Cross-sectional survey links agentic AI learning perceptions with engagement and perceived performance

A cross-sectional, perception-based survey of 398 business school students in India found that perceived autonomy, competence and relatedness were positively associated with learning motivation in agentic AI-supported online learning, while perceived interactivity, information sharing and collaborative environment were associated with social presence. Learning motivation and social presence, in turn, were associated with learner engagement, which had the strongest modeled relationship with perceived learner performance. These findings describe statistical associations among students’ perceptions and do not establish causal effects or improvements in objective academic achievement.

The researchers used non-probability purposive sampling to recruit students enrolled in business or management programs at several business schools across India. The questionnaire was distributed through institutional channels from October to December 2025 and was restricted to students who had used AI-powered platforms for educational purposes during the previous three months. Of 640 invitations, 427 questionnaires were returned; 29 with missing data were excluded, leaving 398 valid responses, or 62.19% of those invited. The sample included MBA students (49%), PGDM students (39.4%) and BBA students (11.6%).

Participants reported using tools such as ChatGPT Agent, Claude with agentic capabilities and Grammarly AI Agents for academic work. The study did not evaluate one particular system or independently verify the tools’ technical architecture. Instead, it operationalized agentic AI through learners’ perceptions of adaptive guidance, task planning, iterative feedback and related capabilities. A seven-point Likert questionnaire measured perceived autonomy, competence, relatedness, learning motivation, interactivity, information sharing, collaborative environment, social presence, learner engagement and perceived learner performance.

The conceptual model combined three theoretical perspectives. Holmberg’s Guided Didactic Conversation Theory supplied the interactional and educational lens for dialogue, guidance, feedback and responsiveness. Self-Determination Theory, described in the study as a popular and comprehensive framework for understanding psychological needs, motivation and well-being, accounted for autonomy, competence and relatedness. Social Presence Theory explained perceived connection and support through interactivity, information exchange and collaboration. The proposed pathways joined these constructs through engagement and perceived performance.

Analysis using Partial Least Squares Structural Equation Modelling found that perceived autonomy had the strongest modeled association with learning motivation (β = 0.504, 95% CI 0.442–0.560), followed by perceived competence (β = 0.425, 95% CI 0.358–0.491) and perceived relatedness (β = 0.189, 95% CI 0.109–0.272). Interactivity had the strongest association with perceived social presence (β = 0.468, 95% CI 0.404–0.528), followed by information sharing (β = 0.323, 95% CI 0.248–0.398) and collaborative environment (β = 0.184, 95% CI 0.112–0.265). All nine tested structural relationships were positive and statistically significant at p < 0.001, with bootstrapped confidence intervals excluding zero.

Both motivational and social-relational measures were strongly associated with engagement. Learning motivation had a path coefficient of 0.533 with a 95% confidence interval of 0.467–0.598, while perceived social presence had a coefficient of 0.493 with a 95% confidence interval of 0.427–0.558. Engagement was then associated with perceived learner performance at β = 0.671 (95% CI 0.615–0.725), the strongest direct relationship in the model. The adjusted R² values reported for engagement and perceived learner performance were 0.553 and 0.449, respectively.

The measurement model showed convergent and discriminant validity: Cronbach’s alpha and composite reliability values exceeded 0.70 for all constructs, average variance extracted values exceeded 0.50, and all heterotrait-monotrait ratios were below 0.85. Variance inflation factors ranged from 1.000 to 1.026, below the cited multicollinearity threshold of 3.3. The model reported an SRMR of 0.044 and an NFI of 0.959, which the authors described as indicators of approximate fit rather than exact fit. The analysis used 5,000 bootstrap subsamples and also examined indirect and serial mediation pathways.

Participants’ reported use of AI was extensive. Most respondents used AI-powered platforms daily (85.9%), and 91.7% reported more than a year of use. All respondents used AI tools for assignments; 98.2% used them for problem-solving, 87.2% for research and 79.9% for case-study analysis. The sample was also nearly balanced by gender, with 211 male respondents (53%) and 187 female respondents (47%).

Several limitations constrain the findings. The cross-sectional design captured perceptions at one point and cannot demonstrate causality or long-term educational impact. All constructs were self-reported, and perceived learner performance was not replaced by exam results, assignment scores or learning analytics. The single-source survey could retain common-method bias despite several checks. Purposive sampling of Indian business students limits generalizability, while unexamined variables—including digital literacy, AI competence, prior online-learning experience, voluntary adoption, institutional affiliation and the specific systems used—may also be associated with the modeled relationships. The authors call for longitudinal, experimental, multi-institutional and probability-based research, as well as studies of technically verified agentic systems.

Subject of Research: Perceived motivational and social pathways linking agentic AI-supported learning with engagement and perceived performance among Indian business school students

Article Title: Agentic AI in educational environments and its association with social and motivational pathways to learning performance

Article References: Pramod, D., & Patil, K. P. (2026). Agentic AI in educational environments and its association with social and motivational pathways to learning performance. Discover Education, 5(1), Article 929. https://doi.org/10.1007/s44217-026-02168-4

Image Credits: AI Generated

DOI: 10.1007/s44217-026-02168-4

Keywords: agentic AI, online learning, higher education, learner engagement, learning motivation, social presence, perceived performance, cross-sectional survey, business school students, Agentic, educational, environments