OpenAI-backed Bocconi study finds ChatGPT access and critical-thinking training have complementary effects on student work

A randomized experiment with more than 1,000 Bocconi University students, run in collaboration with OpenAI Economic Research, finds that ChatGPT access and a short critical-thinking exercise improve different, non-overlapping aspects of student work rather than substituting for each other.

What's new

More than 1,000 first-year undergraduates worked on a real business case, developing marketing recommendations for the university's merchandise store. Students were randomly assigned by class period to one of four groups: access to ChatGPT (GPT-4o), a causal-reasoning training exercise, both, or neither. The training was unrelated to AI, teaching causal-reasoning concepts — explaining why a proposed solution would or wouldn't work — through a game, worked examples, questions, and feedback. Submissions were graded by trained human raters on a five-point rubric and separately scored by automated text analysis for idea count, idea variety, signs of causal reasoning, and similarity to expert-written recommendations.

"The students who had access to ChatGPT scored almost a full point higher on the five-point scale. Their answers included more ideas, followed clearer logic, and were more similar to recommendations written by experts," OpenAI reported. Students were not simply submitting AI output verbatim — they still had to decide what to ask, evaluate the responses, and choose what to include.

The causal-reasoning training had the opposite profile: it did not raise rubric scores, since the rubric only measured how well recommendations addressed two standard marketing goals, but students who completed it produced a wider range of ideas that were more distinct from their classmates' submissions than the ChatGPT-only group's work — a signal of originality the rubric did not capture. Students who received both ChatGPT access and the training showed gains across the widest set of measures, combining the two groups' distinct benefits.

Context

The study is one of OpenAI's more methodologically rigorous attempts to measure AI's classroom effects, using a randomized controlled design rather than surveys or observational data. It arrives alongside a broader OpenAI education push this year, including expanding ChatGPT for Teachers to more U.S. school districts and a separate report on how students and educators use ChatGPT for continuous learning outside class. It also lands amid an active public debate over whether everyday ChatGPT use in education erodes students' capacity for independent reasoning.

Why it matters

The results complicate a common framing of the AI-and-education debate as a binary choice between teaching students to think for themselves and letting them use AI tools. Because ChatGPT access and critical-thinking training moved different metrics — polish and coherence versus originality and idea variety — the study implies that giving students AI tools without also teaching reasoning skills may raise the surface quality of their work without making it more original, while critical-thinking instruction alone does not produce higher-graded outputs by the same rubric. OpenAI frames this as an argument for redesigning assessments to explicitly measure originality and reasoning rather than just polish, and as research-backed support for products like ChatGPT's study mode, which is built around guided reasoning rather than direct answer generation.