Prior work is the need to examine how critical thinking has been conceptualised, assessed, and linked to GenAI in educational contexts

The gap

Evidence profile

Sourced from the recommendations and future work and conclusions and limitations and stated research gap of the source papers, classified as general, drawn from work published between 2024 and 2026, spanning 7 journals. Those papers have been cited 283 times in total.

Supporting evidence — 7 representative gaps

• The Impact of Generative AI Tutors on Metacognitive Monitoring and Self-Regulation in Higher Education: An Experimental Study (2026) · Zenodo (CERN European Organization for Nuclear Research) · doi

  Based on the findings of this experimental study, which demonstrated that generative AI tutors can impair undergraduate students' metacognitive monitoring accuracy and promote cognitive offloading-particularly among novice learners-the following recommendations are proposed for educational practice, instructional design, institutional policy, and future research. 6.1 Recommendations for Educational Practice Educators should integrate explicit metacognitive training into curricula that incorporate generative AI tools. Students must be taught not merely the technical skills of prompting AI systems, but the metacognitive competencies required to use them judiciously. This includes instruction on: ● ● AI assistance. Recognizing the cognitive and metacognitive costs of offloading. Developing strategies for attempting problems independently before seeking ● Critically evaluating AI-generated responses rather than accepting them as authoritative. ● Calibrating self-assessment by comparing internal judgments with external feedback. Furthermore, assessment practices should be diversified to capture process-oriented outcomes alongside final products. Portfolios, reflective journals, and think-aloud assessments can provide insight into students' metacognitive engagement and help counteract the hidden costs of AI use that may not be visible in final outputs alone. Copyright © 2026 by the authors 337 Journal of Contemporary Academic Research and Methodologies (JCARM) Volume 1 | Issue 2 | March 2026 6.2 Recommendations for Instructional Design Developers of generative AI tutors should incorporate design features that promote rather than undermine metacognitive development. Recommended design principles include: ● Metacognitive prompting: AI tutors should be programmed to prompt users for self-assessment before providing answers (e.g., "How confident are you in your answer? Please explain your reasoning before I respond."). ● Structured help sequences: Rather than providing immediate answers, AI tutors should offer graduated assistance, beginning with hints and explanations before revealing solutions. ● Scaffolded verification: When students request verification, AI tutors should encourage self-checking strategies rather than simply confirming correctness. ● Usage analytics: AI platforms should provide students and educators with dashboards displaying patterns of interaction, including frequency of direct answer requests versus explanatory engagement, to foster awareness of help-seeking behaviors. These design features should be developed in collaboration with educational psychologists and subjected to empirical testing to ensure they achieve their intended metacognitive benefits. 6.3 Recommendations for Institutional Policy Highe

  generalrecommendations
  Keywords: metacognitive tutors students design recommendations rather generative educational self assessment help promote cognitive offloading practice
• Generative AI, visual scaffolds, and metacognition in EFL writing (2026) · Language Learning & Technology · doi

  Future implementations of GAI-VGP in L2 writing should address three interrelated priorities. First, scaffolding with fading mechanisms, such as clue-based interactions that prompt inferential reasoning rather than direct correction, can mitigate overdependence by gradually withdrawing support to foster autonomous competence, consistent with Cognitive Load Theory. Second, instructional design should create productive intellectual friction by requiring learners to verify and critically evaluate AI output rather than passively accepting it; Lukešová and Jennings (2026) demonstrate that such constructive engagement shifts attention from surface accuracy toward deeper reasoning. Third, agency-focused instruction should require explicit justification for adopting AI feedback, cultivating agentive literacy and equipping L2 writers to navigate human-AI collaboration critically. To specifically address the monitoring deficit identified in the present study, future designs should incorporate within-task metacognitive prompts that require learners to self-assess coherence and argument development at regular intervals before AI feedback is accessed. Relatedly, because the recorded AI prompts in the present study were relatively limited and appeared to be used selectively at moments of difficulty rather than continuously, future research should examine whether varying levels of AI interaction frequency, and not only its availability, influence the effectiveness of such scaffolding mechanisms.

  generalfuture work
  Keywords: future rather address scaffolding mechanisms reasoning learners critically require feedback present prompts implementations writing three
• Mapping the Research Landscape of AI-Driven Formative Assessment for Argumentation Skill Development in Higher Education: A Bibliometric Study (2026) · TEM Journal · doi

  Future research should focus on the integration of AI with pedagogical principles, the enhancement of collaboration across institutions and nations, and the adoption of an interdisciplinary approach, ensuring that technically sophisticated but also effective in enhancing students’ critical thinking and argumentative skills. For example, the argumentation process dimension has not been studied enough, and in settings.

  generalconclusions
  Keywords: future focus integration pedagogical principles enhancement collaboration across institutions nations adoption interdisciplinary approach ensuring technically
• Learning with large language models: beyond prompt engineering (2026) · Education and Information Technologies · cited 2× · doi

  Future research should explore complementary methods, such as think-aloud pro- tocols or process-tracing approaches, to more directly capture students’ cognitive engagement when interacting with LLMs. Students with more experience using AI tools may demonstrate greater critical thinking, while those with limited exposure may be more likely to accept the responses at face value; future research should examine how prior AI knowledge shapes engagement patterns and verification behaviors. Future research should examine this with a larger sample size and should examine whether gender or other demographic variables influence the way students engage with LLMs for learning, particularly in domains where such disparities are common.

  generallimitations
  Keywords: future students examine engagement llms explore complementary think aloud tocols process tracing approaches directly capture
• Generative AI’s Impact on Critical Thinking: Revisiting Bloom’s Taxonomy (2024) · Journal of Marketing Education · cited 100× · doi

  This study offers valuable insights, but some limitations must be acknowledged. The small, non-representative sam- ple size limits generalizability; however, the idiographic approach allowed for an in-depth exploration of AI-assisted critical thinking (Spencer et al., 2021). Participant perfor- mance varied: one scored moderately on the assignment (63) with full participation (100%), six were high performers with scores between 67 and 72 and full engagement, while Participant 4 had a lower score (53) and 83% participation. As the sample was mainly high-achieving, fully engaged stu- dents, these findings may not fully represent the experiences of lower-performing or less-engaged learners, warranting caution in generalizing results to a wider population. The proposed revision to Bloom’s Taxonomy could face criticism regarding its adaptability across diverse educa- tional contexts and potential overshadowing of traditional critical thinking skills. A comprehensive classification and hierarchy of cognitive processes in AI-assisted education has yet to be developed (Ng et al., 2021), highlighting the need for further research to refine frameworks that address AI-specific competencies. Future research should involve larger, more diverse samples to validate these findings across varied settings, and longitudinal studies to assess AI’s long-term impact on critical thinking (Essien et al., 2024). Expanding the focus to include alternative frame- works, such as those emphasizing intellectual values (e.g., relevance, accuracy, and rigorous reasoning; Carlson, 2013) or decision-making skills (Baldwin et al., 2011), would offer a more holistic understanding of AI’s impact on critical thinking. Table 4 outlines a comprehensive 14 Journal of Marketing Education 00(0) future research agenda for AI-enhanced learning and criti- cal thinking. Finally, while audio diaries provided rich data, they may have influenced participants’ behavior due to self-monitor- ing effects (Dommeyer, 2007) and the reflective nature of journaling, which can prompt deeper thought (Berezan et al., 2023). To gain further insights, future studies could employ reflective essays, cognitive interviews, or other qualitative methods to better understand how critical thinking develops in AI-assisted settings. Incorporating additional metacogni- tive elements would aid in comprehending how students’ awareness, regulation, adaptation, and integration of their learning processes evolve (Parwata et al., 2023). In conclusion, this study highlights the need to rethink Bloom’s Taxonomy in AI-driven education, supporting Mandai et al.’s (2024) call for actionable strategies beyond “general expectations.” As Shaw and Holmes (2014) argue that to effectively foster critical thinking, educators must look beyond rigid objectives and deeply consider the cogni- tive processes involved at each level. The true challenge is to leverage AI as a tool for enriching critical thinking, rather than allowing it to become a superficial aid.

  generallimitationsevidence 5/5
  Keywords: thinking critical assisted processes education future insights must participant varied full participation high lower fully
• The cognitive paradox of AI in education: between enhancement and erosion (2025) · Frontiers in Psychology · cited 179× · doi

  The paper identifies a gap in the understanding of the impact of AI on cognitive development, particularly in critical thinking, problem-solving, and recall. It highlights the need for a balanced approach to AI integration in education to avoid cognitive dependency. The study emphasizes the importance of considering the psychological and cognitive theories, such as Cognitive Load Theory and Self-Determination Theory, in AI-based education.

  generalstated research gapevidence 5/5
  Keywords: paper identifies gap understanding impact cognitive development particularly
• Towards a framework for understanding and assessing critical thinking in generative AI-enhanced learning environments: a scoping review (2026) · European Journal of Psychology of Education · cited 2× · doi

  The gap in prior work is the need to examine how critical thinking has been conceptualised, assessed, and linked to GenAI in educational contexts. Prior work has not fully addressed the concerns about learners' epistemic agency in GenAI-enhanced learning environments. The paper identifies a need to map conceptualisations, measures, and reported impacts of critical thinking in GenAI-enhanced learning contexts.

  generalstated research gapevidence 5/5
  Keywords: gap prior work need examine critical thinking has

Questions about this gap