New Grads From AI-Exposed Majors Are Earning 13 Percent Less. Economists Can't Agree Why.

College graduates who majored in computer science, accounting, journalism, or engineering earned 13 percent less in their first jobs after December 2022 than graduates from the same majors did before ChatGPT launched, according to a U.S. Census Bureau working paper published this month. The authors, Cody Orr, Lee C. Tucker, and Lawrence Warren, call the drop comparable in size to the earnings losses tied to graduating into a large recession.

Where the hiring gap actually shows up

About half of that 13 percent gap comes from lower pay within the same industry, the Census authors found. The other half comes from AI-exposed graduates taking jobs outside their field altogether, often in restaurants or retail. Employment for workers ages 22 to 24 in the most AI-exposed industries fell 12 percent over ten quarters after ChatGPT's release, and coauthor Lee Tucker attributes most of that to companies pulling back on entry-level postings rather than cutting existing staff. The Associated Press put the jobless rate for recent computer science and computer engineering graduates at 7.1 percent this month, well above the rate for graduates overall.

Jack Zemke felt that hiring freeze directly. The Tulane University computer science major sent hundreds of applications that mostly went unanswered before he showed an AI-powered search tool from a class project at an engineering conference. An employee at a civil engineering firm asked if he would work as an internal AI evangelist. Zemke took the job, stayed a year, then moved to an AI startup that summer. "These jobs are expecting you to be able to jump in and kind of operate on the frontier," he said. "There's so much churn."

A second data set, same pattern, different state

The Federal Reserve Bank of Dallas reached a similar conclusion independently, using Texas unemployment insurance records instead of Census data. Economists Samuel Dodini and Tucker Smith mapped O*NET job tasks onto what Anthropic's Claude can automate, then found that for every 10 percentage point increase in a major's automatable task share, a graduate's odds of finding work within a year fell 1.7 percentage points and wages fell 5 percent. Computer science, computer engineering, and language majors ranked most exposed. Nursing, education, and psychology ranked least exposed, matching the Census list almost major for major.

Students appear to be reacting to the data before economists finished writing it up. Undergraduate enrollment in AI-exposed majors at Texas schools fell 4.8 percent between fall 2024 and fall 2025 for every 10-point gap in automation exposure, the Dallas Fed found. Graduate-school enrollment among 2024 graduates from the most exposed majors rose 1.4 percentage points over the same period, and two-thirds of returning computer science students chose to stay in computer science at the graduate level rather than switch fields.

“The decline in hires is the primary cause.”

The economists who disagree

Not everyone accepts the AI explanation. A Stanford Review analysis points to a Federal Reserve study of more than 1 million firms, published in March 2026, that found no link between AI adoption and reduced job postings. A separate 2025 NBER paper covering 25,000 workers across 7,000 workplaces reported a near zero effect on earnings and hours, and replicated the same hiring slowdown at firms that had not adopted AI at all. Economists Daron Acemoglu and David Autor have found no aggregate employment effect from AI in the broader macroeconomic data.

The alternative story runs through interest rates. Pandemic-era near-zero rates pushed tech firms into a hiring binge starting in March 2020: Meta grew from 45,000 to 86,000 employees in three years, and Alphabet grew from 119,000 to 190,000. The fastest rate-tightening cycle in 40 years followed, forcing the same firms to raise their hiring bar sharply. A 2022 change to the tax code, which forced companies to spread software engineer salaries over five years instead of deducting them the year they were paid, made every new engineering hire more expensive on paper. Neither factor requires AI to explain a hiring slowdown.

What the rate-cycle story does not explain is why the pay and hiring gap tracks a major's AI exposure so closely rather than spreading evenly across graduates. Nursing and education majors faced the same tightening cycle and the same 2022 tax change as computer science majors, and neither the Census Bureau nor the Dallas Fed found a comparable drop in either field. The concentration by major, not the overall size of the drop, is the strongest piece of evidence against a purely rate-driven explanation.

Neither camp has the full picture yet. The Census and Dallas Fed papers show a correlation by major, not a controlled test isolating AI as the sole cause, and the counter-studies cited by skeptics measure firm-level AI adoption, a different variable than a major's underlying exposure to automatable tasks. Zemke's route out, an AI side project that got him noticed at a conference, may end up being the more durable data point than either working paper: the graduates treating AI as a resume line rather than a threat are the ones showing up in these numbers with jobs.