GPT-4o pushed marketing assignment grades up by nearly a full letter grade at Bocconi University in a study of over 1,000 students. The problem is stark: schools reward exactly the skills that AI systems fake most convincingly.
The experiment tracked 1,053 Bocconi University students completing marketing assignments with access to GPT-4o. Students using the AI tool saw their grades jump from an average baseline to nearly a full point higher on a five-point scale. But researchers never measured whether these students actually learned anything. That gap between performance and comprehension matters enormously.
This reveals a structural crisis in how education evaluates competence. Marketing assignments, business writing, analytical essays, and similar structured tasks with clear rubrics respond well to large language models. These are exactly the skills that earn top grades in business schools. GPT-4o excels at pattern-matching against grading criteria. It produces polished, well-organized work that scores high on evaluations built around surface-level metrics: clarity, completeness, professional tone.
The risk cuts deeper than just grade inflation. Recent research suggests that AI-assisted work without independent thinking creates long-term learning damage. Students who outsource reasoning to AI systems don't develop the cognitive frameworks needed to solve novel problems independently. They can produce stellar work on assignments while their actual problem-solving capacity atrophies.
The Bocconi findings highlight a mismatch between what grades measure and what skills matter professionally. Top grades increasingly reflect how well AI can assist with the assignment, not how well the student understands the material. A marketing professional who can't think critically without AI prompting will eventually hit walls that polished copy can't solve.
Universities face pressure to either redesign how they evaluate learning or accept that grades no longer correlate with mastery. Some institutions are shifting toward oral exams, live problem-solving under observation, or collaborative projects that require real-time reasoning. Others are integrating AI literacy into curricula, teaching students when to use these tools and when independent thinking becomes essential.
The broader implication extends beyond marketing classes. Any discipline relying on standardized written assignments faces the same problem. Law schools, engineering programs, and MBA curricula all use assignments that GPT-4o can improve to award-winning quality. Without new evaluation methods, institutions risk graduating students with impressive transcripts and brittle capabilities.
Employers are noticing. Some companies now conduct practical assessments during hiring specifically because they distrust grades from students who've graduated during the AI era. They're looking for people who can think in real time, not people who polished their work perfectly offline.
The Bocconi study quantifies what many educators suspected. AI doesn't fail at college-level work. It succeeds too well at producing what grades reward. Schools that don't restructure evaluation around learning outcomes rather than output quality will watch grades drift further from actual competence. The students who benefit most from AI assistance on assignments may face the harshest surprises when their careers demand independent thought.