Seven in Ten Ethiopian Students Use AI Chatbots for Medical Self-Care, Study Finds

When university students in south-eastern Ethiopia feel unwell, a growing number of them are turning not to a clinic or a pharmacist but to a chatbot. A new cross-sectional study at Arsi University has found that just over seventy percent of undergraduates use generative artificial intelligence tools such as ChatGPT and Gemini for medical self-care, a striking figure that offers one of the first systematic measurements of how rapidly these technologies are reshaping health behavior in a low-income setting. The research, published in PLOS Digital Health, surveyed 414 randomly selected students and arrives at a moment when conversational AI systems have quietly become, for millions of young people worldwide, a first point of contact with the health system.

The scale of the phenomenon matters because generative AI occupies an unusual position between information source and medical advisor. Unlike a search engine that returns links, tools built on large language models produce individualized, authoritative-sounding answers to questions about symptoms, medications, and dosages. That conversational fluency can genuinely support informed self-care decisions, helping users understand conditions and weigh options before seeking help. But the same fluency can mask errors: language models can hallucinate drug interactions, misread symptom patterns, and present confident prose where a clinician would express uncertainty. The Arsi University researchers explicitly warned that over-reliance on these systems may lead to incorrect self-medication and delayed consultations with medical professionals when the technology is not managed appropriately.

To quantify the trend, the team administered a self-administered questionnaire to students whose mean age was 22.2 years, collecting responses through the Kobo Toolbox digital data platform. The survey achieved a response rate of 97.87 percent, an unusually high figure that strengthens confidence in the representativeness of the sample. The researchers combined descriptive statistics with binary logistic regression, a statistical technique that estimates how each candidate factor changes the odds of a behavior while holding the others constant. The outcome measure was straightforward: whether a student had engaged in generative AI–assisted medical self-care, meaning the use of tools like ChatGPT or Gemini for activities such as self-diagnosis or self-medication.

The headline result was the prevalence itself. Seventy point zero five percent of the students reported using generative AI for medical self-care, confirming what had previously been only anecdotal: in Ethiopia, as elsewhere, the explosion of social media and free AI tools has dramatically lowered the barrier to personalized health information. University students are a particularly revealing population for this kind of study. They are digitally fluent, they live away from family support structures, and they often face long queues and costs at student health services, making a free, instant, anonymous chatbot an attractive alternative. What the regression models added was a portrait of exactly which students were most drawn to this new form of digital triage.

Several characteristics increased the odds of AI-assisted self-care. Second-year students had nearly three times the odds of engaging in the behavior compared with their peers, an adjusted odds ratio of 2.98, perhaps reflecting a cohort at the intersection of internet familiarity and still-developing confidence in navigating campus health services. Students with good digital health literacy, the ability to find, evaluate, and apply online health information, showed 71 percent higher odds, while those who were aware of generative AI technologies had more than double the odds, at 2.18. A positive attitude toward the technology raised the odds by 63 percent, suggesting that disposition toward AI shapes behavior independently of knowledge alone. Students who had recently visited a health facility were also more likely to use AI for self-care, with an adjusted odds ratio of 2.46, hinting that contact with formal care may prompt follow-up questions that students prefer to ask a machine.

Two factors pointed in the opposite direction, and they may be the most clinically significant findings in the study. Students with good health-seeking behavior, the established habit of consulting professionals when ill, had 39 percent lower odds of AI-assisted self-care, an adjusted odds ratio of 0.61. Those who visited health facilities infrequently had roughly half the odds, at 0.49. Taken together, the pattern suggests that generative AI functions partly as a substitute for formal care rather than a complement: students already inclined to seek professional help rely less on chatbots, while those disconnected from health services appear to fill the gap with algorithms. For public health planners, that substitution effect is the crux of the risk, because the students most dependent on AI advice may be those least equipped to recognize its errors.

The Ethiopian context sharpens these concerns. The study’s authors note a remarkable absence of systematic empirical data on AI-assisted self-care in the country, and by extension across much of sub-Saharan Africa, even as smartphone penetration and free access to English-language chatbots have surged. Generative AI tools are trained predominantly on data and text from high-income countries, and their performance on tropical diseases, locally available medications, and regionally specific clinical guidelines remains poorly characterized. A chatbot that confidently recommends a medication unavailable in an Ethiopian pharmacy, or misjudges a symptom presentation common in the region, could steer a self-medicating student in a dangerous direction with no pharmacist present to catch the mistake.

Yet the researchers stop short of framing the trend as simply harmful, and the data themselves support a more nuanced reading. Awareness and literacy were associated with greater use, not less, which implies that students who understand these tools are not avoiding them but integrating them into their health routines. The positive association with recent health facility visits suggests many users may employ AI as a supplement, researching conditions after or between consultations. The central challenge, on this view, is not to discourage use but to make it safe: to ensure that the fastest-growing source of health advice for young adults is accompanied by the skills to interrogate it, verify it, and recognize when a conversation with a chatbot must become a conversation with a clinician.

That is precisely the policy prescription the study advances. The authors call for formal guidelines governing generative AI use in health contexts and for training programs that build digital health literacy among students, treating the ability to evaluate AI outputs as a core health competency rather than an optional technical skill. Universities, they imply, are the natural venue: campus health services could openly engage with the reality that most of their patients are already consulting AI, offering guidance on prompt quality, verification of recommendations, and red-flag symptoms that demand immediate professional attention. As large language models become embedded in everyday devices, the Arsi University findings offer a rare quantitative baseline from a setting where the shift is happening fast, and a warning that the health system’s newest front door, staffed by machines, is already open whether institutions are ready or not.

Subject of Research: Generative AI-assisted medical self-care practices and associated factors among undergraduate students in Ethiopia

Article Title: Generative artificial intelligence–assisted medical self-care and associated factors among undergraduate students at Arsi University, South-Eastern Ethiopia

Article References: Adem, J. B., & Alhur, A. A. (2026). Generative artificial intelligence–assisted medical self-care and associated factors among undergraduate students at Arsi University, South-Eastern Ethiopia. PLOS Digital Health, 5(9), e0001748. https://doi.org/10.1371/journal.pdig.0001748

Image Credits: AI Generated

DOI: 10.1371/journal.pdig.0001748

Keywords: generative artificial intelligence, ChatGPT, medical self-care, self-diagnosis, self-medication, digital health literacy, Ethiopia, Arsi University, health-seeking behavior, cross-sectional study, PLOS Digital Health, university students