Designing Proactive Behaviors in LLM-Based Systems: A Scoping Review

LLM-based systems are increasingly being designed with proactive capabilities—the ability to initiate actions by anticipating user needs rather than waiting for explicit commands. While technical progress has made such behaviors feasible, ensuring they are helpful rather than intrusive remains a significant design challenge. This study presents a scoping review of 36 empirical studies to examine how proactive behaviors are designed in LLM-based systems. We organize our analysis around four dimensions: design goals (why the system initiates), timing strategies (when it acts), presentation content (what it delivers), and presentation methods (how it presents). We also examine how systems adapt their proactive behaviors through user control and feedback mechanisms, and analyze the roles that LLM capabilities play across these design dimensions. Based on our findings, we identify key gaps in current practice and propose directions for future research toward proactive systems that are meaning-oriented and characterized by just-in-time, user-friendly, and adaptive proactivity. This review offers guidance for researchers and practitioners seeking to design LLM-based proactive systems that balance intelligence with usability, user agency, and well-being.