Mapping the Scaffolding of Metacognition and Learning by AI Tools in STEM Classrooms: A Bibliometric-Systematic Review

Synthesis: A bibliometric–systematic review of AI tools in STEM education:

Systematic review (2005–2025) mapping how AI tools scaffold and co-regulate metacognitive development in STEM classrooms through bibliometric and qualitative synthesis.

Scope and Methodology

Tsakeni et al. (2025) conducted a bibliometric–systematic review of AI tools in STEM education:

• Dataset: 135 peer-reviewed articles (2005–2025) from Scopus and Web of Science
• Core studies: 24 studies selected via PRISMA 2020 protocols for in-depth review
• Methods: Bibliometric mapping (Biblioshiny, VOSviewer) + qualitative thematic synthesis
• Theoretical frameworks: Flavell's metacognition theory, General Systems Theory, Human-Centered vs. Posthumanist paradigms

Key Findings

1. Theoretical Frameworks Evolution

Paradigm	Core Idea	AI Role
Human-Centered (traditional)	AI supports human agency with ethical oversight	Supportive tool under teacher control
Posthumanist (emerging)	Learning = co-regulated process distributed between humans and AI	AI as co-agent in reflective processes

Finding: While most research remains grounded in human-centered conceptualizations, there are emerging indications of posthumanist framings where AI systems are positioned as co-regulators of learning.

2. AI Tools Scaffolding Metacognition

Tool Category	Examples	Metacognitive Function
Intelligent Tutoring Systems (ITS)	Carnegie Learning, ALEKS	Personalized feedback, real-time monitoring, strategic prompts
Adaptive Platforms	Deep RL-based systems	Enhance metacognitive outcomes via adaptive scaffolds
Learning Analytics	Dashboards, tracking tools	Externalize metacognitive processes, support reflection
Generative AI	ChatGPT (GPT-4.1, 4.5, 5), conversational agents	Metacognitive prompts, chatbot feedback, ME-CoT approaches

Core shift: From individual reflection → system-level AI Regulation in Education and distributed cognition.

3. Metacognitive Outcomes

Studies consistently show AI tools improve:

• Reflective thinking and independent inquiry
• Self-regulation and strategic monitoring (planning, monitoring, evaluation)
• Motivation and reasoning (especially for low-achieving and underrepresented groups)
• Academic performance in mathematics, science, chemistry, physics

4. Ethical Concerns and Pitfalls

Despite benefits, scholars warn of:

• Cognitive overload from excessive AI-generated feedback
• Reduced learner autonomy when AI replaces (not scaffolds) reflection
• Algorithmic bias if AI prioritizes automation over deep reflection
• Inconsistent measurement methods across studies (theoretical fragmentation)

Connection to Existing Work

vs. Metacognition in AI Education

• This review maps the field systematically (135 publications, 24 core studies)
• Confirms: metacognition is central to STEM success but challenging to foster
• AI tools offer scalable Scaffolding, but require teacher preparation in AI literacy

vs. Adaptive Learning Systems

• ALEKS, Carnegie Learning highlighted as successful adaptive platforms for metacognition
• Deep reinforcement learning enhances metacognitive outcomes (vs. static adaptive rules)
• Aligns with: system-level regulation > individual reflection

vs. Intelligent Tutoring Systems

• ITS identified as key scaffolding tool for metacognitive development
• Combines content mastery (object level) with reflective monitoring (meta level)
• Consistent with: The Evidence Base on AI in K-12: A 2026 Review — domain-specific tutoring outperforms generic chatbots

vs. AI Literacy

• Critical finding: AI literacy must be integrated into teacher preparation
• Teachers need skills to: select tools, interpret analytics, maintain ethical oversight
• Aligns with: human-centered paradigm (AI as tool, teacher as agentic decision-maker)

The Posthumanist Turn

The review identifies an emerging paradigm shift:

Human-Centered (Traditional):

Teacher → AI Tool → Student
(AI as instrument)

Posthumanist (Emerging):

Teacher ↔ AI System ↔ Student
(Learning as distributed, co-regulated process)

Implications:

• AI systems as co-regulators (not just tools)
• System-level effects (not just individual cognition)
• Challenges the human–machine binary in education

What this means for practice

• Instructors. Treat AI Literacy as a prerequisite for tool selection: understand how a specific tool scaffolds planning, monitoring, and evaluation before adopting it for metacognitive work.
• Schedule reflection around the tool rather than assuming it. Add explicit reflection prompts and self-monitoring checkpoints so AI enhances metacognitive processes instead of replacing them, and cap AI-generated feedback to avoid the cognitive overload the review flags.
• Combine tool categories instead of betting on one: pair an ITS for content mastery with Learning Analytics dashboards that externalize monitoring, and use Generative AI for metacognitive prompts.
• Researchers. Standardize how metacognitive outcomes are measured. The review finds measurement methods inconsistent across studies, which blocks meta-analysis and makes cross-study comparison of effects unreliable.
• Researchers. Prioritize longitudinal and experimental designs: most reviewed work is cross-sectional, so the review can show that AI tools accompany better reflective thinking and self-regulation but cannot establish causal or sustained effects.

Limitations

• The in-depth synthesis rests on 24 studies selected through PRISMA 2020 screening of a 135-article corpus (2005–2025) drawn from Scopus and Web of Science, so work outside those indexes is absent.
• The authors state that few included studies used longitudinal or experimental designs capable of identifying causal or sustained effects of AI-mediated metacognitive interventions.
• Coverage is uneven across STEM: reviewed studies concentrate in mathematics and science education, with limited representation in technology and engineering education and in early-childhood and teacher-education contexts.
• Measurement is inconsistent across the reviewed studies — the review's own stated barrier — because metacognitive outcomes were assessed with differing instruments, so reported benefits cannot be pooled or compared directly.

Citation

Tsakeni, M., Nwafor, S. C., Mosia, M., & Egara, F. O. (2025). Mapping the Scaffolding of Metacognition and Learning by AI Tools in STEM Classrooms: A Bibliometric–Systematic Review Approach. Journal of Intelligence.