AI and Labor Market Demand: What 155 Million Job Postings Actually Show

See also (wiki): ai-talent-workforce-planning · ai-labor-relations

Executive Summary

• Behavioral data from 155 million U.S. job postings (2018–2025) shows no correlation between increased AI adoption and declining overall labor market demand. The fear-based narrative does not match the revealed-preference evidence.
• AI job postings grew 4x from 2022 to 2025 — from 0.28% to 1.13% of all postings — while overall job demand held above pre-pandemic levels. The sectors hiring the most AI-skilled workers are also growing the fastest overall.
• Entry-level job postings reached 12.6% of total postings in 2025, the highest in eight years (excluding the post-pandemic surge). AI is not eliminating the first rung of the career ladder.
• The Yale Budget Lab independently corroborates this: 33 months of Current Population Survey data post-ChatGPT shows “no sign” that AI exposure is correlated with employment or unemployment changes — at a pace matching prior technology disruptions, not exceeding them.
• The gap between fear and data is measurable: 18% of U.S. workers say AI is “very or somewhat likely” to eliminate their job in 5 years (Gallup, n=23,717, Q1 2026). The behavioral labor market data says they are wrong in aggregate — even as specific roles and tasks are genuinely being reshaped.

The Behavioral vs. Attitudinal Gap

Most AI workforce analysis is survey-based: ask workers how they feel about AI, ask executives what they expect to happen. The UMD/LinkUp AI Maps project took a different approach: analyze actual hiring behavior rather than stated intentions.

The dataset is among the largest available for this question: 155 million U.S. job postings from 2018 Q1 through 2025, sourced directly from employer career pages (not third-party aggregators). Lead researcher Anil K. Gupta of the University of Maryland Robert H. Smith School of Business describes the methodology as “revealed-preference” data — what employers actually did, not what they said they were planning to do.

The core finding is a direct challenge to the dominant narrative: “When you look at economy-wide data, a very different picture emerges. AI is not shrinking the labor market — it’s reshaping it.”

Three data points anchor this:

AI hiring is accelerating, not displacing. AI-skill job postings grew from 0.28% of all U.S. postings in 2022 to 1.13% in 2025 — a 4x increase in three years. The sectors with the fastest AI hiring growth are showing stronger overall job demand, not weaker. Companies integrating AI are hiring more, not less, because AI is expanding their operational capacity before it contracts their headcount needs.

Entry-level demand is at an 8-year high. Entry-level postings reached 12.6% of total postings in 2025, the highest share since 2017 (excluding the 2021–2022 post-pandemic hiring surge). This is the clearest available rebuttal to the specific fear that AI eliminates entry-level roles by automating the task profiles those roles cover. Software engineering — the sector most frequently cited as acutely vulnerable — maintains robust entry-level demand.

Aggregate demand is stable. No correlation was found between the period of accelerating AI adoption (2022–2025) and declining overall job postings. This does not mean specific occupations are not being displaced — they are. It means the aggregate economy is absorbing that displacement faster than pessimistic forecasts predicted.

The Yale Budget Lab Corroboration

The University of Maryland/LinkUp findings are not an isolated data point. The Budget Lab at Yale published an independent analysis in October 2025 (Martha Gimbel, Molly Kinder, Joshua Kendall, Maddie Lee) using Current Population Survey data across 33 months post-ChatGPT launch (November 2022 through July 2025).

Their conclusion: “Currently, measures of exposure, automation, and augmentation show no sign of being related to changes in employment or unemployment.”

The Yale team used Duncan and Duncan’s dissimilarity index with 12-month moving averages and compared post-ChatGPT occupational mix shifts against three prior technology disruptions: computers (1984–1989), the internet (1996–2002), and a control period (2016–2019). The pace of change is faster than historical precedent but remains modest relative to the disruption claims circulating in media.

Critically, they found worker distribution across exposure quintiles stable: roughly 29% low AI exposure, 46% medium, 18% high. This distribution has not meaningfully shifted in 33 months — the high-exposure workers who would theoretically show the most displacement are not showing measurably higher unemployment rates.

What the Fear Data Shows

The behavioral data does not mean workers’ concerns are irrational. Gallup’s Q1 2026 survey (n=23,717 U.S. employees, margin of error ±0.9pp) found:

• 18% of all U.S. employees say it is very or somewhat likely their job will be eliminated within 5 years due to AI
• Workers at AI-adopting organizations are more anxious, not less: 23% in AI-adopting organizations share this concern versus 18% overall
• 27% in AI-adopting organizations report their workplace has changed in disruptive ways (vs. 17% at non-adopters)

The Gallup data also shows the other side: 34% of workers in AI-adopting organizations report workforce expansion (vs. 28% at non-adopters), and 65% say AI improved productivity and efficiency. But the disruption signal is real: 23% of AI-adopting organizations are also seeing workforce reductions (vs. 16% at non-adopters).

The synthesis is this: AI is simultaneously creating net positive aggregate employment conditions AND creating real disruption for specific workers, roles, and functions. Both are true. The error is generalizing either direction — claiming AI will eliminate all the jobs, or claiming AI has no workforce impact.

Key Data Points

Source credibility note: UMD/LinkUp data is Tier 1 (April 2026) behavioral data with high methodological credibility — 155M postings from direct employer career pages, not scraped aggregates. Yale Budget Lab is Tier 2 (October 2025) independent academic analysis using government CPS data. Both are substantially more credible than vendor surveys or attitudinal polling for questions about actual labor market behavior. The Gallup data is Tier 1 (Q1 2026) and provides legitimate attitudinal signal that should not be dismissed — disruption is real even where aggregate demand holds.

What This Means for Your Organization

The executive communication problem on AI and jobs is harder than it looks. The honest message is neither “AI eliminates jobs” nor “AI creates all jobs.” It is: aggregate labor market demand is holding while specific workflows, roles, and task profiles are being reshaped — and the pace of that reshaping is accelerating.

For CHROs and COOs managing the workforce communication challenge, the UMD/Yale findings provide credible cover to shift the frame from elimination to redesign without being dishonest. The data supports it. Entry-level hiring is at an 8-year high. The sectors moving fastest on AI are growing fastest. The fear is real; the scale of displacement as measured in aggregate employment data is not yet matching the scale of the fear.

The more important implication is structural: 23% of AI-adopting organizations report workforce reductions alongside 34% reporting workforce expansion. Both are happening simultaneously. The organizations driving workforce expansion through AI are almost certainly doing what the BCG/McKinsey maturity research consistently finds: redesigning workflows before deploying tools, investing in reskilling before automating roles, and treating AI as a capacity multiplier rather than a headcount substitute. The organizations driving workforce reductions are doing the reverse.

The UMD behavioral data does not resolve that strategic choice — it just establishes that the aggregate labor market is absorbing it without a collapse in overall demand. That is meaningful context for a board conversation about workforce strategy, but it does not substitute for having the actual strategy.

If the labor market and workforce implications of AI deployment are live questions for your organization’s leadership team, I’d welcome the conversation — brandon@brandonsneider.com.

Sources

• UMD/LinkUp AI Maps Project — “Tribal Tales vs Hard Data: What Comprehensive Job Postings Data Reveal About Impact of AI on Labor Market Demand” (Anil K. Gupta, University of Maryland Robert H. Smith School of Business; April 17, 2026; n=155 million U.S. job postings, 2018 Q1–2025). Behavioral/revealed-preference data from direct employer career pages via LinkUp. Credibility: HIGH — largest behavioral job-postings dataset analyzed for AI impact; no commercial interest; transparent methodology. PR Newswire

• Yale Budget Lab — “Evaluating the Impact of AI on the Labor Market: Current State of Affairs” (Martha Gimbel, Molly Kinder, Joshua Kendall, Maddie Lee; October 1, 2025). Current Population Survey data, 33 months post-ChatGPT, Duncan and Duncan dissimilarity index. Credibility: HIGH — independent academic institution, government data source, transparent methodology. Yale Budget Lab

• Gallup Q1 2026 AI Adoption Survey — (n=23,717 U.S. employees; fieldwork February 4–19, 2026; margin of error ±0.9pp, 95% CI). Credibility: HIGH — large probability sample, established survey methodology. Gallup

Brandon Sneider | brandon@brandonsneider.com April 2026