Not broadly yet: occupational mix in the fastest AI-adopting industries has moved only modestly, but early-career hiring has already shifted

Summary

Mostly not yet, at the level of which occupations an industry employs overall — but yes, in a narrower and now well-replicated way, for who gets hired into the entry-level rungs of the most AI-exposed jobs. Researchers who built the exact metric this question asks for — the Yale Budget Lab's occupational-mix dissimilarity index, using CPS data — find that Information, Financial Activities, and Professional & Business Services (the highest-genAI-exposure sectors) have seen somewhat larger shifts in their occupational composition since ChatGPT's November 2022 launch than the economy overall, but conclude the shift predates ChatGPT and has not measurably accelerated because of it. At the same time, three independent datasets with three different research teams — ADP payroll data (Brynjolfsson, Chandar & Chen), Census Bureau Quarterly Workforce Indicators (Tucker, 2026), and Anthropic's own Economic Index analysis of CPS hiring flows — now agree on a real, already-visible compositional shift: early-career hiring (ages 22-25) into the most AI-exposed occupations has fallen sharply (estimates range from about 9% to 16% relative decline) since late 2022, even as total sector employment keeps growing and experienced workers in the same occupations are unaffected or growing. Separately, Johnston & Makridis's QCEW-based analysis of nearly all U.S. employers finds that AI-exposed sectors are growing employment overall (+3.9% per one-standard-deviation increase in exposure through 2024), driven by task-complementary (augmentative) rather than fully automatable exposure — consistent with a labor market that is reallocating workers within growing sectors rather than shrinking headcount. So the honest answer: a broad reshuffling of which occupations these industries employ has not yet clearly started (and what movement exists is not clearly attributable to AI rather than pre-existing trends); a narrower, real, and increasingly well-documented shift — fewer entry-level slots specifically in automatable AI-exposed roles — has already started and is growing.

What "fastest-adopting industries" means here, and why the question splits in two

The literature identifies AI-exposure/adoption leaders consistently: Information (software, data processing, media), Financial Activities, and Professional & Business Services rank highest on both theoretical exposure (OpenAI's task-based measure, Eloundou et al. 2024) and observed usage (Anthropic's Claude usage data via its Economic Index, and OpenAI's own workplace usage-pattern data). Johnston & Makridis's industry-by-state exposure mapping and the Census Bureau's QWI paper (Tucker, 2026) both independently identify "four main sectors" concentrated with the highest AI-exposed industry-state cells, matching this set.

Two distinct questions get conflated in casual discussion, and the research is careful to separate them: (1) has the mix of occupations employed within/across these industries changed — e.g., relatively fewer clerical or customer-service workers, relatively more data scientists; and (2) has total employment in these industries changed, and for whom. The evidence differs sharply between the two.

Finding 1 — Broad occupational mix: moving a little faster than history, but not clearly attributable to AI

The Yale Budget Lab (Gimbel, Kinder, Kendall, Lee) built an occupational-mix dissimilarity index from CPS microdata comparing the distribution of workers across occupations at a point in time to a baseline — precisely the metric this question needs. Economy-wide, occupational mix has moved only about 1 percentage point faster since ChatGPT's Nov-2022 launch than during the internet's mid-1990s rollout: a small acceleration, not a dramatic one.

Applied by industry (their Figure 3), the Information, Financial Activities, and Professional & Business Services sectors do show larger occupational-mix shifts than the aggregate economy since November 2022, with Information showing the largest of the three. But extending the window backward, the authors find these same industries' mix shifts were already underway well before ChatGPT existed, and Information in particular has historically been a volatile-mix sector for reasons unrelated to any one technology (its own broader history shows comparable swings before generative AI existed). Their stated conclusion: "there is nothing meaningful we can either attribute or misattribute to AI" in this measure so far. A complementary check — the share of workers in low/medium/high genAI-exposure occupational quintiles — has stayed flat (roughly 29%/46%/18%) since ChatGPT's release, with no rising concentration of AI exposure among the unemployed either.

Anthropic's own analysis (Economic Index labor-market report, March 2026) reaches a similar top-line conclusion using a different design: comparing unemployment-rate trends for the most- vs. least-exposed occupation quartiles in CPS data, the gap has moved only slightly and is not statistically distinguishable from zero since ChatGPT's release.

Finding 2 — Total employment in the fastest-adopting industries is still growing, and growing where AI is a complement

Johnston & Makridis (working paper, using QCEW/LEHD administrative data covering over 95% of U.S. employers, 2017-2024, and a design comparing sectors within the same state and year that differed in pre-pandemic occupational composition) find a one-standard-deviation increase in AI exposure raises sector output by roughly 10% by 2024, employment by about 3.9%, and the aggregate wage bill by about 4.8% — effects that began in 2021, before ChatGPT, coinciding with enterprise tools like GitHub Copilot. This splits sharply by task type: where AI augments workers (requires human collaboration), employment rises about 4% and wages follow; where AI can operate independently (fully automating a task), employment shows no significant effect and total wages fall slightly. The authors' framing: "AI is not destroying jobs" in aggregate, but workers are capturing only about 29 cents of every dollar of AI-driven output growth, with the rest flowing to profits — and the positive employment effects are concentrated in labor markets with high pre-AI worker turnover, implying reallocation of workers into new roles ("coders may become AI architects") within these growing sectors, not stability of the existing occupational mix.

Finding 3 — The one shift now confirmed across three independent datasets: entry-level hiring

Three separate research teams, using three non-overlapping data sources, now find the same pattern for young workers specifically:

• ADP payroll microdata (Brynjolfsson, Chandar & Chen, Stanford Digital Economy Lab, Nov 2025): workers aged 22-25 in the most AI-exposed occupations (software developers, customer-service representatives) saw roughly a 16% relative employment decline since late 2022 versus the least-exposed quintile, controlling for firm-time shocks; in absolute terms, about a 6% decline for this age group in the most-exposed jobs versus a 6-9% increase for older workers in the identical occupations. The effect concentrates in occupations where AI use is "automative" rather than "augmentative," and survives excluding tech firms and remotable occupations.
• Census Bureau Quarterly Workforce Indicators (Lee Tucker, CES-WP-26-27, April 2026): using matched employer-employee administrative data crosswalked from occupational AI exposure to detailed industry-state cells, early-career (22-24) hires fell immediately and persistently after ChatGPT's release in the most AI-exposed quintile of industry-state cells; regression-adjusted employment for this group fell about 12% over the following 10 quarters (over 150,000 jobs), driven almost entirely by reduced hiring rather than increased separations, with only partial (up to about a quarter) of the gap attributable to contemporaneous monetary-policy tightening rather than AI.
• Anthropic Economic Index (CPS panel data, March 2026): while unemployment rates for exposed young workers are flat, the monthly rate at which 22-25-year-olds start new jobs in the most-exposed occupations fell about 14% relative to 2022 levels, a decline visible starting in 2024 and not present among workers over 25 or in less-exposed occupations.

This is real, replicated, and growing evidence of a shift within AI-exposed occupations — but it is a shift in who holds the job (age/experience), not (yet) clear evidence of a shift in which occupations a sector employs overall. Independent studies using LinkedIn data in the US and UK (Hosseini & Lichtinger 2025; Klein Teeselink 2025) find similar entry-level declines, though a Denmark-based study using different labor-market institutions (Humlum & Vestergaard 2025) found minimal effects on entry-level earnings or hours there, underscoring that the pattern is not universal across labor-market institutions and remains actively debated.

What AskAmerica's own data could and could not add

The corpus was searched for a table at this question's exact grain — occupation-by-industry employment tracked over time (BLS OEWS industry staffing-pattern matrices, or CPS/QWI microdata by occupation and industry) — and none exists here. econ.state_occupation_employment (OEWS) covers only 3 broad SOC groups, cross-industry, and a data-coverage scan confirmed it holds only a single year (2025), so it cannot show a mix shift over time. census.acs_industry covers industry-of-employment (not occupation-within-industry) at 13 broad sectors, but is a 5-year rolling ACS average declared only through 2023 and lags too far behind ChatGPT's Nov-2022 launch to isolate a post-2022 shift on its own. Neither table can independently verify or extend the CPS/ADP/QCEW/QWI findings above; this answer rests on the primary research cited rather than a warehouse computation, and that gap is disclosed here and on the dashboard rather than being papered over with a coarser proxy.