Two Answers

A year earlier that gap was 15 percent. Experienced workers show no comparable gap.

The lab's sentence, on its page of 12 August 2026: "Employment among workers ages 22–25 in highly AI-exposed occupations now stands about 19% below where it would be if it had kept pace with employment among similarly aged workers in less-exposed occupations. Experienced workers show no comparable gap." And: "the shortfall was 15% at the July 2025 data vintage and is 19% as of June 2026". The 19% is a shortfall against a kept-pace path, not a fall in jobs: in levels, the page's footnote says, employment of the age group in the two most exposed quintiles fell about 11% between November 2022 and June 2026, while in the three least-exposed quintiles it grew about 10%. The paper's data: ADP payroll records "covering millions of U.S. workers", between 3.5 and 5 million employees a month in the analysis sample.

The page's sentences, whole: "These are descriptive patterns, not causal estimates of the effect of AI." "The gaps between more- and less-exposed young workers shrink when we account for education." The headline: "No Widespread Displacement, but the AI Employment Gap for Young Workers Has Widened to 19%". And the bridge to Yale: "The estimated gaps are also larger in the ADP analysis sample than in national survey benchmarks."

But, it adds, unemployment growth for AI-exposed workers "has tended to be elevated" against comparable unexposed workers since late 2022. In the third quarter of 2026 the difference was +1.1 percentage points.

Yale's own words for it are "volatile", to be "interpreted cautiously".

The update of 14 October 2026, whole: "The Budget Lab's labor market analysis, updated to incorporate September 2026 CPS microdata, does not provide clear evidence of labor market disruption associated with AI." Then: "the growth in unemployment of AI-exposed workers has tended to be elevated relative to growth for comparable unexposed workers since late 2022, and this is the case in the third quarter of 2026. These estimates are volatile and should be interpreted cautiously, particularly as they are not accompanied by other indications of labor market impacts."

The tracker's own data file carries the comparison: the log unemployment rate of AI-exposed occupations against a synthetic unexposed match, and their difference by quarter. In the third quarter of 2026 the difference is +1.14 percentage points, with a 95% interval from −0.19 to +2.48, so it includes zero; it has been positive in 15 of the 16 quarters since late 2022 and outside zero in four of them.

Stanford's exposure measure is Eloundou et al. (2024), in quintiles of occupations; "highly exposed" is the two most exposed quintiles (the page's footnote 1), which hold about two-thirds of the jobs in its sample. Yale's comparison takes the top third of occupations by the Budget Lab's own summary metric as exposed and the bottom third as unexposed, across all ages; its separate chart of workers by exposure level, which uses a top fifth, describes shares and is not the comparison. Stanford reports all ages 22 to 70 as well, but its headline is the 22-to-25 group.

Yale sees unemployment rising a little faster for exposed workers, but calls it volatile, and finds no clear disruption.

Stanford: "We do not see widespread, economy-wide job displacement associated with AI." Yale: the churn and the comparisons "are not consistent with any large-scale disruption to date." Stanford HAI's piece of 21 October reports on the same Stanford dashboards by another measure (employment for 22- to 25-year-olds in software development and customer service "has dropped over 3% annually since ChatGPT's release in November 2022, with the decline sharpening in recent months to 4.4%"); it is the same data, not a third study, and is not set against the 19%.

Every figure from Stanford's and Yale's own releases. Token Runner dot A I.