AI Use Linked to Sharper Thinking Among Nigerian Medical Students, Study Finds

Artificial intelligence has swept into classrooms and lecture halls around the world with remarkable speed, but one of the most pressing questions has been what it does to the minds of the students who use it most intensively. A new study from Ibadan, Nigeria, offers one of the clearest snapshots yet of how future health professionals in a low- and middle-income country are engaging with tools such as ChatGPT, and what that engagement means for their thinking and their mental wellbeing. The research, published in BMC Medical Education, surveyed 365 undergraduate medical and allied health students across three tertiary institutions in Ibadan and found that more than nine in ten of them were already using artificial intelligence for academic purposes. The findings are striking not because they reveal a crisis, but because they complicate the familiar narrative that heavy AI reliance erodes critical thinking and damages mental health.

The study, led by Ishaq Maryam Damilola and Henry Osaro Aisagbonhi of the Department of Health Policy and Management at the College of Medicine, University of Ibadan, was designed as an institution-based analytical cross-sectional survey. The researchers used multistage sampling to recruit undergraduates from three tertiary institutions, with the University of Ibadan contributing the largest share of participants at 44.7 percent. The demographic profile of the sample reflected the composition of Nigerian health professions faculties: 47.1 percent of respondents were aged between 18 and 21 years, 63.0 percent were female, and nursing students formed the largest programme group, accounting for 36.2 percent of the total. Data were collected with a study-specific structured questionnaire, and the analysis combined descriptive statistics, t-tests, analysis of variance, Pearson’s correlation, chi-square tests, and multivariable linear regression, with statistical significance set at a p-value below 0.05.

The headline finding on adoption is unambiguous. Overall, 92.3 percent of the students reported using artificial intelligence for academic purposes, and 43.0 percent described themselves as daily users. ChatGPT emerged as the dominant tool, a pattern consistent with its global popularity but notable in a setting where access to paid platforms and reliable connectivity cannot be assumed. The frequency of use was treated as the primary exposure in the analysis, categorised as rare, occasional, weekly, or daily. This granularity matters, because the relationship between AI and cognition is unlikely to be a simple yes-or-no question; what the researchers wanted to know was whether intensity of use tracked with measurable differences in how students reason and how they feel.

To measure reasoning, the team constructed a 16-item composite Cognitive Score with a possible range of 16 to 80, designed to capture cognitive reflection, the capacity to override intuitive but incorrect first answers and engage in deliberate, analytical thought. Psychological distress was assessed with a 24-item composite Psychological Score ranging from 24 to 120, where higher values indicate greater distress. These composite instruments allowed the researchers to test two separate hypotheses in a single design: whether AI use is associated with sharper or duller analytical thinking, and whether it is associated with better or worse psychological outcomes. Both questions have been fiercely debated in the education literature, but rarely examined together, and almost never in the context of health professions training in sub-Saharan Africa.

The correlational results point in two different directions. AI-use frequency showed a weak positive correlation with cognitive scores, with a Pearson’s correlation coefficient of 0.120 and a p-value of 0.028, suggesting that students who used AI more often scored slightly higher on cognitive reflection. At the same time, AI-use frequency showed a moderate negative correlation with psychological scores, with a coefficient of -0.256 and a p-value below 0.001. Because higher psychological scores indicate greater distress, this negative correlation means that more frequent AI users reported lower levels of psychological distress. On the surface, then, the students leaning hardest on artificial intelligence were also the ones thinking slightly more analytically and feeling somewhat less distressed.

Correlation, however, is not causation, and the researchers were careful to push their analysis further. When they entered AI-use frequency into a multivariable linear regression adjusted for demographic and institutional factors, the association with cognitive reflection remained statistically significant and independent. Each increment in AI-use frequency was associated with an increase of 0.96 points on the cognitive score, with a 95 percent confidence interval of 0.23 to 1.68 and a p-value of 0.010. The psychological association, by contrast, did not survive adjustment. The regression coefficient for psychological outcomes was -0.76, with a confidence interval spanning from -2.17 to 0.66 and a p-value of 0.293. In other words, once institutional and demographic context was accounted for, AI use could no longer be distinguished from other factors as a driver of psychological wellbeing.

This divergence between the cognitive and psychological results is arguably the most important technical insight of the study. The authors conclude that psychological outcomes appear to be shaped more strongly by institutional context than by AI use itself. That interpretation fits the pattern in the data: the raw correlation between AI use and distress was moderate, but it dissolved under adjustment, which is exactly what one would expect if the students’ mental state were being driven by the environment of their institution, their workload, or their stage of training rather than by their chatbot habits. The cognitive effect, though modest in magnitude, was robust to the same adjustments, hinting at a genuine, if small, association between engaging with AI tools and exercising analytical reasoning.

The study also identified who was using AI and where. AI use varied significantly with age, with a p-value of 0.047, and varied even more sharply across institutions, with a p-value below 0.001. Gender, however, made no measurable difference, with a p-value of 0.450. The institutional variation is particularly telling for policymakers: if two students of similar age and gender studying comparable disciplines show markedly different AI adoption depending on which campus they attend, then access, culture, and institutional policy are doing much of the work. In a country where universities differ widely in digital infrastructure, this finding suggests that the AI divide in health professions education may be less about individual enthusiasm and more about institutional provision.

Yet the study does not read as an unqualified endorsement of AI in medical training. One of its most sobering results concerns verification behaviour. Daily users of AI were substantially more likely than occasional users to accept AI-generated responses without checking them, at 62.8 percent versus 41.8 percent. For future doctors, nurses, and allied health professionals, that gap is not a trivial statistic. Large language models can produce fluent, confident, and occasionally incorrect clinical information, and a habit of uncritical acceptance formed in the classroom could translate into risky shortcuts in professional practice. The finding suggests that familiarity breeds trust, and that the very students who benefit cognitively from frequent AI engagement may also be the most vulnerable to its failure modes.

The authors frame their conclusions as a call for institutional action rather than individual restriction. They recommend institution-level AI literacy programmes, governance frameworks governing appropriate use, and curricula that explicitly promote critical appraisal of AI-generated content and responsible use of these tools in health professions education. The study received no external funding, was approved by the University of Ibadan/University College Hospital Ethics Review Board under approval number UI/EC/25/0530, and was published open access on 7 October 2026. As generative AI becomes as ordinary a study aid as the textbook, evidence like this from Nigeria will be essential: it shows that the technology’s effects on young minds are neither uniformly harmful nor uniformly benign, but depend heavily on the structures that surround its use.

Subject of Research: Artificial intelligence use and its cognitive and psychological correlates among medical and allied health students in Nigeria

Article Title: Artificial intelligence use and its cognitive and psychological correlates among medical and allied health students in Ibadan, Nigeria

Article References: Damilola, I. M., & Aisagbonhi, H. O. (2026). Artificial intelligence use and its cognitive and psychological correlates among medical and allied health students in Ibadan, Nigeria. BMC Medical Education. https://doi.org/10.1186/s12909-026-10558-3

Image Credits: AI Generated

DOI: 10.1186/s12909-026-10558-3

Keywords: artificial intelligence, medical education, cognitive reflection, psychological distress, ChatGPT, health professions education, Nigeria, cross-sectional study, digital learning, AI literacy, students, mental health