For years, arguments about AI and jobs have run ahead of the evidence. A company announces layoffs. A headline adds “AI-driven.” Someone else points to a healthy employment number and says the fear is overblown. Everybody walks away with the answer they brought with them.
California just made that argument a little harder to conduct carelessly.
A monthly reality check
On July 23, the California Employment Development Department and the nonpartisan California Policy Lab at the University of California launched the AI-Unemployment Tracker. It pairs state unemployment-insurance claims with measures of how exposed different occupations are to AI. The public dashboard can also sort the results by education, age, industry, region, race and ethnicity, and gender.
And it will update every month. That matters. One release can become a headline. A recurring public record lets people watch what changes.
Quiet statewide, sharper underneath
The first statewide finding was fairly quiet. Researchers found no statistically significant post-2022 increase in the share of claims coming from AI-exposed occupations.
Look underneath that average, though, and the picture gets more complicated. The release reported elevated patterns among college-educated workers, in the San Francisco Bay Area, and in the Information and Professional Services sectors.
That does not prove AI caused those layoffs. The tracker groups claims using occupational exposure scores. It cannot tell us why a particular person lost work. A calm statewide result can still exist alongside real pressure in a specific region, industry, or type of job.
The people outside the frame
The dashboard also cannot see everyone. Unemployment-insurance records miss people who never file, workers who are ineligible, many self-employed and gig workers, some people who find work quickly, and people who leave the labor force. Occupations are self-reported. Small groups may be hidden to protect privacy, and the figures can be revised.
So this is a better window into the AI-jobs question. It is still a window, with edges.
The next time a headline says AI is taking jobs, or insists that it is not, ask one question: What does the evidence actually measure: occupational exposure, observed AI use, unemployment claims, worker displacement, or proof that AI caused the job loss?