The impact of generative AI on critical thinking skills: a systematic review, conceptual framework and future research directions

Purpose The purpose of this study is to systematically review and critically analyze the emerging body of research on how generative artificial intelligence (GenAI) tools impact individuals’ critical thinking skills. It asks: How can GenAI tools increase or decrease the fundamental processes of interpretation, analysis, evaluation and creative inference? Design/methodology/approach The authors developed a comprehensive search string comprising 15 keywords that combined GenAI terms with higher-order cognitive descriptions. For the 2023–2025 timeframe, this search yielded 79 Web of Science papers and 142 Scopus papers. They analyzed and synthesised 68 peer-reviewed papers after filtering, duplication removal and full-text eligibility checks. Findings This study proposes the dual-impact generative-AI critical thinking (DI-GAI-CT) framework, which maps GenAI affordances and mirror-image pitfalls onto five cognitive-metacognitive mediators (prompt quality, self-regulation, engagement, trust, metacognitive critique); three inhibitors (hallucination, automation bias and quick-solution dependence); Murphy’s five-stage critical thinking staircase; and four boundary moderators (task specificity, task complexity, ethical-AI literacy and general AI literacy). A forward-looking agenda then outlines six priority research streams such as multiwave causal tracking, full-constellation modeling and cross-cultural replication. Practical implications In theory, DI-GAI-CT provides the first mechanism-rich model for explaining both uplift and erosion in higher-order reasoning driven by GenAI. In practice, the agenda provides domain-specific levers to organizational leaders, AI designers and educators, such as prompt engineering, metacognitive scaffolding and dual-impact governance, to increase reflective judgment while dampening automation bias. Originality/value To the best of the authors’ knowledge, this is the first review to incorporate a diverse evidence set into a multilevel, dual-stream process model, indicating precisely when, how and why GenAI may either strengthen or undermine critical thinking abilities.