Large Language Models (LLMs) are powerful AI agents but struggle with open decision-making in complex scenarios. To address this, we propose the Multi-Perspective Team Tactic (MPTT) framework, using the language logic game “Who is Undercover?” (WIU) as a testbed. MPTT enhances LLMs’ language logic, multi-dimensional thinking, and self-awareness. Through alternating speaking and voting sessions and techniques like self-perspective, identity determination, self-reflection, and multi-round collaboration, LLM agents make rational decisions by strategically concealing information and building trust. Experiments show that MPTT, paired with WIU, harnesses LLMs’ cognitive capabilities to simulate societal decision-making, supporting minority group communication and promoting fairness. We also introduce a word network dataset with multidimensional semantic relations for WIU and other language-intensive tasks. Human-in-the-loop experiments further reveal LLMs’ ability to learn and align with human behavior, showcasing their potential in societal decision-making.