Decoding the Student-AI Dialogue

by

Author: Denis Avetisyan

New research reveals how students engage with AI feedback within learning analytics dashboards, and how those interactions vary based on individual learning styles.

A dialogue analysis of student questioning patterns with a generative AI assistant in a learning analytics feedback system highlights the need for more personalized and context-aware educational technology.

Despite the increasing prevalence of learning analytics dashboards, students-particularly those with weaker self-regulation skills-often struggle to interpret and act upon the feedback provided. This research, ‘What Students Ask, How a Generative AI Assistant Responds: Exploring Higher Education Students’ Dialogues on Learning Analytics Feedback’, investigates authentic interactions between students and a generative AI assistant integrated into such a dashboard, revealing distinct questioning patterns linked to students’ levels of self-regulated learning. Findings demonstrate that while low-SRL students sought clarification and reassurance, those with higher SRL skills probed technical aspects and requested personalized strategies-though the assistant’s ability to provide truly tailored responses remained limited. How can future AI-driven feedback systems become more adaptive, context-aware, and trustworthy to effectively support all learners?

Unveiling Individual Learning Profiles

Historically, educational systems have largely operated on a “one-size-fits-all” model, delivering instruction at a pace and in a manner designed for the average student. However, cognitive science reveals significant variation in how individuals learn – differences in prior knowledge, learning styles, motivational levels, and cognitive capacities. This inherent diversity often leaves students either struggling to keep up or feeling unchallenged, leading to disengagement and, ultimately, suboptimal learning outcomes. When instruction doesn’t resonate with a student’s unique profile, it can erode their intrinsic motivation, foster feelings of inadequacy, and hinder their ability to reach their full potential. The consequences extend beyond academic performance, impacting self-esteem and future learning trajectories, highlighting the critical need for approaches that acknowledge and address individual student needs.

The burgeoning field of learning analytics presents a compelling opportunity to move beyond standardized educational models, yet the sheer volume of data generated by student interactions is insufficient on its own. True personalization requires sophisticated algorithms and interpretive frameworks capable of transforming raw data – encompassing everything from time spent on tasks to error patterns and resource utilization – into actionable insights. These insights must then be seamlessly integrated into adaptive learning systems, providing educators and students with targeted recommendations, customized content, and precisely timed interventions. Without this critical translation step, learning analytics remains a descriptive tool rather than a transformative one, failing to unlock the potential for truly individualized educational experiences and optimized learning outcomes.

Individuals approach learning with markedly different levels of self-regulation, impacting their ability to plan, monitor, and evaluate their own progress. Some students excel at proactively identifying knowledge gaps and seeking appropriate resources, while others struggle with these metacognitive processes, requiring more explicit guidance. Consequently, a one-size-fits-all educational approach often proves ineffective; instead, adaptable support systems are crucial. These systems must dynamically assess a student’s current SRL competence-determining whether they benefit from detailed step-by-step instructions or more open-ended prompts-and adjust the level of scaffolding accordingly. Recognizing this inherent variability is fundamental to fostering genuine learning potential, as interventions tailored to an individual’s self-regulatory skills demonstrably improve academic outcomes and cultivate lifelong learning habits.

Truly effective personalized learning transcends simply delivering tailored content; it necessitates a deep understanding of a student’s unique information-seeking behaviors and feedback interpretation processes. Research indicates students don’t uniformly approach learning – some actively explore, while others prefer direct instruction, and their responsiveness to corrective feedback varies considerably. Consequently, static personalized systems fall short. Instead, a dynamic, conversational support system is needed-one that observes how a student navigates information, identifies knowledge gaps through interaction, and adjusts its guidance accordingly. This approach allows the system to not only present relevant material but also to model effective learning strategies, offering nuanced feedback and scaffolding that aligns with the individual’s cognitive style and promotes self-directed learning. Ultimately, such systems aim to cultivate not just knowledge acquisition, but also the metacognitive skills essential for lifelong learning.

A Conversational Bridge for Adaptive Support

The Generative AI Assistant functions as an interface between Learning Analytics Dashboards and individual students, processing quantitative data – such as assignment scores, time spent on tasks, and participation rates – to formulate personalized conversational prompts. This system moves beyond the passive presentation of learning data by actively initiating dialogues with students, aiming to facilitate self-regulated learning. The assistant’s interpretation of dashboard data is not simply a restatement of metrics, but rather a translation into accessible language designed to prompt reflection, identify areas for improvement, and encourage proactive learning strategies. Data from multiple sources within the Learning Analytics Dashboard are synthesized to provide a holistic view of student performance and engagement, informing the content and direction of these conversations.

The system employs dialogue-based feedback as an alternative to traditional, static learning analytics reports. Rather than presenting data points, the Generative AI Assistant engages students in conversational interactions designed to facilitate understanding of their learning progress. This approach moves beyond simply telling a student their performance; instead, the AI co-constructs knowledge with the student through questioning, clarification, and personalized guidance. This iterative process allows the AI to address individual misconceptions, promote self-reflection, and collaboratively build a shared understanding of the student’s strengths and areas for improvement, ultimately supporting their self-regulated learning (SRL) process.

Context awareness within the Generative AI Assistant is achieved through the integration of student-specific data points, including performance metrics from Learning Analytics Dashboards, historical interaction logs with the system, and identified learning goals. This data informs the AI’s response generation, allowing it to move beyond generalized feedback and address the student’s unique needs and progress. Specifically, the system analyzes patterns in a student’s activity – such as frequently accessed resources, common errors, and time spent on tasks – to understand their current learning situation. This understanding is then used to tailor the language, complexity, and content of the AI’s responses, ensuring relevance and maximizing the potential for constructive dialogue.

Prompt engineering is the process of designing and refining textual inputs, known as prompts, to elicit desired responses from the Generative AI Assistant. Effective prompts are not simply questions; they include specific instructions regarding the desired response format, length, and tone, as well as contextual information drawn from the student’s learning analytics data. These engineered prompts guide the AI to deliver feedback that is directly relevant to the student’s performance, offers actionable steps for improvement, and supports the student’s self-regulated learning (SRL) process by encouraging reflection and goal-setting. The quality of the prompt directly impacts the AI’s ability to provide helpful and personalized guidance, making iterative prompt refinement a critical component of system performance.

Mapping Cognitive Processes: A Mixed-Methods Investigation

Epistemic Network Analysis (ENA) was employed to characterize student information-seeking behavior by mapping the relationships between three distinct query types: Technical Queries, which request specific procedural assistance; Clarification Queries, indicating a need for explanation of concepts; and Personalized Insight Queries, reflecting a desire for tailored guidance or advanced analytics. This network-based approach allowed for the visualization of how students navigate information needs, revealing patterns of inquiry and dependencies between query types. By analyzing the frequency and co-occurrence of these queries within individual student interactions, we identified characteristic information-seeking profiles and potential areas where students struggle to effectively locate or process information. The resulting network maps displayed the connections between these query types, offering a quantitative representation of student cognitive processes during learning.

Conversation Analysis of student interactions with the GenAI assistant revealed several key themes in feedback interpretation. Students frequently required iterative clarification of responses, particularly when dealing with complex analytical outputs. Challenges emerged around discerning the level of detail provided; some students indicated a need for more concise summaries, while others requested expanded explanations of underlying data. Furthermore, analysis identified instances where students misinterpreted the assistant’s suggestions as directives, highlighting the importance of framing AI-generated feedback as recommendations rather than prescriptive instructions. This qualitative data emphasized the need for the AI to adapt its communication style based on individual student preferences and demonstrated learning styles.

Analysis of student interactions revealed a correlation between self-regulated learning (SRL) competence and the types of queries submitted to the AI assistant. Students identified as having lower SRL competence frequently posed questions requiring clarification of basic concepts and expressed a need for reassurance regarding their understanding. Conversely, students with high SRL competence primarily submitted queries requesting advanced analytics, strategic guidance for problem-solving, and data-driven insights into their learning progress. Epistemic Network Analysis (ENA) quantified this difference, demonstrating a 12.0% variance in discourse networks between the two groups, indicating distinct patterns in how they approached information seeking and utilized the AI assistant’s capabilities.

Analysis of student interactions with the GenAI assistant revealed that responding to expressed emotional states was significantly correlated with continued engagement. Of the 34 students enrolled in the study, 22 utilized the assistant, and those who engaged with the assistant’s emotional response capabilities demonstrated a higher frequency of follow-up queries and a longer average session duration. This suggests the AI’s capacity to acknowledge and address student affect is not merely a rapport-building feature, but a functional component in sustaining student motivation and promoting deeper interaction with the learning platform.

Cultivating Adaptive Support: Impact and Future Trajectory

The AI assistant distinguishes itself through a conversational approach to feedback, moving beyond simple error identification to engage students in a dialogue aimed at practical improvement. This system doesn’t merely tell a student what is wrong; instead, it prompts them to articulate their understanding, identify gaps in knowledge, and collaboratively construct solutions. By framing feedback as a series of guided questions and prompts, the assistant encourages active participation and fosters a deeper comprehension of the underlying concepts. Consequently, students are better equipped to translate the feedback into concrete actions, leading to more effective learning and demonstrable progress – a shift from passively receiving criticism to actively implementing strategies for growth.

The implementation of this AI-driven learning system cultivates a uniquely supportive environment, proving particularly beneficial for students who find self-regulation challenging. These learners often struggle with initiating tasks, staying focused, and managing their learning process; however, the system’s consistent feedback and guidance offer external scaffolding that promotes greater agency. This, in turn, demonstrably increases student engagement and motivation, as the AI assistant helps break down complex tasks into manageable steps and provides encouragement tailored to individual progress. The result is a learning experience where students feel more empowered and less overwhelmed, fostering a positive feedback loop that reinforces proactive learning behaviors and sustained effort.

The ongoing development of this AI assistant benefits from a robust data feedback loop, combining measurable metrics with direct student input. Quantitative analysis reveals overall performance, while qualitative feedback – in this case, student ratings – pinpoints specific strengths and areas for improvement. Notably, students consistently scored the assistant’s clarity highly, averaging 4.63 on a 5-point scale – the most positively received aspect of the system. This detailed understanding, achieved through the interplay of numbers and nuanced opinions, isn’t simply descriptive; it actively informs iterative refinements to both the assistant’s responses and its personalization strategies, ensuring continuous enhancement of the learning experience.

Efforts are now directed towards expanding the reach of this AI-assisted learning system by integrating it with existing learning management systems, envisioning a future where personalized education is broadly accessible. Current evaluations reveal strong student perception of the system’s clarity, yet highlight a need for enhanced personalization strategies; while students rated clarity at 4.63 out of 5, personalization received a score of 3.56. Consequently, ongoing development prioritizes refining algorithms to better tailor feedback and learning pathways to individual student needs, moving beyond simply understandable guidance towards a truly adaptive educational experience that dynamically responds to each learner’s unique progress and challenges.

The study illuminates how students’ approaches to learning analytics feedback are deeply connected to their existing self-regulated learning skills. It reveals a spectrum of inquiry, from those seeking clarification of data to those proactively questioning underlying assumptions. This echoes David Hilbert’s assertion: “We must be able to answer the question: What are the prerequisites for the possibility of mathematical thinking at all?” Just as Hilbert sought foundational principles for mathematics, this research probes the prerequisites for effective engagement with learning analytics. Understanding these foundational differences in questioning-the invisible boundaries of a student’s approach-is crucial for designing AI feedback systems that truly support personalized learning pathways and anticipate potential weaknesses before they manifest as frustration or disengagement.

Where the Conversation Leads

The study reveals a predictable truth: students do not simply want data; they seek justification, explanation, and, crucially, a conversational partner who understands the shape of their questions. The observed differences in questioning patterns, tied to levels of self-regulated learning, suggest that a single, ‘smart’ AI assistant is a chimera. If the system survives on duct tape, it’s probably overengineered. The real challenge lies not in predicting what a student will ask, but in recognizing why – and that requires a model of the learner that extends beyond surface-level engagement metrics.

Current learning analytics dashboards, and their increasingly sophisticated AI companions, often mistake access to information for genuine understanding. Modularity without context is an illusion of control. Future work must move beyond presenting data about learning and focus on facilitating a dialogue within learning-one that acknowledges the messy, iterative, and deeply personal nature of knowledge construction.

The ultimate test will not be whether the AI can answer the questions, but whether it can ask the right questions in return, prompting students to refine their own understanding and, ultimately, become more effective architects of their own learning journeys. Only then will the conversation truly begin.

Original article: https://arxiv.org/pdf/2601.04919.pdf

Contact the author: https://www.linkedin.com/in/avetisyan/

See also:

2026-01-11 21:15