Concrete labor-market evidence of AI’s impact has finally arrived. The picture is more uneven than either AI optimists or pessimists like to argue. Fresh annual data from the Bureau of Labor Statistics shows that 18 occupations the agency previously flagged as exposed to AI saw a 0.2 percent drop in employment between May 2024 and May 2025. The broader US labor market grew 0.8 percent over the same period. Those 18 occupations together account for roughly 10 million jobs. NewsTrackerToday previewed exactly this divergence in coverage of last year’s BLS release, when the early signs of a split were already showing up in customer service and clerical categories.
Strip out the noise and the gap gets ugly. Excluding the fast-growing medical secretaries and assistants category, which is being pulled up by the healthcare boom, employment in the remaining 17 occupations fell 1.6 percent for a second consecutive year. The category-by-category breakdown is where the real story lives. Customer service representatives lost 130,180 jobs in the year through May 2025. That’s 4.8 percent. Credit authorizers, checkers and clerks are down 26.2 percent since May 2022, the last data point before ChatGPT went live. Broadcast announcers and radio disc jockeys have shed 20.8 percent of their workforce over the same window. Sales engineers are down 13.2 percent.
Ethan Cole, a macroeconomics and central banks specialist, kept it tight: “Clear divergence between AI-exposed and AI-insulated parts of the labor market. The aggregate effect on unemployment is still small. Composition of jobs is shifting sharply once you disaggregate. Central banks watching for AI-driven productivity gains have to read the next BLS releases more carefully than they’ve been doing. The labor reallocation is in the headline numbers now.”
BLS hedged its own list, which matters. When the agency published the 18-occupation grouping in late 2024, it noted that the list “should not be considered exhaustive or definitive”, just a set of examples where a reasonable expectation of AI-driven impact already existed. The data confirms a directional trend, not a settled causal story. And parallel research from the Federal Reserve Bank of New York has argued that job-posting declines in exposed occupations actually began before ChatGPT’s release, which complicates the AI-only narrative. NewsTrackerToday cross-referenced both the BLS and New York Fed datasets earlier this spring and ended up at the cleanest read: structural reallocation rather than a sudden AI shock.
Liam Anderson, who follows financial markets, came in from a different angle entirely: “Honestly, the cause barely matters. AI or general productivity. The investment implication lines up either way. Companies in highly exposed categories are pocketing the productivity gains instead of reinvesting in headcount. That improves operating leverage short term. Supports the equity valuations we have right now. But it raises a much harder question. Where does consumer demand come from once enough sectors make the same trade? The market hasn’t priced that second-order effect. Not even close.”
The political layer is starting to harden. Executives keep citing AI as the cause of layoffs in earnings calls and press statements, even when the underlying research suggests AI is not the main driver of the overall US labor slowdown. That mismatch is dangerous. It feeds public anxiety. Invites legislative responses. Complicates the story that AI will create as many jobs as it eliminates. NewsTrackerToday flagged the same pattern in March, when layoff-announcement language across the S&P 500 started converging on “AI-driven restructuring” as the preferred frame regardless of underlying cause.
Demographic effects compound the picture in unexpected directions. Workers in exposed occupations are disproportionately women and disproportionately college-educated. That inverts the historical pattern of automation hitting blue-collar work first. Research from the Federal Reserve Bank of Dallas has found that young workers in high-AI-exposure occupations are seeing slower inflows into employment rather than higher outflows. The entry door is narrowing even where layoff rates hold steady. Long-term consequences for early-career earnings and skill formation are still being mapped.
Companies overstating the AI cause for short-term reputational cover may end up on the receiving end of regulation written for a problem they themselves described. The trap is closing faster than many boards seem to notice. The next round of BLS data will probably accelerate that closure, not ease it. News Tracker Today already documented two state-level proposals that explicitly cite executive layoff statements as evidence of AI-driven harm. Similar drafts are moving in three more legislatures this quarter.
The numbers don’t settle the AI-and-jobs debate. But they end the era when the debate could be conducted purely in theory. The uncomfortable takeaway, the one nobody quite wants to say out loud: workers in exposed roles are losing ground in measurable ways. Employers are pocketing the productivity gains without rehiring at the same scale. Policymakers now have hard data they can quote in hearings. However the next two annual releases break, the question of who bears the cost of AI-driven productivity stopped being hypothetical the day this release dropped.
