Claims about what increasingly capable AI does to judgment, learning, creativity, coordination, wellbeing, and decision making are arriving faster than the field can evaluate and accumulate them. A key bottleneck is methodological: "AI use" is often reported as if it were a single, stable intervention, even though outcomes depend on what system was used, how collaboration was structured, and what constraints the platform imposed. Without clear reporting of model/version and interface, enabled capabilities, turn structure and initiative, task allocation and integration rules, and constraints such as tier differences, rate limits, and updates, along with records of prompts, and what participants actually saw, results are hard to interpret, compare, or reproduce. This article proposes a practical response: a compact human-AI collaboration reporting guide, CoCREATE-AI, that specifies the minimal information needed to interpret, compare, and reproduce findings, while also limiting overgeneralized claims about "AI" that exceed what the reported intervention can support.
Our study examined how individual creativity translates into group-level outcomes in collaborative versus independent trio settings, and whether group performance aligns with the predictions of the additive or the disjunctive model, while applying the same analytical framework to fluid intelligence for comparison. Using a longitudinal design, 131 participants completed the Test for Creative Thinking-Drawing Production and the Advanced Progressive Matrices independently in Session 1. Two months later, 120 participants returned and were assigned to intelligence-stratified trios in Session 2, performing the same tests in an either collaborative or and independent setting. Bayesian analyses showed strong evidence for the effect of collaboration in Session 2. Model based predictions of Session 2 outcomes in collaborative settings revealed strong evidence against the additive model and substantial evidence in favor of the disjunctive model. These findings suggest that at least for tasks emphasizing immediate output quality, in situations where small ad hoc teams composed of largely unfamiliar individuals work over a short period of time, organizations are unlikely to benefit from collaborative teamwork, as highly capable individuals may perform comparably both alone and as the most capable member of a small ad hoc team. In other settings, however, teamwork may be more conducive.
