somalia; therefore, the discussion must focus on how structural factors—including socioeconomic inequality, digital literacy, and the availability of locally validated health information—will shape its emerging use. The central paradox is that while GenAI may reduce information scarcity, it simultaneously risks reproducing or even amplifying existing health inequalities.

From a technical and clinical standpoint, the risks associated with GenAI outputs are significant. Established evidence demonstrates that LLMs can generate false or misleading medical content, and fluency does not equate to truth. In specialized fields, such as oncology, meta-analyses have reported hallucination incidences of approximately 23%, while testing under deliberately adversarial clinical conditions revealed substantially higher hallucination rates, ranging from 50% to 82%. It is crucial to note that these estimates are evaluation-specific and do not represent the prevalence of hallucinations during routine consumer use. These systems can be exploited to generate credible-sounding misinformation at scale, making both inadvertent technical error and deliberate manipulation relevant public health threats. The equity implications are equally pressing, as inadequate representation in training data, poor transportability across diverse populations, and deployment without local adaptation can encode assumptions from high-income contexts that do not travel well in fragile settings.

The path forward requires a nuanced, phased governance roadmap that prioritizes safe access over premature restriction. In the immediate term, feasible actions include disseminating low-cost public guidance on the appropriate uses and inherent limits of GenAI for health, alongside building basic AI and digital-health literacy for both the public and frontline workers. Simple transparency norms, requiring disclosure when health information is AI-generated, are essential. Medium-term priorities must pivot toward local validation and institutionalization, involving multidisciplinary oversight groups, Somali-language and medical-translation evaluations, and establishing minimum clinical safety standards for public-facing tools. This stage also requires practical misinformation monitoring and shared reporting channels to build a coherent infodemic response infrastructure.

Long-term sustainability hinges on durable governance capacity, which includes independent auditing, bias and fairness assessments across diverse population groups, and the development of robust Somali medical-language datasets. Public trust must be treated as a multidimensional component of responsible AI deployment, requiring not only technical performance improvements but also clear accountability, transparent communication, and credible mechanisms for oversight and redress. The policy logic is cumulative: one must first establish guidance and human oversight, then build validation and coordination, and only subsequently move toward advanced, scaled integration into fragile health systems. This approach ensures that the potential benefits of GenAI—such as enhancing health literacy through personalized learning or reducing administrative burdens for healthcare workers—are realized without undermining the foundational principles of safety, equity, and accountability.

Subject of Research: The public health implications of using generative AI for health information seeking in resource-constrained settings like 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