Review of 27 studies finds AI skill assessment tied to better vocational learning

The studies were drawn from major international and Chinese databases under PRISMA 2020 guidelines. Results varied considerably across study designs, outcome measures and implementation contexts, and the authors say regional and institutional subgroup differences should be read cautiously because of small samples and mixed methods.

From the evidence the authors propose a conceptual framework for AI-supported skill assessment. It separates empirically grounded components from forward-looking extensions tied to generative AI, and is meant as a baseline for future research and system design rather than a prescription.

The authors say research on AI-enabled assessment is fragmented and methodologically uneven, particularly in vocational settings, despite growing interest in generative AI in higher education. This review sought to consolidate the empirical evidence.

Twenty-seven studies met the inclusion criteria. The random-effects model gave a pooled Hedges' g of 0.72 for skill-related outcomes, a moderate positive association, though effects varied widely between studies.

Subgroup analyses suggested differences across regional and institutional settings. The authors caution that small sample sizes and diverse methods limit interpretation of these differences.

The conceptual framework they propose is evidence-informed but not prescriptive. It is intended to support future research, system design and responsible integration of generative AI in assessment, and the authors call for more theory-informed and transparent studies.