The BaseResearchThe Budget Lab at YaleMay 7, 2026

What We Do and Don't Know About How AI is Affecting the Labor Market

The Budget Lab at Yale compares employment, wages and unemployment in AI-exposed occupations with a weighted group of unexposed occupations, using Current Population Survey data. Gimbel, Kendall and Nunn report no statistically or economically significant effect so far, and caution that exposure measures are imperfect, the survey is thin for narrow groups, and the picture could change quickly.

Open the source ↗

Why it matters

The same thing, for different readers.

What it means for you

Jobs most open to AI show no significant change in employment or pay against a matched group of other jobs so far, though the authors say that could change quickly.

For lawmakers

This analysis gives lawmakers no clear AI effect on jobs or pay to cite yet; the Budget Lab says it will update as new labor market data arrive.

For researchers

Exposure scores are informed guesses that can average away opposite effects, the design needs AI's arrival separable from events like remote work, and the survey is weak for narrow groups.

In their own words

Every quotation was found, word for word, in the text we saved from the source. That shows the words are there; it does not make them true, or prove the speaker meant them as quoted.

From the source’s own text.

“Using an approach called synthetic differences-in-differences (SDID) that addresses these challenges, we generally find no statistically or economically significant effects as of yet.”

Martha Gimbel, The Budget Lab at YaleThe authors' headline result, from the opening paragraphs: employment and wages in AI-exposed occupations compared with a weighted group of unexposed ones.

Found word for word in the source

“These figures indicate positive effects in the most recent quarter—roughly half a percentage point increase in the entire sample, and more for the 16-34 year old subsample—but both are statistically insignificant as of the first quarter of 2026.”

Martha Gimbel, The Budget Lab at YaleFrom the main text, on the appendix figures for the unemployment rate in exposed occupations, for all workers and for ages 16 to 34.

Found word for word in the source

“This means that analysis of exposure metrics could actually hide effects if (for instance) half of “exposed” occupations see their jobs expand and half see their jobs contract—the average could be zero.”

Martha Gimbel, The Budget Lab at YaleThe authors explain why occupation-level exposure scores may miss real effects, which is why the Budget Lab's own tracker relies on other measures.

Found word for word in the source

“First, LLMs are constantly improving, such that the economic impacts from early models seem likely to be smaller than those of recent models.”

Martha Gimbel, The Budget Lab at YaleThe first of the caveats the authors list on their main finding.

Found word for word in the source

“If labor market effects of AI are currently limited to a narrow slice of the workforce, other datasets and research designs would be better suited to identify them.”

Martha Gimbel, The Budget Lab at YaleThe authors' closing caveat about what the Current Population Survey can detect in narrow groups such as recent graduates.

Found word for word in the source

How the finding was reached

In our words, from the source. Weigh the finding by how it was made.

What was done: Synthetic difference-in-differences on Current Population Survey data (2020 omitted). Occupations in the top third of the Budget Lab's AI-exposure ranking are compared, from ChatGPT's release in the fourth quarter of 2022 onward, with a weighted mix of occupations in the bottom third. Outcomes: employment share of the civilian population, median real hourly wages, and, in the appendix, unemployment rates, including ages 16-34.
What it found: No statistically or economically significant effect of AI exposure on occupational employment or real hourly wages. For the latest quarter's employment share, the point estimate is virtually zero and the top of the 95% confidence interval is about 5% of the average exposed occupation's share (just above 0.16% of the civilian population). The estimated effect on unemployment in exposed occupations is positive in the latest quarter, roughly half a percentage point (more for ages 16-34), but not statistically significant as of the first quarter of 2026.
What it does not show: Exposure scores are informed guesses and do not mark jobs at risk, so averaging could hide opposite effects within the exposed group. The method assumes AI's arrival can be told apart from other changes such as remote work, and the authors say impacts of early models seem likely to be smaller than those of recent ones. The Current Population Survey is underpowered for narrow groups: among 22-27-year-olds, few occupations met the 50-observations-per-quarter rule. Unemployment estimates by occupation are imprecise.
Status: A report

The record

Questions: What happens to work, income and the economy?
People: Martha Gimbel (Wrote it, The Budget Lab at Yale) · Joshua Kendall (Wrote it, The Budget Lab at Yale) · Ryan Nunn (Wrote it, The Budget Lab at Yale)
What we read: We read the whole source.
Found by: A research pass: work, income and the economy, gathered October 13, 2026, via research-work (tier 1: the source’s own channel or an official record)
Cite this: The Budget Lab at Yale, “What We Do and Don't Know About How AI is Affecting the Labor Market”, May 7, 2026. Entry in the Ctrl AI Base, https://ctrlai.com/base/budget-lab-2026-05-07-ai-labor-market-effects

Our notesFull article and appendix read from the page. The page names three authors and gives no individual roles; quotations are attributed to the first-named author on behalf of all three, and the title shown is the Budget Lab as the site gives it. The page says the Budget Lab will update the assessment, so check for a later version. The research pass's excerpt also mentions a Budget Lab tracker (updated 15 September 2026) and an October 2025 baseline report; neither is part of this page and neither is described here. Found no earlier Base entry that this clearly bears on.