SHAP analysis shows self-testing habits as the strongest predictor of cognitive skill, with GenAI-assisted notes and active learning following closely; for study efficiency, AI usage frequency leads, with research skills next.

The study uses a five-layer framework on cross-sectional survey data from Oct 2024 to Feb 2025, with 14,949 valid responses spanning undergraduates to doctoral students in seven languages.

Limitations include self-reported measures of cognitive skill, potential selection bias, geographic imbalance across ten countries, and no causal inference; authors call for longitudinal designs and objective outcomes like GPA trajectories.

Robustness checks reveal disciplinary differences: STEM predicts cognitive skill less well than social sciences or the full sample, possibly due to AI-use measures focused on writing; private funding at institutions correlates with higher efficiency gains, raising equity considerations for access.

Unlike prior work that treated AI interaction as a single frequency metric, this study treats AI use as predictors of two outcomes (cognitive skill and efficiency) and considers emotional states as pathways.

The multinational study spans nearly 15,000 students across more than 40 countries, emphasizing metacognitive habits over sheer usage frequency in shaping cognitive skills and study efficiency.

The five-stage methodology includes confirmatory factor analysis for validity, OLS baseline predictors, and benchmarking four ML models; Gradient Boosting and Random Forest provide the strongest predictive power for cognitive skill and efficiency.

Task type matters: GenAI-assisted academic writing strongest predicts cognitive skill (standardized coefficient 0.170, p<0.001), with creative and professional writing also influential; anxiety and motivation help predict outcomes, while anxiety reduces efficiency in ML analysis.

K-means clustering identifies four learner profiles—Augmentor, Optimiser, Traditionalists, and Substitutors—showing that Traditionalists still benefit from AI use, challenging the idea that avoiding AI yields better learning.

The main takeaway is that effective human-AI interaction enhances cognitive skill and efficiency through metacognitive scaffolding, task-appropriate applications, and disciplined study habits, with policy implications for structured, equity-aware AI integration in higher education.