Scaffolding Metacognition with GenAI: Exploring Design Opportunities to Support Task Management for University Students with ADHD
Zihao Zhu, Junnan Yu, Yuhan Luo · 2026 · Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI '26) · doi:10.1145/3772318.3790697
Summary
This paper applies a metacognitive lens to academic task management for university students with ADHD, asking how Generative AI (GenAI) tools like ChatGPT, Gemini, and Claude could scaffold the awareness and regulation of one's own thinking processes that this population struggles with most. Rather than instructing participants to design 'for metacognition' (which would have biased ideas toward researcher language), the authors ran 70-100 minute one-on-one co-design sessions with 20 university students diagnosed with ADHD (11 female, 9 male; ages 19-24; majoring across humanities, sciences, engineering, and the arts; 17 in mainland China and three abroad), inviting them to sketch GenAI-powered task management ideas on the Excalidraw whiteboard in response to their own lived challenges. After analysis, the authors interviewed five ADHD experts (three intervention specialists and two coaches) to assess feasibility, clinical relevance, and potential pitfalls of the resulting ideas. Bottom-up thematic analysis of the co-design transcripts produced 764 initial codes, the expert interviews added 245 more, and the team consolidated these into four metacognitive challenge themes (lack of awareness, barriers to initiation, attention control difficulty, struggles to regulate emotions) and three GenAI design directions (cognitive scaffolding, reflective task execution, emotional regulation).
Key findings
Participants surfaced metacognitive breakdowns at every stage of the task cycle. In organization, fragmented information across syllabi, emails, and chat apps overwhelmed working memory; many set unrealistic 'do everything at once' to-do lists, conflated task importance with personal interest, and could not estimate task duration. In execution, perfectionism delayed initiation, distraction and hyperfocus disrupted attention, and progress was hard to gauge without external visualization. In adjustment, sudden schedule changes triggered black-and-white rigidity. Emotional barriers (anxiety, fear of disappointing supervisors, learned helplessness) compounded each stage. Co-designed GenAI features included pulling tasks from chat logs and the clipboard, AI-estimated durations padded with a 'cheat the brain' buffer to ease initiation, interest-driven task prioritization (experts noted ADHD users often have lower baseline dopamine and need interest-based reward pathways), collaborative task decomposition with progressive subtask reveal to avoid overload, motivational reminders voiced by favorite anime characters or virtual companions, real-vs-planned progress visualization, and adaptive plan adjustment after disruptions. Experts validated the cognitive value of these ideas but warned against three pitfalls: GenAI's sycophancy can reinforce avoidance ('I'm just bad at time management'), interest-only prioritization can let high-stakes deadlines slip, and over-reliance on AI emotional companions can erode the user's own self-regulation skills.
Relevance
For accessibility practitioners and product teams building productivity, planning, or studying tools, this paper is a sharp reminder that ADHD support is not a notification problem - it is a metacognitive scaffolding problem. Three design takeaways carry directly: design GenAI to promote reflection rather than full automation (otherwise users outsource the very monitoring skills they need to grow); calibrate interest against reality so high-stakes work is not silently deprioritized; and frame AI emotional support as a focus partner, not an endlessly soothing companion, to build resilience rather than dependency. The work also concretely surfaces ADHD-specific cognitive realities - lower baseline dopamine, working-memory load from context-switching, alexithymia and interoceptive gaps, learned helplessness, planning-fallacy time underestimation - that should inform any neurodivergent design brief. Limitations to flag: all 20 student participants are Chinese-speaking and recruited via Mandarin social media, so cultural norms around autonomy, family expectations, and AI use may shape generalizability. The study also intentionally focused on metacognition over alternative designs (e.g., physical, calendar-based), so non-AI solutions deserve parallel investigation.
Tags: ADHD · metacognition · generative AI · large language models · co-design · neurodivergence · task management · higher education · emotional regulation · cognitive scaffolding