Factors Associated with Students' Adoption of Artificial Intelligence Technology in Tertiary Education: A Meta-Analytic Review

Synthesis: In brief: Teo, Scherer, Fung, and Fung meta-analyze 233 correlations from 32 studies (N = 16,977) to synthesize the factors shaping post-secondary students' adoption of AI. A three-level model yields moderate positive correlations for individual (r = 0.57), contextual (r = 0.53), and technological (r = 0.50) factors — no single category dominates — with usage intentions the strongest specific predictor (r = 0.64). The review argues the field over-relies on traditional technology acceptance models (TAM, UTAUT) and neglects AI-specific factors such as anthropomorphism and Ethics.

This meta-analysis addresses the scattered evidence on why tertiary students adopt AI for learning. Synthesizing 233 correlations from 32 studies across 16,977 participants, the authors group predictors into individual, contextual, and technological factors and use a three-level correlated-effects model. The result is a balanced picture: individual (r = 0.57), contextual (r = 0.53), and technological (r = 0.50) factors all matter moderately, with usage intentions the single strongest predictor (r = 0.64). Notably, perceived risks and Trust showed weaker associations than expected. Large heterogeneity (I² > 98%) was partly explained by construct diversity and students' prior experience with AI.

The review's theoretical contribution is its critique of the field's reliance on traditional technology acceptance models (TAM, UTAUT) that predate modern intelligent systems. It argues these frameworks miss AI-specific factors — such as anthropomorphism, ethics, and the adaptive/human-like nature of modern AI — and that addressing this gap matters for advancing theory and evidence-based policy. For practice, it recommends targeting specific constructs (learning objectives, intentions) and adapting integration to students' experience levels.

Key Findings

• 233 correlations from 32 studies (N = 16,977) synthesized via a three-level correlated-effects meta-analysis.
• Moderate positive correlations for individual (r = 0.57), contextual (r = 0.53), and technological (r = 0.50) factors — no single category dominates.
• Usage intentions are the strongest predictor (r = 0.64); perceived risks and trust showed weaker associations.
• Large heterogeneity (I² > 98%), partly explained by construct diversity and students' AI experience.
• The field over-relies on TAM/UTAUT and neglects AI-specific factors (anthropomorphism, ethics).
• Practice implication: target specific constructs and adapt integration to students' experience levels.

What this means for practice

• Instructors. Target specific, measurable constructs rather than generic "AI acceptance": usage intentions were the strongest predictor of adoption (r = 0.64), so design integration around students' intent to use a tool for learning.
• Instructors. Calibrate integration to students' prior AI experience, which moderated the association with individual factors — samples with at least some AI experience showed higher correlations than those with little, none, or unstated experience.
• Administrators. Do not chase a single lever: individual (r = 0.57), contextual (r = 0.53), and technological (r = 0.50) factors all matter moderately with no category dominating, and the authors tie closing the theory gap to evidence-based policy.
• Researchers. Move past TAM and UTAUT: the review argues these frameworks predate modern intelligent systems and leave AI-specific factors — anthropomorphism, perceived intelligence, and ethical considerations — largely unexamined, so new studies should draw on AI-specific frameworks such as AIDUA.
• Instructors. Do not assume trust is the binding constraint: perceived risk and Trust showed weaker associations than expected, which the authors attribute to measurement approaches that fail to capture the ethical dimensions of AI.

Limitations

• Heterogeneity was extreme (I² > 98%) and the authors report likely publication bias, so they caution that the weighted average effect for each construct category must be interpreted carefully.
• The statistically significant advantage of individual over technological factors was small in magnitude; the authors attribute it to high statistical power rather than a meaningful practical distinction.
• The evidence base is dominated by quantitative self-report surveys and cross-sectional designs — only one included study was longitudinal — so it cannot trace how students' perceptions and adoption behaviors evolve over time.
• Primary studies concentrated on large language model chatbots, making the synthesis most representative of conversational AI adoption, and their limited granularity prevented analysis of how adoption varies with specific technological functions; average study size was 530.5 participants but ranged from 64 to 2,240 (SD = 492.9, Mdn = 376), and only construct type and students' AI experience explained heterogeneity.

Citation

Teo, T., Scherer, R., Fung, A. S. K., & Fung, C. S. L. (2026). Factors associated with students' adoption of artificial intelligence technology in tertiary education: A meta-analytic review. Educational Research Review, 52, 100804.