Whom to Query for What: Adaptive Group Elicitation via Multi-Turn LLM Interactions

Eliciting information to reduce uncertainty about latent group-level properties is a central problem in collective assessment, preference modeling, and opinion aggregation, and is especially important in survey-based studies. While natural language interactions provide a flexible interface, existing methods typically rely on fixed questionnaires and static respondent sets, and do not adapt to partial or missing responses across rounds. To address this gap, we study adaptive information elicitation through multi-turn interactions between a large language model and a group of individuals, where both queries and respondents are adaptively selected to infer latent group properties. We propose a theoretically grounded framework that, at each round, jointly selects a query and a subset of respondents based on previously observed responses to efficiently reduce uncertainty about a target latent quantity (e.g., group-level political inclination). Motivated by practical survey constraints, such as limited questions and costly participation, our strategy maximizes information gain under a fixed budget. To handle missing and incomplete responses, we combine graph neural networks for aggregating/imputing partial group information with an information-theoretic criterion that guides per-round selection. Across three real-world opinion datasets, we achieve consistent improvements in population-level response prediction under constrained budgets, including over a 12% relative gain on CES at a 10% respondent budget.

Understanding what a group of people thinks about politics, policy, or social issues usually means asking everyone the same fixed set of questions. But this is slow, expensive, and ignores the fact that some questions and some respondents are far more informative than others. What if we could run smarter surveys that learn as they go, asking the right questions to the right people at each step? We developed an AI-driven framework that conducts surveys as an adaptive conversation. At each round, the system selects both which question to ask and which individuals to ask it to, based on what it has already learned, much like a detective who focuses follow-up questions where the answers are most revealing. When some people do not respond, the system fills in the gaps using a network-based model that draws on the responses of similar individuals. We tested our approach on three real-world public opinion datasets and found it consistently outperforms standard survey methods under tight budgets, achieving over 12% better accuracy in predicting group opinions while surveying only 10% of respondents. This work could make opinion polling, market research, and policy surveys faster, cheaper, and more accurate, helping organizations better understand the communities they serve with fewer resources.