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 fail to account for individual learning styles and environmental factors. Large Language Models (LLMs), with their ability to analyze text-based input and personalize responses, offer a potential solution. Unlike traditional methods or rule-based chatbots, LLMs can dynamically interact and integrate various procrastination management strategies. Their ability to tailor advice based on individual circumstances (e.g., deadlines, work schedules) increases the practicality and effectiveness of interventions. However, challenges exist. LLMs lack the capacity to perceive emotions or navigate complex social contexts, requiring users to provide detailed information. Understanding user expectations regarding personalization (e.g., reminders vs. time management tips) is crucial for effective prompt engineering. The study aims to understand how users envision LLMs in managing academic procrastination, focusing on task initiation, deadline management, emotional regulation, and the impact of social and environmental factors. The research questions 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 literature review covers two main areas: the application of LLMs in academic settings and interventions to manage procrastination. Regarding LLMs, the review highlights their use in personalization (analyzing student data to suggest improvements), adaptive tutoring (providing context-specific hints in programming), and personalized learning pathways (recommending relevant courses and materials). Ethical concerns, such as potential negative impacts on critical thinking and the risk of disseminating inaccurate information, are also discussed. The review then examines existing procrastination management interventions, including self-reflection applications, conversational systems, technological interruptions, and therapeutic sessions. It highlights the limitations of these interventions in addressing nuanced, individual-specific factors that contribute to procrastination, setting the stage for the potential of LLMs to address this gap.

The study employed a mixed-methods approach involving 15 university students (aged 18-25, self-identified as at least average procrastinators) and 6 experts (clinical psychology, education, cognitive science). Data was collected through individual interviews and focus group discussions. Participants interacted with a web-based technology probe, SPARK, which uses GPT-4 to offer context-specific advice. SPARK includes four components: a seed message based on existing procrastination management strategies (cognitive insight, psychological flexibility, value alignment, implementation intention); interface options to generate context-specific messages (allowing users to specify tone, directedness, length, and inclusion of instructions); a feature for prompting future action (drafting an email to their future selves); and instructions and examples to guide user input. After interacting with SPARK, participants shared their thoughts on the tool's utility and potential features. Thematic analysis was applied to the transcripts, followed by expert interviews to validate findings and explore potential design challenges. Ethical considerations included informed consent, participant comfort and safety measures, and training in suicide risk assessment for interviewers.

The key findings are organized into five themes: 1. **Aspirations for Structured Action Steps:** Participants valued customizable features (tone, directedness, length) but also emphasized the need for structured, step-by-step guidance, comparing it to following a recipe. Experts cautioned against overreliance, suggesting a balance between structured guidance and independent thought. 2. **Deadline-Driven Interactions:** Participants appreciated the email-to-future-self feature for deadline management and suggested integrating the tool with existing platforms (Google Calendar, etc.) for timely reminders. Experts emphasized the diverse motivational needs and suggested providing multiple motivational prompts. 3. **Guided versus Unguided Questions:** Opinions varied on the optimal number and type of questions. Some preferred guided questioning, while others were concerned about overwhelming users, particularly during busy periods. Experts recommended adaptive questioning based on contextual factors (time of day, workload). 4. **Concerns and Boundaries in Providing Emotional Support:** Participants expressed a need for emotional support, but experts cautioned against LLMs providing therapeutic advice. They emphasized the need for clear disclaimers and links to mental health resources. 5. **Providing Support on the Use of LLM-based Tools:** Participants desired guidance on how to phrase requests for optimal results, suggesting video tutorials, examples, and collaborative features (shared prompt library). Experts agreed but stressed the importance of clearly distinguishing helpful from unhelpful contexts.

The findings address the research questions by showing that users envision LLMs as valuable tools for structured, deadline-focused procrastination management, but also highlight potential limitations. The need for structured guidance with adaptability underscores the importance of creating tools that offer both systematic task breakdown and flexible customization. The integration with user routines and deadlines points to the need for seamless integration with existing platforms and dynamic task prioritization. The collaborative support aspect highlights the benefits of incorporating shared resources and experiences. The study also emphasizes the crucial need for ethical considerations, particularly in addressing emotional support and avoiding overreliance on the tool.

This study provides valuable insights into user expectations and potential limitations of LLM-based tools for managing academic procrastination. Key recommendations include designing tools that offer structured guidance with adaptive complexity, setting boundaries in addressing user emotions, and incorporating collaborative support. Future research should focus on longitudinal studies to assess long-term effectiveness, explore diverse cultural contexts, and expand the application of these tools to other behavioral challenges.

The study's limitations include a geographically limited participant pool (North America) and the lack of a longitudinal study to assess long-term effectiveness. Furthermore, the findings are specific to academic procrastination and may not generalize to other contexts. Future research should address these limitations by including more diverse participants, conducting longer-term studies, and exploring broader applications of LLM-based tools.