Is AI Taking Jobs From Young People? What the Payroll Data Shows

Is AI taking jobs from young people? The clearest evidence available says yes, in one narrow place: workers aged 22 to 25 in the most AI-exposed occupations show a real, measured employment gap against similarly aged peers in less-exposed fields, while workers who already have a few years in those same occupations show no such gap at all.

That distinction comes from Stanford Digital Economy Lab's "Canaries in the Coal Mine?", a study by Erik Brynjolfsson, Bharat Chandar and Ruyu Chen built on ADP payroll records rather than a survey or a job title rated for how automatable it sounds. Actual employment counts, tracked over time, are a harder thing to argue with than an opinion poll.

What the payroll data actually shows

Employment among 22 to 25 year olds in highly AI-exposed occupations now sits about 19% below where it would be had it kept pace with similarly aged workers in less-exposed occupations, as of the June 2026 data. That shortfall was 15% a year earlier, at the July 2025 data vintage, so the gap is widening, not holding steady.

Why the gap sits with hiring, not with who already has the job

The researchers state plainly that experienced workers show no comparable gap. The people already doing exposed work are not being pushed out in the data. What moved is reduced hiring of young workers rather than increased separations, the study's own words for the mechanism. Employers are not emptying desks. They are filling fewer of the entry seats that used to sit next to them.

That is a hiring story, not a layoff story, and the two call for different responses. A layoff wave would mean nobody is safe once hired. A hiring gap means the door into the field is narrower than it used to be, which is a different problem with a different fix.

What "AI-exposed occupation" means here

The study sorts occupations using an established task-exposure measure, not a guess dressed up as data. Our own explainer on how job risk scores get built covers how that kind of classification usually works and where it can go wrong. This is a different dataset than the Anthropic usage study we covered separately, which matched real chat conversations to occupations and found exposure concentrated in higher paid office work. The two studies use different methods and different data, and they agree on the same underlying shape: exposure sits with desk work, not service work, and the researchers here found no widespread, economy-wide job displacement associated with AI once you look past the youngest entrants.

Why the youngest workers carry it

The likely mechanism is not mysterious. A junior hire has historically built a track record on exactly the routine, well-defined tasks a language model now handles cheaply, the first draft, the standard summary, the boilerplate analysis. An experienced person already has years of judgment calls on record to point to. A 23 year old does not, and the tasks that used to be the proving ground are the ones most exposed. That gap in provable track record, not a gap in ability, is the more precise way to state the problem.

What this changes if you are early in an AI-exposed field

The honest fix is not waiting for a hiring market that used to exist to come back. It is closing the track-record gap directly, through checkable proof rather than years alone. Our piece on why entry level jobs keep asking for experience covers the broader hiring pattern behind that, and our rundown of how to prove a skill to an employer covers the specific moves: a work sample, a documented judgment call, a scored assessment. If you want a read on your own six domains of judgment rather than an occupation average, the check is built for exactly that gap.