Will AI take entry-level jobs? What the hiring data says

What the payroll data shows: the Stanford "canaries" study

• -What it is: Erik Brynjolfsson, Bharat Chandar and Ruyu Chen, "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence", Stanford Digital Economy Lab, August 2026 version.1
• -Data: monthly ADP payroll records for a fixed panel of firms, between 3.5 and 5 million employees a month, January 2021 to June 2026. Jobs are rated by how exposed their tasks are to AI, mainly with the "GPTs are GPTs" scores and the Anthropic Economic Index.1

Main findings

• -Employment of 22- to 25-year-olds in AI-exposed occupations "now stands 19% below where it would be had it kept pace with that of their less-exposed peers; experienced workers show no comparable gap."1
• -The gap is growing: it was 15% in the July 2025 data and 19% in June 2026.1
• -By June 2026, employment of 22- to 25-year-olds in the sample was about 2% below its November 2022 level, while every older age group had grown, by up to 11% for workers aged 35 to 40.1
• -It "operates primarily through reduced hiring of young workers rather than increased separations."1 Young people already in these jobs were not being pushed out faster. Fewer were getting in.
• -Declines are concentrated in jobs where people mostly use AI to automate tasks. Where AI mostly complements workers, employment is "flat or rising, especially for experienced workers."1

Caveats the authors stress themselves

• -They call the findings "early, descriptive indicators—canaries in the coal mine—rather than causal estimates."1
• -More- and less-exposed jobs were already moving differently before ChatGPT, especially around the pandemic, and the gap shrinks when they control for education.1
• -The pattern is weaker in national survey data. Over 2022 to 2024 the gap between the most and least exposed jobs was −0.022 in the American Community Survey (range −0.055 to +0.011, so not clearly different from zero) against −0.132 in the ADP sample. In professional, information and financial services the two sources agree closely (−0.213 and −0.231).1
• -The ADP sample over-represents larger firms and more AI-exposed occupations.1

What this means: the strongest evidence so far points to a narrowing entry door in some jobs, not mass firing. It is a signal worth watching, not proof that AI caused it.

College grads and AI: what the unemployment data says

The New York Fed tracks "recent college graduates" (ages 22 to 27 with at least a bachelor's degree, not enrolled in school) from Census and BLS survey data.2

• -For the second quarter of 2026 the NY Fed summarises: "The unemployment rate stayed elevated at about 5.6 percent, and the underemployment rate edged up to 42 percent."2 Underemployment means working in a job that typically does not need a degree.
• -Recent graduates are now above the overall rate. In the NY Fed series they have been above all workers in every month since January 2021. Before 2018 that was rare in the series, which starts in 1990. They are still below young workers without a degree.2
• -By major (2024 data, released February 2026): unemployment for recent graduates was 7.0% in computer science and 7.8% in computer engineering, against 2.1% in nursing, 1.2% in elementary education and 2.6% in accounting.2 These are one year of data and they don't track AI exposure neatly: accounting is an AI-exposed field with low graduate unemployment, while fine arts (7.7%) and anthropology (7.9%) are high.2
• -The Burning Glass Institute (July 2025) found that the unemployment rate of 20- to 24-year-olds with at least a bachelor's degree rose from 5.2% in 2018–19 to 6.2% in the two years to June 2025, while rates for people the same age with less education generally fell.4

Is it AI? New York Fed researchers Natalia Emanuel, Emma Harrington and Amanda Pallais (June 2026) estimate that remote work "can explain 64 percent of the recent increase in unemployment among young college graduates," because it makes training and mentoring new hires harder. They add that "the uptick in youth unemployment rates predates the rapid diffusion of AI."3 They also write that generative AI "may play a more primary role" from here.3 So this is one research team's estimate, set against Stanford's, not a settled answer.

What job postings show

Postings are a different lens: they show what employers ask for, not who gets hired.

• -Indeed Hiring Lab (July 2026): entry-level postings on Indeed have "been trending downwards since peaking in 2022, declining 7.5% year-over-year as of May 2026," while senior-level postings were up 14.7% year over year. Entry-level roles are still the biggest group: about 46% of postings, against about 14% for senior roles.5 As possible reasons Indeed lists AI, remote work, the Fed's rate hikes and companies slimming down after the pandemic hiring boom, and says "all of these factors may be at least partially at play."5
• -Where entry-level postings are scarce: in software development only 4.5% of postings were entry-level in the first quarter of 2026, against 91.3% in personal care and home health.5
• -More competition for each opening: in May 2026, 30% of applications to entry-level postings on Indeed came from people with 10 or more years of experience, more than any other group.6
• -Graduates are searching harder: the share of new bachelor's graduates who created or updated an Indeed profile in their graduation year rose 67% between 2023 and 2025. Internship postings ended 2025 at their lowest level since 2020, then picked up in early 2026.7
• -Burning Glass Institute (July 2025): the drop in entry-level postings (three years' experience or less) is "concentrated almost entirely in occupations with medium to high LLM exposure," while junior openings in low-exposure roles "have actually grown since 2022."4

Which entry-level jobs are most exposed?

No source gives a reliable list of "jobs AI will take". Here is what the sources above point to:

• -In the Stanford paper, the occupation case studies for the AI-exposed group include software development and customer service.1
• -Burning Glass's "Growth Roles", where it expects employers to need fewer juniors and more experienced staff: marketing specialist, project manager and financial analyst. Its reason: generative AI is good at "the kind of foundational work that junior employees traditionally perform: research, first drafts, basic data analysis, and routine communications."4
• -Software engineering for new grads: on Handshake, software engineer was a top-five posted role every year from 2018 to 2023 but fell to ninth in the 2024–25 school year.9
• -Less exposed, with steadier entry routes: health care and personal care roles, and fields with built-in placements such as nursing and teaching, where graduate unemployment is low.25 Low exposure is not a guarantee, and these jobs have their own pressures.

Exposure means AI can help with many of a job's tasks. It is not a forecast that the job will disappear. To look up one occupation, use our job lookup.

Gen Z, AI and jobs: what the surveys say

• -Handshake (class of 2026 survey, August 2025): 61% of the class of 2026 were somewhat or very pessimistic about their careers, and almost half of those said generative AI was at least part of the reason.8 Among computer science majors, 70% were at least somewhat pessimistic.9
• -But the same Handshake report found "not a clear trend toward steeper hiring slowdowns for job roles that are thought to be more exposed to AI" on its platform.8
• -Employers want AI skills: the share of full-time job descriptions on Handshake mentioning generative AI rose almost five times since 2023.8

Feeling worried is understandable. If it is getting heavy, our guide to AI job anxiety is about exactly that.

What about AI layoffs?

Challenger, Gray & Christmas counts job cuts that US employers announce and the reasons they give. Through September 2026, AI was cited in 120,136 announced cuts, about 21% of all cuts, the leading reason so far this year.10 These are announcements, not confirmed job losses, and Challenger does not break them down by age or experience. Most of the entry-level effect in the studies above shows up as hiring that didn't happen, which layoff counts can't see.1

What's still unclear

• -Cause. AI, remote work, interest rates, post-pandemic overhiring and a rising number of graduates all fit parts of the data.13458 No study has cleanly separated them.
• -Which data to trust. The payroll data (ADP) shows a strong pattern; national surveys show a weaker one.1 Postings data shows employer intentions, not hires.
• -The overall youth picture is mixed. BLS figures for 2026 are better than 2025 on some measures: the unemployment rate for 16- to 24-year-olds in July 2026 was 9.1%, down from 10.8% in July 2025.11 The rate for 20- to 24-year-olds was 8.0% in September 2026, against 9.2% a year earlier.12
• -What economists expect. In Indeed Hiring Lab's September 2026 survey of more than 120 US academics and economists, the panel leaned slightly towards AI-driven job loss becoming more likely for college-educated workers over the past year (a diffusion index of 55.6, where 50 is neutral; 48.1 for workers without a degree), and expected more downward pressure on college-educated workers' real wages from AI than a quarter earlier.13 That is a survey of expert opinion, not a measurement, and it is about all degree holders, not only new graduates.
• -Whether it lasts. The Stanford authors say they "do not make predictions about future impacts."1 Hiring could recover, or the gap could keep widening as it has since 2025.

Will AI replace entry-level jobs?

Our honest answer: the evidence so far says AI is changing entry-level work more than erasing it. In some AI-exposed fields, especially software, customer service and junior office and analysis roles, employers are hiring fewer young people than you'd expect, and the gap has grown.145 Most entry-level job postings still exist, and almost half of all postings on Indeed are entry-level.5 Whether this becomes a lasting shrinking of the first rung or a slow patch depends on things nobody can measure yet.