Researchers randomly assigned participants to 68 human-human pairs or 72 human-AI pairs. The pairs alternated who responded first while completing an Alternate Uses Task, which asks for unusual uses for an object, and a creative short-story task.

Human-human collaboration produced stronger creativity that increased over time. Although GPT-4o generated more verbose responses, working with it did not improve the originality or diversity of the group’s ideas and was linked to less creative responses from its human partners.