Firm or Friendly? AI Assistance Styles and Personality in Live Negotiation

Abstract: Generative AI is increasingly deployed to assist negotiators, yet we know little about how real-time AI assistance affects live negotiations or how different assistance styles interact with individual personality. Integrating the dual-concern model of negotiation with the interpersonal circumplex, we designed four AI assistance styles varying along agency (dominant vs. submissive) and communion (warm vs. cold) dimensions. Using Pairit, an online experimentation platform, we conducted a pilot study (N=98) of a live human-human negotiation with real-time AI assistance and are completing a preregistered study (N=800). Pilot results revealed no main effect of AI assistance, but significant heterogeneity by style: dominant assistance increased value claimed while warm assistance increased counterpart subjective value. We also found evidence for reinforcement dynamics (dominant-cold assistance most effective for dominant-cold individuals) and compensation dynamics (dominant-warm assistance most effective for submissive-cold individuals). These findings demonstrate that AI assistance effectiveness depends on both how the AI assists and who it assists.

Keywords: AI assistance, live negotiation, personality, AI style