Academic procrastination, the intentional delay of tasks despite negative consequences, is prevalent among college students, negatively impacting academic performance and well-being. Existing interventions, focusing on task completion, time management, and motivation, often fall short due to their inability to account for individual learning styles and environmental factors. Large Language Models (LLMs), with their ability to analyze text-based inputs and provide personalized responses, offer a potential solution. This study investigates how LLMs can be used to tailor strategies for managing academic procrastination, focusing on task initiation, deadline management, emotional regulation, and the impact of social and environmental factors. The research questions explored are: RQ1: How do users envision the role of LLMs in tailoring strategies for managing academic procrastination? RQ2: What challenges or tensions might arise when using LLMs for tailoring strategies to manage academic procrastination? The study employed a mixed-methods approach, involving interviews and focus groups with 15 university students and interviews with 6 experts in clinical psychology, education, and cognitive science. A technology probe, SPARK, was used to provide a functional context for participants' feedback. SPARK utilizes GPT-4 to offer context-specific advice based on four strategies: Cognitive Insight, Psychological Flexibility, Value Alignment, and Implementation Intention.

The literature review examines the application of LLMs in academic settings, their use in personalization and contextual adaptation, their role as supportive scaffolds, and associated ethical challenges. LLMs have shown potential in personalizing learning experiences, providing task-oriented support, analyzing student essays for improvement, acting as adaptable tutors in programming courses, and recommending learning materials. However, ethical concerns include the potential negative impact on critical thinking, over-reliance on LLMs, and the risk of inaccurate information. The review also covers existing interventions for managing procrastination, including self-reflection applications, conversational systems, technological interruptions, and therapeutic sessions. While these interventions focus on self-regulation, self-efficacy, and social support, they often struggle to address the nuanced, individualized factors contributing to procrastination. The study addresses this gap by exploring how LLMs can enhance existing interventions and address these nuanced issues.

Fifteen university students (10 women, 5 men; ages 18-25) who self-identified as at least average procrastinators (scoring 24 or higher on the Irrational Procrastination Scale) participated through individual interviews (n=10) and focus group discussions (n=5). Six experts in clinical psychology, learning and education, and cognitive psychology were also interviewed. The study used a semi-structured approach. Participants interacted with SPARK, a web-based technology probe utilizing GPT-4. SPARK comprises four components: a seed message based on established procrastination management strategies; interface options for generating context-specific messages by adjusting tone, directedness, and length; a feature for prompting future action by drafting an email to their future selves; and instructions and examples to guide user input. Participants explored SPARK independently, then discussed their experience, preferences, and potential improvements. Thematic analysis was used to analyze the transcribed interviews and focus groups (approx. 9.5 hours of recordings). Expert interviews were then conducted to validate and expand upon the findings. The researchers followed a thematic analysis approach, using open and consensus coding to identify and refine codes. These codes were then grouped into broader themes through axial coding. Ethical considerations included informed consent, participant safety (Columbia-Suicide Risk Assessment protocol), and compensation.

The study identified five key themes: 1) **Aspirations for Structured Action Steps**: Participants valued structured guidance and step-by-step instructions, but experts cautioned against over-reliance on the tool. 2) **Deadline-Driven Instructions**: Participants appreciated features like drafting emails to their future selves and integrating the tool with calendars for timely reminders, with experts noting the diverse motivational needs at different times. 3) **Guided Versus Unguided Questions**: Opinions varied on the optimal level of guidance in prompting user input, with some preferring guided questions while others preferred less intrusive prompts, particularly during busy periods; experts suggested adaptive questioning based on contextual factors. 4) **Concerns and Boundaries in Providing Emotional Support**: Participants expressed the need for emotional support, but experts stressed the importance of clear disclaimers and avoiding therapeutic advice. 5) **Providing Support on the Use of LLM-based Tools**: Participants desired more examples and instructions on how to use the tool effectively and proposed a collaborative element (shared prompt library) and feedback mechanisms. The need for direct solutions, step-by-step guidelines, structuring recommendations, and adaptive breakdown were frequently mentioned. The importance of deadlines and using calendars for organization were also significant. Participants discussed the need for varied levels of guidance and concerns regarding overreliance on the tool. The limitations of LLMs in addressing emotions and the need for clear disclaimers were highlighted.

The findings address the research questions by illuminating user perceptions of LLMs in tailoring procrastination management strategies and highlighting potential limitations and tensions. The desire for structured guidance with adaptability emphasizes the need for tools that balance systematic task management with individual needs. Integration with user routines and deadlines underscores the importance of context-aware task prioritization. The collaborative support for using LLM-based tools reflects the broader need for shared learning and best practices. The study highlights ethical considerations concerning emotional support and the potential for over-reliance, emphasizing the need for careful design and transparent communication of limitations. The findings contribute to the broader literature on human-LLM collaboration by demonstrating the nuanced ways in which users engage with LLMs, highlighting their potential as collaborative partners while acknowledging ethical considerations.

This study contributes to understanding the design implications of LLM-based tools for managing academic procrastination. It emphasizes the need for structured, deadline-focused guidance, user support mechanisms, and adaptive questioning, while cautioning against using LLMs for therapeutic advice. Future research should explore longitudinal deployments, investigate diverse populations, and broaden the scope to other behavioral or psychological challenges. These findings offer valuable insights for developers and researchers aiming to design effective and ethical LLM-based intervention tools.

The study's limitations include a geographically limited participant group (North America) and the absence of a longitudinal study to assess long-term effectiveness. The findings may not directly generalize to other behavioral areas. Future studies should address these limitations by incorporating diverse populations, employing longitudinal designs, and exploring applications in various contexts.