Artificial Intelligence-Supported Learning and Its Effect on Knowledge Acquisition and Practical Performance Among Nursing Students: A Systematic Review and Meta-Analysis

Artificial intelligence (AI) tools, including chatbots, generative language models, and AI-enabled simulation, are increasingly used in nursing programs, yet it remains unclear whether improvements in knowledge translate into observed clinical performance. This systematic review and meta-analysis, reported in line with PRISMA 2020, synthesised comparative studies of AI-supported learning among nursing students, treating knowledge, practical performance, and reasoning outcomes separately. PubMed, CINAHL, Scopus, and ScienceDirect were searched from inception, and screening and data extraction were conducted independently by the primary author and verified by a research colleague. Risk of bias was appraised with ROBINS-I (Risk Of Bias In Non-randomized Studies - of Interventions) for non-randomized studies and RoB 2 (Risk of Bias tool version 2) for randomized trials, and standardized mean differences were to be pooled with random-effects models where studies reported comparable outcomes. Not all eligible studies could be obtained in full text, so the synthesis was restricted to those that were accessible. These were non-randomized, single-site studies, most of them at serious risk of bias. Findings for knowledge were inconsistent: one study showed a very large benefit for a purpose-built chatbot, another showed no clear difference, and the disagreement was too great to support a meaningful pooled estimate. None of the analyzed studies assessed observed clinical performance, and one study of a general-purpose language model reported poorer rubric-scored work than textbook-based learning. Self-reported reasoning outcomes were uncertain and rested on unblinded self-rating. Current evidence does not show that AI-supported learning improves clinical performance among nursing students, and inaccessible full texts and the methodological weaknesses of the primary studies limit the conclusions. Future research should use randomized designs, validated observed-performance measures with blinded assessment, and active comparators matched for time and content.