What researchers tested
More than 1,000 first-year Bocconi University students completed a marketing case for the university store. Class periods were randomly assigned to receive GPT-4o access, a causal-reasoning exercise, both interventions or neither.
Human graders used a five-point rubric, while automated analysis measured idea count, variety, causal reasoning and similarity to expert submissions. OpenAI reports that ChatGPT access raised average rubric performance by almost one point and produced more expert-like work.
Different interventions, different gains
The causal-reasoning exercise did not raise the conventional rubric score, but students produced a wider range of ideas and explained why proposals might work or fail. Students receiving both interventions retained the broader idea variety while matching the stronger rubric performance of the ChatGPT group.
That split matters because a polished final answer and evidence of learning are not interchangeable. A rubric optimized for conventional quality can miss originality, process knowledge and whether students can apply a method without assistance.
Limits for schools
The experiment concerns one institution, one business task and one model configuration. It does not establish retention, independent transfer or effects across age groups and disciplines. The OpenAI collaboration and automated measures should also be considered when interpreting the findings.
AINewsInu's homepage and How-to coverage track practical AI literacy. Schools can use this result to design better trials: evaluate process artifacts, oral explanation, source verification and later unaided performance alongside the finished assignment.
Sources & further reading
Social-media activity is treated as a signal of attention, not proof. Product claims are attributed to the linked publisher or announcement.