Assessment of AI’s Impact on Critical Thinking in Higher Education: A Systematic Review of Evaluation Methods

Abstract

Background: The integration of artificial intelligence (AI) tools in higher education has accelerated rapidly, prompting growing interest in their impact on students’ critical thinking (CT) skills. While some studies suggest that AI can enhance analytical reasoning and reflective learning, others raise concerns about overreliance and cognitive offloading. Despite this growing literature, the conceptualization and measurement of “critical thinking” remain inconsistent across studies, complicating efforts to synthesize findings. Objectives: This systematic review examines how recent empirical studies (2022–2025) define, operationalize, and assess critical thinking in the context of AI-enhanced learning in higher education. Specifically, it investigates the theoretical frameworks employed, the assessment tools used, the types of AI tools integrated, and the reported outcomes on CT development. Methods: Following the PRISMA 2020 guidelines, we conducted a systematic search in Scopus, Web of Science, and ERIC for studies published between 2022 and February 2025 that involved higher education student populations, AI-based tools, and CT-related outcomes. A total of 22 eligible studies were identified and analyzed using narrative synthesis and thematic coding. Risk of bias was assessed using JBI and CASP tools. Results: Only 8 of the 22 studies provided a formal definition of critical thinking, and even fewer used dedicated CT assessment instruments. Most studies relied on mixed methods and domain-specific performance tasks. The findings indicate that AI tools can support CT development, particularly when embedded in human-facilitated learning environments that promote reflection, evaluation, and dialogue. However, studies also reported risks such as superficial learning and diminished metacognitive engagement when AI was used as a cognitive substitute. Conclusions: AI’s impact on critical thinking in higher education is shaped by tool design, instructional context, and the clarity of CT conceptualization. This review highlights the need for consistent definitions, theoretically grounded assessments, and pedagogical models that combine AI affordances with reflective, instructor-guided learning. Future research should emphasize longitudinal designs and the development of CT-specific instruments aligned with validated frameworks.