ChatGPT use, perceived usefulness, and self-reported clinical reasoning among undergraduate nursing students in Uganda: a cross-sectional study

Generative artificial intelligence is increasingly used in health-professions education, yet multi-university evidence from African nursing programmes remains limited. This study examined patterns of ChatGPT use, perceived usefulness, and their associations with self-reported clinical reasoning among undergraduate nursing students in Uganda. A quantitative cross-sectional survey was conducted from 2 to 27 February 2026 among undergraduate nursing students from 13 Ugandan universities (six public and seven private) who had previously used ChatGPT for nursing-related academic or clinical learning. Students were sampled by year of study within participating institutions. Of 480 invited students, 420 submitted questionnaires meeting the prespecified completeness criterion, yielding an 87.5% response rate and a final analytical sample of 420. A 41-item structured questionnaire assessed eight ChatGPT-use activities, perceived usefulness, self-reported clinical reasoning, challenges, and responsible-use practices. Data were analysed in IBM SPSS Statistics version 31.0 using descriptive statistics, Pearson correlation, group comparisons, and exploratory multiple linear regression. Students reported frequent ChatGPT use across eight nursing-learning activities, including understanding disease processes, clarifying nursing concepts, examination preparation, care-plan development, clinical procedures, patient-case analysis, medication information, and comparison of nursing interventions (descriptive grand mean = 4.12, SD = 0.81). Perceived usefulness was very high (grand mean = 4.28, SD = 0.73), and self-reported clinical reasoning was high (grand mean = 4.17, SD = 0.76). ChatGPT usage was moderately and positively associated with self-reported clinical reasoning ( r = 0.661, 95% CI [0.603, 0.712], p < 0.001). The exploratory regression model was significant, F(7, 412) = 78.34, p < 0.001, adjusted R² = 0.563; multicollinearity was not indicated by the VIF range of 1.08–1.98. A null random-intercept model indicated minimal between-university clustering (ICC = 0.008; 95% CI [0.000, 0.071]). Because the design was cross-sectional and all principal constructs were self-reported, these findings represent associations between reported behaviours and perceptions rather than causal effects or objectively measured clinical-reasoning performance. ChatGPT use was common and positively associated with students’ perceptions of their clinical-reasoning abilities, particularly in the context of high perceived usefulness and reported verification practices. These findings support cautious, faculty-guided educational use of generative AI, but they do not demonstrate that ChatGPT objectively improves clinical reasoning. Future studies should use validated measurement models, objective performance-based outcomes, longitudinal or experimental designs, and multilevel or cluster-robust analyses when institution-level dependence is relevant.