Integrating Generative Artificial Intelligence into Student Learning: A Systematic Review from a TPACK Perspective

Synthesis: Liu and Zhong (2025) systematically review 71 high-quality empirical studies (74 effect sizes) of generative AI integration into student learning, published after ChatGPT's release (November 30, 2022), through the lens of the TPACK framework. A random-effects meta-analysis finds a medium-to-large overall effect on learning outcomes (Hedges' g = 0.752, 95% CI [0.566, 0.937]), with strong cognitive (g = 0.831) and affective (g = 0.729) gains but a negligible effect on behavioral engagement. The review identifies two critical priorities: developing students' GenAI literacy and establishing GenAI-TPACK professional development for teachers.

Key Findings

• Overall positive effect. Across 47 journal articles and 74 effect sizes, learners using GenAI outperformed control groups (Hedges' g = 0.752, p < 0.001), a medium-to-large effect.
• Domain-specific outcomes. GenAI showed a strong effect on cognitive outcomes (SMD = 0.831, 95% CI [0.596, 1.066]) and a medium-to-large effect on affective outcomes (SMD = 0.729, 95% CI [0.414, 1.044]), but only a small, non-significant effect on behavioral engagement (SMD = 0.057, p = 0.828).
• Uneven adoption. GenAI is more widely applied in higher education than in K-12; quasi-experimental designs dominate; most studies focus on pedagogical development rather than curriculum development.
• GenAI as a tool role. GenAI most often acted as a learning tool, with tool roles varying by discipline. Pedagogical Scaffolding of GenAI use was frequently incomplete.
• Moderating variables. Research method/focus and key instructional-design variables (learning content, tool, strategy, assessment, outcome) moderated learning outcomes.
• Two critical priorities. (1) GenAI literacy development for students, and (2) GenAI-TPACK professional development for teachers.

Research Landscape

The 71 studies span publishing year, country/region, authorship type, sample size, learning stage, educational setting, duration, and research method. GenAI-integrated student learning has attracted global attention, is concentrated in higher education, and is dominated by quasi-experimental designs that require larger samples and longer durations than other methods. The review emphasizes the need for international collaboration, exploration of pedagogical strategies for K-12 higher-order thinking, more rigorous controlled experiments, and systematic curricula for introducing GenAI.

The TPACK Lens

The review applies the TPACK framework to identify what knowledge teachers need to integrate GenAI effectively into student learning. It synthesizes GenAI-related competencies across the framework's domains — technological knowledge (operating GenAI, examining its benefits/limitations and ethical implications), pedagogical knowledge (scaffolding student use), and content knowledge (integrating AI within subject-matter instruction). This positions GenAI-TPACK professional development as a core requirement for realizing AI's benefits while mitigating misuse risk, connecting to the knowledge base's Teacher AI Competency and Educational Development research.

What this means for practice

• Faculty developers. Build GenAI-TPACK rather than delivering isolated tool training: teachers need integrated technological, pedagogical and content knowledge to judge when and how GenAI serves a subject-specific learning goal.
• Instructors. Pair GenAI access with structured pedagogical Scaffolding, because scaffolding of GenAI use was frequently incomplete across the 71 reviewed studies and the tool alone did not raise behavioral engagement (SMD = 0.057, p = 0.828).
• Instructors. Teach students to evaluate and use GenAI responsibly as a distinct curriculum goal — the review's first critical priority — instead of assuming literacy follows from exposure.
• Learners. Use GenAI as a domain-specific learning tool with deliberate study strategies; the cognitive payoff is real (g = 0.831) but it comes from how the tool is used, not from access.
• Administrators. Extend GenAI integration beyond Higher Education and fund systematic AI curricula for K-12, where adoption and pedagogical strategy for higher-order thinking remain underdeveloped and quasi-experimental designs still dominate.

Limitations

• The review's scope is deliberately narrow: only empirical studies using GenAI with student participants and reporting learning outcomes were included, all published after ChatGPT's release on November 30, 2022, so teacher-facing studies, reviews and theoretical work were excluded by design and the evidence base is young.
• The meta-analysis pools 71 papers (47 journal articles, 74 effect sizes) that are dominated by quasi-experimental designs of uneven rigor; the authors' own quality appraisal flagged recurring problems such as short duration, small sample sizes and non-disclosure of research limitations.
• The behavioral-engagement finding is a null result (SMD = 0.057, p = 0.828) and is much weaker evidence than the cognitive (g = 0.831) and affective (g = 0.729) effects, so the overall g = 0.752 should not be read as a uniform treatment effect.
• Instructional-design variables (learning content, tool, strategy, assessment, outcome) significantly moderated outcomes, so pooled effect sizes conceal substantial variation between interventions.

Citation

Liu, X., & Zhong, B. (2025). Integrating generative Artificial Intelligence into student learning: A systematic review from a TPACK perspective. Educational Research Review, 49, 100741.