How Does Students' Perception of ChatGPT Shape Online Learning Engagement and Performance?

Synthesis: This SEM-based study tested whether students' perception of ChatGPT (knowledge, willingness to use, and concerns) shapes online learning engagement across behavioral, emotional, and cognitive dimensions, and whether engagement in turn drives academic performance. Using survey data from 305 graduate students in Iran, the authors found that ChatGPT perception accounts for 19.2% of variance in engagement, while perception plus engagement together explain 40.4% of variance in academic performance. Online learning engagement significantly mediates the perception→performance link, underscoring its pivotal role in translating AI adoption into academic gains.

Key Findings

• Sample: Self-reported survey data from 305 graduate students (master's and doctoral) at Shiraz University, Iran (2024–2025 academic year), drawn from a population of ~7,382 and exceeding the 251 required by power analysis.
• Model fit: PLS-SEM showed students' perception of ChatGPT explains 19.2% of the variance in online learning engagement; combining ChatGPT perception and online engagement explains 40.4% of variance in academic performance.
• Direct paths: ChatGPT perception had significant direct effects on online learning engagement (β = 0.438) and on academic performance (β = 0.268).
• Mediation: Online learning engagement significantly mediated the perception–performance relationship, with a significant indirect effect of β = 0.206 — confirming engagement as a crucial mediator (Hypothesis 4 supported).
• Key correlations: Willingness to use generative AI correlated with academic performance (r = 0.412, p < .01); cognitive learning engagement correlated most strongly with performance (r = 0.563, p < .01). Importance–Performance Map Analysis (IPMA) ranked willingness to use ChatGPT as the most important driver of academic performance, followed by cognitive and behavioral engagement.

Study Design & Method

• Design: Cross-sectional, self-report questionnaire using a five-point Likert scale (1 = strongly disagree to 5 = strongly agree); stratified sampling across master's and PhD levels for representativeness.
• Theoretical grounding: Integrated the Technology Acceptance Model (TAM; Davis, 1989) and Bandura's Social Cognitive Theory (SCT, 1986) to frame how personal, behavioral, and environmental factors shape ChatGPT adoption and engagement.
• Perception of ChatGPT (adapted from Chan & Hu, 2023): three subscales — Knowledge of Generative AI Technology (KGAT, 6 items), Willingness to Use Generative AI Technologies (WUGAT, 8 items), and Concerns About Generative AI Technologies (CGAT, 4 items); Cronbach's α = 0.711.
• Online learning engagement (adapted from Kuo et al., 2021): 14 items across behavioral (3), emotional (5), and cognitive (6) engagement dimensions; Cronbach's α = 0.921.
• Academic performance (adapted from Yu et al., 2010): four items; Cronbach's α = 0.889.
• Analysis: Structural Equation Modeling (SEM) via Partial Least Squares (PLS), with Confirmatory Factor Analysis (CFA), Composite Reliability, and Average Variance Extracted (AVE) to validate the measurement model; bootstrapping used for mediation analysis. Reliability (α) ranged 0.711–0.927 and CR 0.661–0.947 across constructs; skewness/kurtosis within acceptable normality thresholds.
• Hypotheses: H1 perception→engagement; H2 engagement→performance; H3 perception→performance; H4 engagement mediates perception→performance. All were supported.

What this means for practice

• Instructors. Optimize the engagement channel, not the tool: online learning engagement carried a significant indirect effect (β = 0.206) between ChatGPT perception and academic performance, so activities need to raise behavioral, emotional, and cognitive participation rather than distribute answers.
• Instructors. Target adoption attitudes before awareness: willingness to use generative AI was the most important driver of performance in the importance–performance map and correlated with it at r = 0.412, which points to perceived usefulness and ease of use as the levers worth designing for.
• Instructors. Watch the dimensions the study ranks highest — cognitive engagement correlated most strongly with performance (r = 0.563) — and intervene where online participation thins out, since engagement is the mediator that converts perception into achievement.
• Administrators. Treat students' concerns as part of the model rather than as complaints to manage: worry about privacy, plagiarism detection, and lost transferable skills loads the perception construct, and the authors flag third-person effects on the equity gap in AI access and digital literacy.
• Administrators. Commission ChatGPT activities that require interaction and engagement by design rather than as answer shortcuts, the design the authors present as the counter to overreliance and reduced critical thinking.

Limitations

• Cross-sectional, self-report survey of 305 graduate students (master's and doctoral) at a single Iranian university, drawn from a population of approximately 7,382 and exceeding the 251 required by power analysis; the PLS-SEM path model is correlational and cannot show that perception or engagement causes performance.
• Perception, engagement, and performance were all measured with five-point Likert self-reports, which the authors note are open to bias related to students' perceptions and behaviors; they call for experimental designs in live settings with behavioral measures such as time-on-task.
• The study measured none of the access conditions the authors flag as consequential — quality of internet access, use of VPNs to reach ChatGPT, and ChatGPT availability — all of which can shape perceptions, interactions, and outcomes in online learning.
• Participants came from one institution and one national context, which the authors say limits applicability where perceptions of ChatGPT and its application differ across universities and cultures.

Citation

Mehrvarz, M., Salimi, G., Abdoli, S., & McLaren, B.M. (2025). How does students' perception of ChatGPT shape online learning engagement and performance?. Computers and Education: Artificial Intelligence, 9, 100459.