<p>This scoping review synthesises 29 empirical studies examining how generative artificial intelligence (GenAI) has been conceptualised, assessed, and linked to critical thinking (CT) outcomes in educational contexts. A systematic search of Scopus, Web of Science, and Google Scholar (November 2022–January 2025) yielded 1991 records, of which 29 met the inclusion criteria. The findings indicate a discernible reorientation in how CT is framed in GenAI-related research. While 48% of studies retained classical conceptions of CT as reflective judgement and reasoned decision-making, a substantial proportion re-operationalised CT in AI-specific terms, emphasising learners’ capacity to detect GenAI errors, evaluate credibility, and verify sources. Assessment approaches mirrored this conceptual diversification: although 37.9% of studies relied on standardised psychometric instruments, the majority evaluated CT through context-embedded performance tasks—such as structured debates, synthesis activities, proof correction, and appraisal of GenAI-generated claims—designed to capture CT as enacted within human–AI interaction rather than as a decontextualised cognitive trait. Regarding reported impacts, the evidence reveals a bimodal pattern. In most studies (72.4%), GenAI functioned as a scaffold by reducing lower-order cognitive demands (e.g. information retrieval or drafting), thereby enabling greater investment in higher-order reasoning and synthesis. However, a contrasting offloading pathway was also identified, indicating that unregulated or frictionless use may encourage superficial engagement and reduced epistemic effort. Overall, the synthesis suggests that GenAI is more likely to strengthen CT when instructional designs sustain epistemic accountability through systematic verification, justification, and self-regulated use, rather than treating AI-generated outputs as epistemic endpoints.</p>