According to a  global survey  of 500 executives conducted by the Capgemini Research Institute, the active use of AI in team meetings is anticipated to more than triple  in the next three years. The executives surveyed said that they expect group use of AI to lead to better meetings that arrive at higher-quality outcomes. Leaders should not let this optimism obscure the challenges ahead. Our  research  suggests that integrating AI into team settings doesn’t happen naturally, and introducing AI into meetings without laying the proper groundwork can narrow participation, fragment discussions, or shift ownership away from the team.

Fortunately, there is an approach that overcomes these pitfalls: we call it “Human-AI Team Chemistry.” Our research indicates three practices to help build this new capability as AI becomes more embedded in organizations:

  1. Engage with AI as a team.  Participants should introduce themselves and involve the AI in a collective dialogue so that it addresses the group, considering the various expertise at the table.
  2. Leverage AI’s role fluidity. AI should be used not just as a note-taker but as a multi-role team member, deliberately switching roles (such as stakeholder representative, challenger, customer, competitor, etc.) to enrich the team discussion.
  3. Maintain collective ownership of the interactions with AI.  When team members treat prompting as a collective act, debate alternative directions, and pause to judge AI’s output throughout the meeting, interactions with AI will advance their thinking rather than outsourcing  it.

These  recommendations emerged from a  five-month experiment  involving 60 managers from 12 companies across diverse industries. In each organization, a team of three to four managers – all with prior individual experience using generative AI – was tasked with designing a  platform-based  solution to address a strategic business challenge of comparable scope and complexity. To ensure consistency, all teams followed the same methodology, developed by two of us (Daniel and Tommaso), and used OpenAI’s ChatGPT model. Each team met in person five times, for a total of 30 hours of collaborative work. To understand how teams actually worked with AI and experienced the process, we used three complementary methods: we observed team interactions with the AI in real time, analyzed the complete chat transcripts from each session, and collected post-session surveys to capture participants’ feedback. This combination allowed us to see not just what teams produced, but how they collaborated and where team-AI collaboration thrived or faltered.

Based on this research, we believe that teams that engage in these practices  will achieve higher‑quality outcomes  and reduce the risk of falling into common AI‑related traps.