AI Slows Hiring for Young Workers, Especially Women

Generative AI is changing entry points, not headcount

Generative AI is not eliminating jobs in the aggregate, but it is quietly reshaping who gets hired—and the shift is hitting the youngest workers hardest. A new working paper by Magnus Lodefalk, Lydia Löthman, Michael Koch, and Erik Engberg uses full-population Swedish register data and 4.6 million job advertisements to show that while total employment in AI-exposed occupations has held steady, hiring of workers aged 22-25 has fallen sharply since late 2022. Young women, concentrated in administrative and customer-service roles with high AI exposure, face the steepest declines.

The research, first reported by ProMarket, draws on monthly employer tax declarations covering Sweden's entire workforce. By tracking the same individuals month to month, the authors can distinguish new hires from departures and compare how employment in AI-exposed occupations evolved relative to less-exposed roles within the same employer. By the first half of 2025, employment of 22-25 year olds in the most exposed occupations had fallen 5.5 percent relative to less-exposed roles in the same firms. Workers aged 31-49 saw essentially no change, and those over 50 showed a small gain.

The adjustment happens through hiring freezes, not layoffs

The decline operates almost entirely through reduced hiring rather than increased separations. For workers aged 22-25, the drop in new hires is large and precisely measured, while layoffs and quits changed only about one-quarter as much. Swedish employment law enforces a last-in-first-out principle that constrains dismissals of incumbent workers, so employers are not clearing out junior staff. Instead, they are slowing entry: not replacing departing workers and not bringing in new cohorts at the previous rate.

Recent U.S. evidence from Seyed Hosseini and Guy Lichtinger, covering roughly 62 million workers across 285,000 firms, shows the same pattern. Junior employment falls sharply in firms adopting generative AI, while senior employment remains stable. The hiring-freeze mechanism appears to cross institutional boundaries.

Young women face disproportionate exposure

Women make up about 54 percent of the 22-25 age group in the most AI-exposed occupations, driven by occupational sorting. Payroll administrators are 82 percent female and among the most exposed roles in Sweden; customer service agents and receptionists are 66 percent male. Software development is highly exposed but 80 percent male and represents a smaller share of the affected group.

Employment in high-exposure occupations fell about 1.6 percent for young women, roughly twice the 0.7 percent decline for young men. About two-fifths of that gap reflects occupational concentration—young women are simply more often in the jobs AI reaches first. The International Labour Organization estimates that female-dominated occupations globally face almost twice the generative-AI exposure of male-dominated ones.

Why it matters

Entry-level positions are not just jobs; they are training grounds where workers build applied skills, professional networks, and tacit knowledge that become mid-career capability. When employers stop bringing in new cohorts, they weaken the pipeline of experienced workers they will need later. If the hiring slowdown persists, the result may be a generational skill-formation gap—fewer workers learning by doing, fewer junior employees moving into senior roles, and a thinner future stock of occupational expertise. That gap will show up in career trajectories, firm capabilities, and productivity growth long before it appears in headline unemployment figures.

The research also clarifies that the broad decline in job postings since 2022 tracks monetary tightening more closely than the launch of ChatGPT. Swedish job postings began falling in April 2022, immediately after the Riksbank's first rate hike and seven months before ChatGPT's release. The additional decline after November 2022 was statistically imprecise and not robustly different from zero.

The findings were detailed in a working paper by Lodefalk, Löthman, Koch, and Engberg, with funding from the Torsten Söderberg Foundation and WASP-HS, and first reported by ProMarket.