Behavioural engagement patterns in an AI-powered student support chatbot

An exploratory learning analytics study

Bibliographic Data

ID22434802
AuthorsNondumiso Shabangu (Mangosuthu University of Technology), Phiwayinkosi Richmond Gumede (0000-0002-9203-8457, Mangosuthu University of Technology), Khulekani Yakobi (0000-0001-7568-5065, Durban University of Technology)
Year2026
Volume11
Publication date2026-07-17
Peer ReviewedYes
Open AccessYes
TypeARTICLE
VenueFrontiers in Education (JOURNAL)
Journal identifiersISSN: 2504-284X • E-ISSN: 2504-284X
PublisherFrontiers Media SA (PUBLISHER • CH)
DOI10.3389/feduc.2026.1851891
OpenAlexW7169559214
LanguageEN
References cited44

Introduction The increasing adoption of AI-powered conversational systems in higher education has created new opportunities for scalable student support and digitally mediated engagement. However, limited research has examined how students behaviourally engage with these systems using real-world interaction data, particularly within resource-constrained institutional contexts. This study investigates patterns of student engagement within an institutional chatbot-supported student support environment using an exploratory learning analytics approach. Drawing on the Technology Acceptance Model (TAM) as an interpretive framework, it conceptualises engagement sustainability as a behavioural lens for understanding continued interaction beyond initial system use. Methods An exploratory descriptive learning analytics design was employed to analyse 1,495 chatbot interaction records collected from a hybrid AI-assisted student support system deployed at a South African University of Technology between July and November 2025. Behavioural interaction data were analysed using Microsoft Power BI to identify engagement patterns, interaction persistence, support-query trends, and system-classified resolution outcomes. Results The findings indicate that chatbot-supported student engagement occurs within a hybrid interaction environment combining automated conversational processing with human-assisted institutional support. Student interaction patterns were highly uneven, with a small proportion of users accounting for a disproportionately large share of system activity, while many users disengaged after initial interactions. Chatbot usage was concentrated primarily within administrative support domains, and system-classified completion rates were high for structured queries. However, extended response times associated with escalated interactions reflected institutional workflow processes rather than chatbot processing performance. Discussion This study demonstrates the value of behavioural interaction log analysis for understanding AI-mediated student engagement in higher education. By interpreting engagement sustainability as an observable behavioural phenomenon, it extends existing applications of the Technology Acceptance Model beyond adoption intentions to actual usage behaviour. The findings further highlight the importance of considering interaction patterns, institutional context, and hybrid human–AI support models when evaluating the effectiveness and sustainability of chatbot-supported student engagement in higher education.

Analytics · Chatbot · Dashboard · Descriptive statistics · Exploratory research · Learning analytics · Student engagement · Workflow · AI in Service Interactions · Artificial Intelligence in Healthcare and Education · E-Learning and COVID-19