Over the years, conversational search has transformed how users engage in information seeking through multi-turn dialogues. Traditional conversational search systems are primarily reactive, responding to explicit user-issued queries at each turn [2, 10]. In contrast, modern conversational search treats information seeking as a dynamic process in which users refine goals, explore alternatives, and make decisions across turns [7, 9]. However, most conversational search systems remain fundamentally reactive. They support search by clarifying ambiguous queries [4], maintaining conversational context [5], retrieving relevant documents across turns [6], and generating responses [3], but they rely on explicitly expressed user queries. As search tasks grow more complex, users' information needs are often incomplete, evolving, and not fully articulated. In such settings, users may not know what to ask next or what information is missing. This limitation motivates a shift from reactive systems to proactive ones. This paradigm, known as proactive conversational search (PCS), aims to anticipate users' future information needs [1] by inferring implicit intent from the natural flow of conversation, and to proactively surface relevant recommendations that the user may not have explicitly considered, ultimately reducing user effort over the course of the search process. Importantly, the goal is not merely to retrieve information earlier, but to identify implicit information needs that are not yet expressed, supporting users in complex decision-making scenarios. Thus, proactivity is a task-dependent capability that becomes essential as user intent becomes more complex, implicit, and uncertain.

In the information retrieval (IR) domain, evaluation plays a crucial role in optimizing search experiences and supporting diverse user intents.In the recent LLM era, research has been conducted to automate document relevance labels.These labels have traditionally been assigned by crowd-sourced workers, a process that is both time consuming and costly.This study focuses on LLM-generated usefulness labels, a crucial evaluation metric that considers the user's search intents and task objectives, an aspect where relevance falls short.Our experiment utilizes task-level, query-level, and document-level features along with user search behavior signals, which are essential in defining the usefulness of a document.Our research finds that (i) pre-trained LLMs can generate moderate usefulness labels by understanding the comprehensive search task session, and (ii) pre-trained LLMs perform better judgment in short search sessions when provided with search session contexts.Furthermore, we investigate whether LLMs can capture the unique divergence between relevance and usefulness, along with conducting an ablation study to identify the most critical metrics for accurate usefulness label generation.In conclusion, this work explores LLM-generated usefulness labels by evaluating critical metrics and optimizing for practicality in real-world settings.