The University After ChatGPT

The question 0:01

Almost every university student now uses AI. So what has ChatGPT actually changed about the university itself?

This is WAC Signal, from the World Assessment Council. Evidence and ideas for the changing world of learning.

Episode twelve. The university after ChatGPT.

The signal 0:30

In May 2023, a UNESCO survey of more than four hundred and fifty schools and universities found that fewer than ten percent had formal guidance on generative AI. One report cited by Times Higher Education found that as many as a third of Russell Group universities banned student use of AI that year.

Then the direction reversed. In under three years, universities went from banning AI to buying it by the campus. In early 2025 the California State University system gave its four hundred and sixty thousand students and sixty-three thousand faculty and staff access to ChatGPT Edu, in a deal for which Inside Higher Ed reported it paid OpenAI about seventeen million dollars. In 2026 it renewed for three years, at thirty-nine million dollars.

In September 2025 Oxford became the first UK university to offer ChatGPT Edu free to all its staff and students, after a year-long pilot with about seven hundred and fifty people. And from fall 2026, every new undergraduate at Purdue must meet an AI working competency before graduating.

Students got there first. In a 2026 survey by the Higher Education Policy Institute, ninety-five percent of UK undergraduates used AI in at least one way, up from sixty-six percent in 2024.

So the university has changed its rules and its contracts. The question is whether it has changed how it teaches, assesses and prepares people for work.

The evidence 2:22

Start with strategy. In Inside Higher Ed's 2026 survey of three hundred and seventy-six chief academic officers, roughly seven in ten provosts said they use AI in their own work at least weekly. But only one in ten said their institution has a centralized AI strategy, and only one in five agreed it has a coherent vision of how AI will change what and how it teaches.

The returns so far are narrow. Provosts said AI's value has come mostly through individual productivity, at thirty-nine percent, and administrative efficiency, at thirty-six percent. Only eight percent reported institution-wide transformation. And only eight percent said their institution uses AI for personalized learning, often presented as AI's central promise for education.

Then there are the licenses themselves. When the California State University system renewed, the Cal Poly Pomona student newspaper reported that roughly forty-seven percent of staff and students had activated their accounts. Our own rough calculation, from the reported figures, puts the cost at about twenty-five dollars per eligible user a year, and about fifty-three dollars per activated account. And according to NPR, about a quarter of the students who used ChatGPT used the version the system paid for. Most used the free one.

What do students say about AI in their courses? The Digital Education Council's 2026 global survey drew more than twenty-seven thousand student responses in thirty-five countries. Only fifteen percent of students said AI is integrated into many of their courses, and forty-three percent had experienced no integration at all. Among those who had, only five percent said AI had transformed how they learn, and twenty-four percent said it brought no clear learning value. In the United States and Canada, that no-value share reached forty-two percent.

Meanwhile, assessment has moved under everyone's feet. HEPI found that the share of UK undergraduates using generative AI for assessments rose from fifty-three percent in 2024 to eighty-eight percent in 2025. A Guardian survey of UK universities counted almost seven thousand proven cases of AI cheating in 2023-24, or five point one per thousand students, up from one point six the year before.

And the job market that degrees feed is shifting. Using payroll data through June 2026, Erik Brynjolfsson and colleagues at Stanford find that employment of twenty-two to twenty-five year olds in AI-exposed occupations is nineteen percent below where it would be had it kept pace with less-exposed peers, mainly through reduced hiring. The New York Fed reports that recent graduates' unemployment stayed at about five point six percent in the second quarter of 2026, with underemployment edging up to forty-two percent.

Universities are responding in the curriculum. Ohio State announced in June 2025 that, from the Class of 2029, every undergraduate is expected to graduate fluent in AI in their field. Purdue identified or revised twenty-two qualifying courses and updated more than three hundred plans of study for its new requirement.

The tension 6:23

First, much of the change is happening without the faculty. In a report from the American Association of University Professors, based on a survey of about five hundred members, seventy-one percent said AI decisions were led overwhelmingly by administrators, and majorities said AI had worsened the teaching environment and student success. The AAUP itself notes that its responses do not necessarily represent all faculty.

The skepticism is regional. In the Digital Education Council's data, faculty intent to use AI in teaching in the United States and Canada fell from seventy-six percent in 2025 to sixty-seven percent in 2026, the lowest of any region. And forty-three percent of students there said they would support an institution-wide ban.

Second, students say they are not being taught to use what they are given. In HEPI's 2026 survey, sixty-eight percent of UK undergraduates said AI skills are essential, but only forty-eight percent said teaching staff are helping them develop those skills. Globally, only twenty-nine percent of students believe their instructors are well equipped to guide them on AI, even though sixty-four percent of faculty report having taken AI literacy training.

Third, the evidence on new requirements does not exist yet. The requirements at Ohio State and Purdue are too new to have outcome data, and our report found no source showing that a mandated AI course changes graduates' learning or employment.

And the labor-market signal is early. The Stanford authors call their findings early, descriptive indicators, canaries in the coal mine, rather than causal estimates. Some of the divergence predates generative AI.

The implication 8:36

A campus license buys access, not a pedagogy. The university after ChatGPT is not defined by which model it licenses, but by whether its courses, assessments and careers advice have changed around it.

That means the measure of success cannot be seats or activations. It has to be what students can do on their own, after the course, and whether their degree still certifies it.

The WAC view 9:09

In our AI in Learning Report 2026, we propose the WAC AI Learning Readiness Framework. It asks five questions, each a P.

Purpose. What learning problem is AI for, and for whom? A program without that is a procurement, not a strategy. Pedagogy. Does the design make students think, or does it hand them answers? People. Are teachers and students prepared, guided and kept in charge? Proof. Is learning measured without AI, and after a delay, rather than usage counted as impact? And Protection. Are students' data, wellbeing, fairness and rights secured?

For a university, three actions follow. Tie every license to a named learning problem and a measure of success set before you buy, and track activation and learning before you renew. Fund assessment redesign, because secure, supervised assessment needs staff time and invigilation. And bring faculty into the governance, because licenses negotiated centrally, without their support, risk paying for seats that teaching staff then discourage students from using.

The framework is a synthesis of published evidence, not the result of a study we ran, and its levels are informed judgments rather than validated cut-offs.

What to watch 10:47

One. Renewal decisions. As the first campus-wide contracts come up for renewal, watch whether institutions publish activation and learning data, or renew on price alone.

Two. The first outcome data from AI graduation requirements at Purdue, Ohio State and their followers, and whether they show anything beyond course completion.

Three. Hiring of new graduates. Whether the squeeze on entry-level jobs in AI-exposed work deepens, and whether universities answer it with more work-based learning, not only more AI literacy.

The sources for every figure in this episode are listed on the episode page, along with the full AI in Learning Report 2026, at worldassessmentcouncil.org.

WAC Signal is produced by the World Assessment Council, with AI-assisted research and AI voices, and every claim is checked against its source before publication.

This has been WAC Signal. Evidence and ideas for the changing world of learning.