Why are students choosing the degrees AI is most likely to change?

When ChatGPT arrived in November 2022, the jobs that looked most exposed were the white-collar ones built on words, code and numbers: programming, accounting, law. If you were 17 and filling in your UCAS form that winter, you might reasonably have steered clear of those subjects.

That is also what economics predicts. Students, the textbook says, move away from subjects whose job prospects are getting worse. A working paper published in August by researchers at King’s College London tested that prediction against six years of UK university applications. Students did the opposite.

How the study worked

The researchers treated each subject at each university as its own degree, so economics at one university and economics at another count separately. They used millions of LinkedIn profiles to see where each degree’s graduates had gone to work before ChatGPT existed, and scored each degree by how exposed those jobs are to AI.

“Exposed” has a specific meaning here. It is the share of a job’s tasks that a large language model could do in half the time at the same quality. It is not a forecast that the job will disappear.

Computing, maths and engineering came out as the most exposed. Nursing and other subjects allied to medicine, education and medicine came out the least. The researchers then followed applications from UK students to 1,090 degrees across the 2020 to 2025 application cycles.

Towards, not away

Before ChatGPT, applications to more and less exposed degrees moved in step. Afterwards they split. A degree one notch more exposed (one standard deviation, in the paper’s terms) saw applications grow about 6% faster than less exposed ones. For a typical degree, that is roughly 60 extra applications a year. The gap was 4% in the 2023 cycle and 8% in 2024, then stopped widening in 2025.

The raw numbers make the shape plain. Comparing the three cycles after ChatGPT with the three before, applications rose 22% in computing, 14% in engineering, 13% in law and 10% in maths. They fell 17% in education and 11% in subjects allied to medicine, which includes nursing.

This wasn’t more people going to university. Total applicant numbers stayed broadly flat, so students were moving between subjects. The shift pointed the same way for men and women and for every ethnic group the researchers looked at.

It’s worth seeing how reasonable this looks from a 17-year-old’s chair. In the winter of 2022, the loudest thing about AI was that it was new and exciting, and computing looked like the place the future was being built. Picking the subject at the centre of the story isn’t foolish. It just answers a different question from the one that matters three years later. Not “what’s interesting now?” but “who will be hiring people like me when I finish?”

Meanwhile, at the bottom of the ladder

If those subjects’ job prospects had improved, the move would make sense. So the researchers checked, using the same LinkedIn data to follow jobs in the fields each degree’s graduates usually enter.

They found no sign of improvement. The share of junior and entry-level positions in those fields fell slightly faster the more exposed the degree: 0.25 percentage points for each notch of exposure. Job adverts in exposed fields also dipped, though that result only held in some versions of the analysis.

These are small effects so far, and they describe whole fields rather than any particular employer. But they point the wrong way for the students heading towards them. In the authors’ summary, students are “moving toward the subjects in which entry-level job opportunities are declining.”

The junior share is the number we’d watch most closely. A field can keep growing overall while its front door narrows: if employers lean on experienced staff and AI tools, they can do more work while taking on fewer people with no experience. For a graduate, the size of an industry matters much less than how many beginners it takes on each year. That is the figure a degree choice is really a bet on, and it’s the hardest one to see from a UCAS form.

Why would students do that?

The paper offers three explanations and doesn’t claim to know which is right.

The first is that students simply don’t see it. Earlier research finds young people hold optimistic beliefs about subjects they find attractive, and changes in junior hiring may take years to reach someone choosing a degree, if they reach them at all.

The second is that students see AI as an opportunity rather than a threat. A computing degree may now attract people who want to build the technology, not compete with it. AI may also widen where a degree can take you: the authors’ own example is a politics graduate becoming a software engineer because coding has become easier to pick up.

The third is that something else pushed the same way. Inflation may have sent students towards subjects they associate with good pay, which happen to be the exposed ones. The authors note that those pressures began a year or more before applications shifted, and that the shift lines up with ChatGPT’s release. They still can’t rule out other causes, and universities may have recruited harder for those courses over the same period.

Our view: the first explanation is the one worth worrying about, because it’s the one that can be fixed. The authors point to experimental evidence that accurate information about jobs corrects students’ beliefs. The second isn’t foolish either. As we wrote about Accenture, some employers are still hiring juniors while redesigning their work around AI. But “AI is exciting” and “AI-exposed fields are taking on more juniors” are different claims, and so far the data supports only the first.

The deeper problem is timing. A student choosing a degree in 2023 was relying on a picture of the job market drawn by people who graduated years earlier, and will graduate into one nobody can see yet. That gap, between the market as it’s described and the market as it is, is a big part of why Clive exists.

What it means if you’re graduating

Students who applied in the 2023 to 2025 cycles graduate between 2026 and 2028. If the pattern holds through to graduation, more people with computing, engineering, law and maths degrees will be looking for junior roles in fields where those roles make up a slightly smaller share of the work.

There’s a plainer point too. A subject getting more popular tells you what other 17-year-olds think, not what employers will need. If anything, a rush into a subject means more people competing for the same junior roles at the end of it.

That isn’t a reason to panic about your subject. Exposed is not the same as replaced, and the shifts the paper measures so far are small. It is a reason to look at the market as it is now, rather than as it was described when you chose your course.

Check how many entry-level roles are actually open in the area you want, not just whether the subject sounds employable. And look at where graduates from your own course went: the study found that exposure can vary a lot between universities offering the same subject.

The authors end by saying the next few application cycles will show whether students change course once AI’s effect on jobs is easier to see. For anyone already holding the degree, the useful part is the same either way: know what the entry-level market looks like before you plan around it.

References

01

Vossos, Andronic, Klein Teeselink and Akileswaran (King’s College London). Artificial Intelligence and the Future Supply of Skills: Evidence from UK University Applications. Working paper, 26 August 2026. Pages 1–15 and 23. Supports the application, exposure and junior-share figures; it is not yet peer reviewed, covers domestic UCAS applications to 1,090 university-subject pairs over the 2020–2025 cycles, excludes psychology, creative arts and languages (26.5% of applications), and reports associations around ChatGPT’s release rather than proven causes.