Shaduf.Research preview
Will AI Take My Job?/AI-exposed jobs posted higher pay. Their entry-level share still collapsed.

Research update · 25 September 2026

AI-exposed jobs posted higher pay. The entry-level share still collapsed.

Indeed’s new U.S. analysis finds faster advertised-pay growth in occupations with high generative-AI exposure. It also finds that entry-level ads became a much smaller share of openings in those occupations. The result changes how we read wage headlines. It does not show that AI caused the pattern.

What was measured

Indeed Hiring Lab analyzed millions of U.S. job postings that reported a salary. It sorts occupations using a score for skills that generative AI could transform. It tracks posted salaries and the seniority stated in ads from 2021 to mid-2026.

This is advertised labor demand. It does not measure a worker’s paycheck, an accepted offer, total employment, layoffs or a firm’s actual AI use.

The pay result and the entry result point in different directions

Advertised pay in the most AI-exposed occupations rose about 46% from 2021 to mid-2026, versus 25% in the least-exposed group. In the authors’ post-ChatGPT models, the estimated premium is 5.7% with occupation controls and 4.7% when a job title is compared with its own earlier ads.

But the mix of openings changed sharply. In the most-exposed group, entry-level ads fell from 29% to 10% of postings. Senior ads rose from 22% to 47%. A reader who sees only the pay figure can miss the more immediate question: who is being recruited into the better-paid work?

Why the causal claim remains weak

A shift toward senior openings can lift average advertised pay without raising pay for comparable workers. When Indeed holds seniority mix constant inside occupations, its estimated premium narrows to 2.4% and is not statistically significant. The exposure index also measures potential task transformation, not whether a named employer deployed AI.

Industry cycles, changes in occupation mix and demand for experienced workers can contribute to the pattern. The result is a useful structure-of-work signal, not a causal estimate of an AI wage gain or a count of junior jobs eliminated.

How this fits the rest of the evidence

Texas administrative records find lower first-year employment and earnings for graduates from more AI-exposed majors. A separate Census study finds weaker initial employment and earnings in highly exposed majors. Those results are also exposure-based, but they measure outcomes after entry rather than job ads.

The new Indeed result does not replicate either study. It makes their concern more legible: higher posted pay can coexist with fewer junior opportunities. At the same time, Yale’s broad U.S. tracker has found no clear economy-wide AI footprint through August 2026. The evidence does not establish a general wage collapse or a mass layoff wave.

What would change the conclusion

  • Employer-level studies that observe AI rollout, posted pay, offers, accepted wages, hiring and seniority together.
  • Evidence that the junior-share change remains after accounting for remote work, industry cycles and other non-AI changes.
  • Longer follow-up that shows whether a thinner entry route leads to lower employment, wage loss, different job matches or later catch-up.

Search published pools, pages, reports, and evidence.