Answers are grounded in this paper and cite their sources.
Generative artificial intelligence (GenAI) can improve task performance when support is available, yet these gains do not necessarily indicate what learners retain independently. This Mini Review examines three assessment conditions: assisted task performance, immediate testing without AI, and delayed testing without AI. Relevant evidence was identified through searches of the Web of Science Core Collection and supplementary source identification. Across the reviewed experiments, outcomes varied with task demands, instructional design and the skills assessed. Unrestricted AI assistance was associated with weaker subsequent performance in some comparisons, whereas structured AI-supported instruction showed benefits for selected outcomes. None of the twelve selected studies documented all three conditions with verified AI exclusion at both no-AI posttests. Variation in baseline measurement, follow-up intervals and instructional comparators further limits cross-study interpretation. Retrieval attempts, elaboration and feedback use offer plausible explanations, but their roles as mediators remain largely untested. Future studies should document AI availability during learning and testing, specify the constructs assessed, and report learning activities between assessments.
No takes yet. Share an insight, caveat, or question.
Junjie Wei (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: