Su Xu, Dewen Liu
Purpose This research investigates how collaborative task type (exploratory vs. exploitative) and user AI literacy interact to determine the optimal communication strategy (polite vs. direct) for an AI collaborator. Design/methodology/approach Three experiments were conducted. Study 1 (N = 439) used a live interactive chatbot to enhance ecological validity, while Studies 2 (N = 395) and 3 (N = 437) used static stimuli to ensure experimental control and examine the moderating role of user AI literacy. Findings A polite communication style is more effective for exploratory tasks, whereas a direct style is more effective for exploitative tasks. This effect is symmetrically mediated by the AI's perceived experience and agency, respectively. Importantly, this psychological mechanism is significant only among users with low AI literacy. Originality/value By challenging the assumption that polite AI is universally beneficial, this study reframes human-AI collaboration through a task-contingent lens. It unveils a symmetrical psychological pathway where polite AI signals experience to foster exploration, while direct AI signals agency to drive exploitation. Furthermore, identifying AI literacy as a critical boundary condition, this research demonstrates how technical sophistication attenuates anthropomorphic heuristics, providing actionable insights for designing adaptive AI interfaces.