Student teams scored higher when AI joined the brainstorm early

One of five experimental conditions gave student teams DeepSeek V3.2 during a 15-minute brainstorm; another opened it for the 15-minute selection stage of a 40-minute product-design exercise. Creativity combined 1–7 ratings of originality, appropriateness and elaboration with feature-count and naming contributions. Early access averaged 4.88, against 4.52 for late access and 4.39 without AI. Continuous access scored 4.87, giving no clear reason to switch the assistant off after brainstorming.[1]

The October preprint studied one Chinese university, used stage-specific prompts and relied on four industry raters whose agreement was modest. The early–late difference survived the planned statistical adjustment, but two stricter corrections for multiple comparisons removed significance. These are provisional findings about judged proposals.[1]

Make room for ideas to become alternatives

A useful way to interpret the result comes from design practice. The Design Council’s Double Diamond alternates broad exploration with focused decisions. Its delivery stage includes small-scale tests, rejection and improvement. The framework also allows a team to return to an earlier question when testing exposes something it missed; it is not a rigid conveyor belt.[3]

Imagine a team arranging signs for a school event. An assistant proposes coloured routes, pictograms, volunteer guides and printed maps. Put those options on the table early and the group can ask who would use each one, where it would fail and how much work it requires. Introduce them after everyone has committed to a map and the group faces a scheduling decision: reopen the choice, or use the remaining time to make that map workable. This is an illustrative scenario, not a task tested in the experiment.

For that team, “we have four options” would be an inadequate checkpoint. A more useful one is “we can explain why the chosen option fits the visitors and the venue”. The second statement demands a comparison and a reason. Generating another attractive suggestion does not supply either automatically.

Ask for difference, then make a choice

A separate 2024 study in Organization Science gives that distinction some empirical texture. In a circular-economy ideas challenge, human–AI proposals received stronger ratings for value and overall quality, while human-only proposals were judged more novel. A prompting approach that asked for successive ideas to differ from earlier ones improved novelty compared with independent searches. This was a different experiment, not a replication of the timing study.[2]

The practical experiment for a creative team is therefore small and observable: ask an assistant for meaningfully different options at a named point in the meeting, record the options, and write down the reason for the final choice. If the group reopens that choice later, record what new information justified it. This is an editorial suggestion for examining a workflow, not a productivity gain demonstrated by either paper.

A lively brainstorm can leave a team with more to discuss. A useful decision leaves it with something specific to develop. Giving both activities their own place in the meeting makes it easier to see which contribution the assistant is actually making.

Report an error or suggest a correction