Why are there more top grades at university? ChatGPT is to blame

The number of top grades at universities has been rising for a long time. But since 2022, when ChatGPT appeared, that figure has jumped by 30%. The increase is concentrated in courses with heavy writing or programming workloads, where using AI is easier. In contrast, in classes that rely on oral presentations, the effect is smaller.

That pattern emerges from a study based on data from half a million students across 319 courses at a Texas university. The research also finds that average grades have risen overall and that grades are increasingly skewed upward. One curious detail: the marks most likely to turn into top grades are the ones just below them.

“The biggest jumps aren’t among students on the verge of failing, but among those already in the middle‑to‑high range,” says Igor Chirikov, the study’s author and a researcher at the Center for Studies in Higher Education at the University of California, Berkeley. AI does not rescue those who were going to fail so much as it nudges upward students who were already doing fairly well.

This inflation of top grades is only one trend in AI’s disruptive arrival on university campuses. On Thursday, Science published another study showing that, based on 2024 data, 9% of students cheat using AI. That figure reflects only the first full year of ChatGPT’s existence. It is now higher: “The latest evidence suggests it has continued to grow. We’re currently finishing data collection for 2026, and preliminary results from several universities show a significant increase in students’ use of AI,” Chirikov says. The grade‑inflation study, which includes data through fall 2025, shows that courses exposed to AI — especially those built around homework — have seen substantial increases in top marks.

The fields with the most cheating are economics (17%) and journalism (16%), while biology students are the least likely to use AI to cheat, at just 5%. The authors, who gathered data from a survey of more than 95,000 students at 20 U.S. institutions, do not have a clear hypothesis for why some majors cheat more than others.

“Misuse rates may vary depending on the kinds of students who choose each major and the types of assignments used to evaluate them,” says René Kizilcec, co‑author of the study and a professor at Cornell University’s Ann S. Bowers College of Computing and Information Science, where he directs the Future of Learning Lab.

The fields with the highest cheating rates are economics (17%) and journalism (16%), while biology students use AI to cheat the least, only 5%. The authors, who obtained the data from a survey of more than 95,000 undergraduates at 20 U.S. institutions, do not have a clear hypothesis for why some majors have higher cheating rates: “Misuse rates may vary depending on the type of students who choose each major and the kinds of assignments used to assess them,” says René Kizilcec, co-author of the paper, professor at the Ann S. Bowers College of Computing and Information Science at Cornell University, and director of the Future of Learning Lab.

Grade inflation has actually been happening for decades. At Harvard University, the share of top grades rose from 24% in 2005 to 60.2% in 2025, according to another academic paper. But unlike today, back then it was the professors who had incentives to inflate grades for their own benefit: first, generous grading led to better student evaluations and improved promotion prospects, and higher grades helped students gain admission to competitive graduate programs — prompting other universities to follow suit.

Now something different is happening: assignments simply look better, without the student having done anything to deserve it. Some students submit top‑grade work they did not write. The author calls this a “task shift,” in which AI replaces the student rather than enhancing their abilities.

A global problem

Both studies were conducted with data from U.S. universities. But it is a global phenomenon: “The mechanism is not unique to the United States: if students have access to AI and courses rely on writing, programming or similar assignments, grades can rise without a corresponding increase in real competencies,” Chirikov says.

However, the magnitude of the problem may vary by location, he says: “It could be smaller where assessment relies more on in-person exams, oral exams, external grading, or stricter grading systems, or where instructors have already redesigned their assignments. It may also depend on students’ access to AI, institutional policies, and how well AI tools work in their language.”

Despite these possible variations, this rise in top grades and cheating at universities is no surprise to professors and students. These studies quantify what was easy to suspect. Solutions, however, are more complicated. Some instructors have opted, for example, to require students to disclose deliberate AI use on each assignment. But that is not an option that works for every task or course, says Chirikov.

He proposes three directions: “Use supervised environments when individual student performance must be verified; clarify, case by case, which AI uses are acceptable and which are not; and redesign assignments so AI is either limited or intentionally incorporated.”

The authors stress that there is no single AI‑proof solution. Reform must be tailored to each course and aligned with what the assessment is meant to measure.

The hardest part is convincing students that taking the easy path early on will come back to haunt them. “It is the same problem behavioral economists have faced when trying to encourage people to save early,” Kizilcec says. “Exams that include tasks that cannot be done with AI can help motivate students to learn the fundamentals, because they know they will be assessed on them sooner or later. There is nothing wrong with a cognitive shortcut — we use them constantly — but developing metacognitive skills, the judgment to know when it’s worth investing real effort in a task, that is crucial for students to learn.”

Sign up for our weekly newsletter to get more English-language news coverage from EL PAÍS USA Edition