- A 5-month study with 60 managers from 12 companies shows that introducing generative AI into group meetings does not automatically improve collaboration as many leaders expected.
- A survey from the Capgemini Research Institute states that the number of meetings using AI is expected to more than triple in the next 3 years, but the research team found that AI can make members passive, reduce interaction, and erode collective ownership of decisions.
- In the early stages, many groups almost “silently watched the screen,” letting AI lead the entire discussion instead of debating together. Team engagement levels were rated low after the first session.
- Groups often made 3 major mistakes:
- Using AI as a tool for individuals rather than for the whole team.
- Assigning AI a fixed role such as “expert.”
- Entering short prompts lacking context, allowing AI to navigate the meeting.
- The research team built the “Human-AI Team Chemistry” model with 3 principles:
- Interacting with AI as a team.
- Giving AI many different roles.
- Maintaining collective control over prompts as well as AI outputs.
- After applying the new method, teams began introducing each member’s role to the AI, using it as a critic, simulated customer, storyteller, or competitor to expand perspectives.
- Results showed that participation increased by about 30%, and about 66% of participants reported that the quality of collaboration and group discussion improved significantly.
- About 3/5 of participants noted that collective evaluation helps reduce the risk of over-trusting AI or being led by AI toward groupthink.
- The study recommends that leaders pre-design “AI slots” in meeting agendas, prepare multi-role prompts, and use transcripts to evaluate how the team interacts with AI after each session.
📌 This study shows that AI does not automatically help teams work better and can even reduce interaction if implemented incorrectly. A test with 60 managers across 12 companies proves that when a team collectively controls prompts, uses AI in multiple roles, and maintains collective debate, engagement increases by 30% and decision quality clearly improves. AI is most effective when it becomes a “flexible member” of the team rather than a passive answering tool.
