A systematic review of student engagement research in adaptive learning platforms

Synthesis: Simon, Zeng, and Fryer (2026) conducted a PRISMA-aligned systematic review of the factors that shape Student Engagement in Adaptive Learning platforms and of which engagement types dominate the literature. Searching six databases for 2012–2022 and retaining 44 peer-reviewed studies, they applied a hybrid inductive/deductive thematic analysis organized around a four-component engagement framework (behavioral, affective, cognitive, agentic). Four overarching themes emerged — system quality, information quality, learner characteristics and preferences, and affordances that mimic real-life social interactions. Behavioral engagement was the most discussed type, agentic engagement the least, prompting the authors to call for an extended engagement framework for human–computer interaction. All four themes were found to interact, and most factors ultimately trace back to system design.

Core Finding

Student engagement in adaptive learning platforms is driven by four interacting theme clusters — system quality, information quality, learner characteristics and preferences, and social-interaction affordances — and these factors are entangled: content quality, learner agency, and social support all depend on how the platform is designed. Across the literature, behavioral engagement is the most studied type and agentic engagement the least studied, a gap the authors attribute to adaptive platforms reducing the student–teacher interactions through which agentic engagement is normally expressed.

The Four Engagement Themes

Drawing partly on the Information Systems Success Model and partly on inductive coding, the review organizes the drivers of engagement into four clusters.

• System quality — how well the platform works: adaptivity (tailoring activities to the learner's needs and emotional state), Accessibility and "anytime, anywhere" navigation, gamification, structure and design, progress tracking, and novelty. Adaptive tailoring and the exploratory novelty of environments like iTalk2Learn kept primary students on task.
• Information quality — the value of the content: conciseness and relevance, challenging tasks with progressive difficulty, diversity of formats (texts, visuals, animations, video), and interactive practice questions that must be answered until correct, delivering immediate feedback.
• Learner characteristics and preferences — self-directed learning at one's own pace, plus socio-demographic variables (educational aspiration, parental education/SES, gender, age/year level) and learning emotions.
• Affordances mimicking real-life social interactions — affective learning companions, adaptive peer robots that switch between tutee and tutor roles, regular/immediate elaborated feedback, positive reinforcement, and a "teammate effect" (as in Wayang Outpost) that increases Help-Seeking.

Engagement Types: Behavioral, Affective, Cognitive, Agentic

The review classified studies using the extended framework of Fredricks et al. (behavioral, affective, cognitive) plus Reeve and Tseng's agentic dimension. Rater counts showed behavioral engagement discussed most (≈29–35 articles), followed by affective (25–27), cognitive (17–19), and agentic (3–6). The relative omission of agentic engagement is telling: agency in Reeve and Tseng's conception is typically expressed in interaction with a teacher, which adaptive platforms minimize. The authors argue students should be given ways to express preferences, customize their environment (as ALEKS allows), and use help buttons to support agentic engagement, and they call for an engagement framework revised for human–computer interaction.

Disengagement and the Novelty Effect

The review also surfaces the "negative side" of adaptive learning. Disengagement arose from complicated, frustrating layouts, browser/technical issues, overly long or irrelevant modules, lack of detailed feedback, and — notably — a novelty effect: engagement declined over time in longitudinal studies of ALEKS and W-Pal once the novelty of the tool faded. Sustaining motivation after the exploration period is flagged as a key open problem.

Self-Determination and Design Implications

The authors interpret the social-interaction findings through self-determination theory: positive reinforcement, teammate effects, and learning companions affirm competence, convey relatedness, and support autonomy. Practically, schools should invest in platforms matched to their students' needs, involve experienced teachers in content creation, orient students to platform features, balance structure and flexibility, and integrate adaptive peer robots/companions to reduce the solitude of online learning. Well-designed feedback systems can also unburden teachers in blended settings so they can focus on quality student–teacher relationships.

What this means for practice

• Instructors. Audit platform usability before you commit to it. Complicated layouts, browser and operating-system incompatibilities, missing automatic feedback, and hard penalties for minor mistakes were among the disengagement triggers students reported most often.
• Instructors. Orient students to platform features at the start of a course and give them ways to act on the system — preference settings, customization, and a visible help button — because agentic engagement was the least-studied and least-supported type in this literature.
• Instructors. Plan for the novelty effect. Engagement declined over time in longitudinal studies of ALEKS and W-Pal, so build in refreshers, new task types, and check-ins rather than assuming initial motivation holds.
• Designers. Design the practice loop, not just the content: interactive questions answered until correct, immediate elaborated Feedback, and progressive difficulty were the information-quality features linked to engagement, and most factors that look like learner characteristics trace back to system design.
• Designers. Build social-interaction affordances — adaptive peer robots or learning companions — into fully online and large-enrollment courses to reduce the solitude of online learning, and check for gender mismatch between student and companion.

Limitations

• The review retained 44 studies drawn from six databases for 2012–2022; gray (non-peer-reviewed) literature and non-English manuscripts were deliberately excluded, which the authors acknowledge may have removed national literatures published in other languages.
• The database search used only "adaptive learning" and "student engagement" as keywords, so studies using terms such as artificial intelligence or intelligent tutoring systems instead may have been missed.
• Engagement measurement was uneven across the evidence base: of the 44 studies appraised, only 10 reported both reliability and validity data for their instruments and 17 reported no psychometric information at all, while just eight were randomized controlled trials (12 non-randomized, eight descriptive, 15 mixed-method, one qualitative).
• Effect sizes were out of scope, so the review names factors without giving any of them relative weight; the authors recommend meta-analysis to establish relative impact and experience sampling methods (ESM) for more ecologically valid engagement measurement.

Citation

Simon, P. D., Zeng, L. M., & Fryer, L. K. (2026). A systematic review of student engagement research in adaptive learning platforms. Computers and Education Open, 10, 100360.