A new experiment from Bocconi University and OpenAI Economic Research with over 1,000 first-year undergraduates found that ChatGPT access and critical-thinking training produced distinct, complementary effects on student work quality and originality. Students were randomly assigned to one of four groups: ChatGPT access (GPT-4o), causal reasoning training, both, or neither. They worked on a real-world marketing case for the university’s merchandise store.

Students with ChatGPT access scored nearly a full point higher on a five-point grading rubric. Their submissions included more ideas, clearer logic, and greater similarity to expert-written recommendations. AI helped novices produce more professional-looking work, but students still had to decide what to ask and which responses to incorporate.

The causal reasoning exercise—a form of critical thinking unrelated to AI—produced a different effect. It did not improve rubric scores (which measured standard marketing goals), but text analysis showed these students generated a wider, more distinct range of ideas compared to peers. The training encouraged broader thinking that a traditional rubric failed to capture.

Students who received both interventions showed benefits from each: idea variety matching the training-only group, rubric scores matching the ChatGPT-only group, and stronger logical coherence with more evidence of questioning assumptions. The randomized design allowed researchers to isolate these separate and combined effects.

The experiment highlights a challenge for schools: if AI helps students produce polished, expert-like answers, then final outputs alone reveal less about actual understanding. The authors suggest assignments and evaluations may need to evolve to reward originality, reasoning, and multiple approaches, not just conventional polished answers.