Economists at the Federal Reserve Bank of Cleveland published a new Economic Commentary on 9 November with one of the cleaner framings yet of what generative AI has actually done to the labour market so far: it has changed who gets in the door, not who gets pushed out of it. Alexander Cline and Barış Kaymak tracked workers aged 23 to 54 month by month through the US Census Bureau’s Current Population Survey, sorting 923 occupations into 20 groups by AI exposure, and comparing the period since ChatGPT’s launch (May 2025 to April 2026) against a pre-AI baseline of 2015 to 2019.
🔍 THE BOTTOM LINE: Unemployment has risen about one percentage point more in the most AI-exposed occupations than the least exposed — but the flows data shows the mechanism is slower hiring, not faster firing. Job-finding rates fell roughly six percentage points more in exposed occupations, implying average unemployment stretching from about 10.5 weeks to more than 13, while separation rates show no consistent AI-era change. The authors read it as a “no hire, no fire” market, and the gap is widest for workers aged 23 to 30 — with strong counterweights, including the fastest job growth sitting outside software and a rebound in US hiring for AI-heavy industries.
What the flows show — and what they don’t
The study’s headline is the unemployment gap, but the interesting machinery is underneath. When the authors decompose the change into two transition rates — how often workers leave jobs, and how quickly the unemployed land new ones — only one moves. As the paper states: “We do not find consistent evidence of elevated job separation rates among workers in occupations highly exposed to AI.” Job-finding, by contrast, fell “by roughly six percentage points more” in the most exposed occupations. Against an average monthly job-finding rate around 34 percent — meaning about a third of unemployed people find work each month — the shift implies average unemployment duration rising from roughly 10.5 weeks to over 13.
The age split sharpens the story. Moving across the exposure range raises predicted unemployment among 23-to-30-year-olds by about two percentage points; for workers 45 and older, at most one. The authors’ interpretation: younger workers compete for entry-level roles, exactly the positions where AI tools substitute for routine cognitive tasks first, so “AI’s disparate impact on younger workers is consistent with a gradual restructuring of occupations from the bottom up.”
Three caveats belong next to those numbers. The authors describe the work as an “early and necessarily incomplete look” at a long adjustment; the associations are observational, not proof of causation; and the CPS records an unemployed person’s most recent occupation rather than the one they are seeking — so the study undercounts first-time job seekers who have never held an exposed job.
The counterweights — and why “no hire, no fire” cuts both ways
The no-hire reading is genuinely two-sided, and the same week’s data supplies the other side. First, the door is slower all round, not only for AI-exposed workers — the Dallas Fed’s August 2026 analysis found lower employment for young workers in high-AI-exposure occupations was “mainly driven by a lower inflow” rather than by layoffs or separations, matching the Cleveland pattern. Second, the aggregate market’s problem is thin hiring, not mass dismissal: US jobless claims have been printing near historic lows for months, which we covered recently in our low-hire, low-fire labour market explainer. And third, the boom is real but displaced — as Lenny Rachitsky reported in his late-2026 tech job market survey, the biggest AI hiring boom “isn’t in software”, with tech jobs “concentrating” in AI-heavy regions even as software-specific lines stall.
There is a positive angle hiding inside the slower-door finding too. If AI adoption so far expresses itself as hiring hesitancy rather than redundancy letters, then the AI-era career risk is concentrated at specific moments — graduation, job change, re-entry after a break — rather than spread as constant background threat. Workers already inside exposed occupations have, on this evidence, kept their jobs. The Dallas Fed’s same analysis found experienced workers’ wages have been rising even as entry hiring shrank. The strategic career implication: the risky transitions are the planned ones, and planning earlier is worth more than panic later.
What it means for New Zealand
New Zealand’s labour market is running roughly a year behind the US cycle. Stats NZ has unemployment at 5.7 percent in the September 2026 quarter — an 11-year high, up from 5.6 percent the quarter before — and youth unemployment at 15.2 percent, nearly triple the national rate, as we covered in our NZ youth jobs evidence check. Cleveland’s mechanism travels cleanly into that context: in an economy where firms are already cautious, any AI-linked hesitation about entry-level roles lands on the group with the least experience to fall back on. The September quarter release shows hours worked edging up for the first time in nearly two years, an early cyclical green shoot — but the hiring-gate effect the Fed describes is about job composition, and that is the statistic NZ watchers should track next: not just how many young people are employed, but which entry occupations the vacancies sit in.
❓ FAQ
What does “no hire, no fire” mean? A labour market pattern where firms respond to AI uncertainty by slowing new hiring rather than laying off existing staff — using attrition and hiring freezes instead of redundancy waves. The Cleveland Fed researchers find the data consistent with this dynamic.
Did the Cleveland Fed study find AI is causing mass layoffs? No. It found no consistent evidence of elevated job separations in AI-exposed occupations. What changed is the pace at which unemployed people in those occupations find new work.
Which workers are most affected? Workers aged 23 to 30 in the most AI-exposed occupations show roughly a two-percentage-point higher unemployment gap than the least exposed; workers 45 and older show at most one point. Entry-level competition is the suggested mechanism.
Does this apply to New Zealand? The study is US-based and its authors call it early evidence. NZ-specific effects are not yet measured, but NZ’s own youth unemployment (15.2 percent, September quarter) and elevated general unemployment mean the hiring-gate mechanism compounds an already tight market.
📰 Sources
- Federal Reserve Bank of Cleveland — Job Finding and Unemployment in the Age of AI (Economic Commentary 2026-26, Cline & Kaymak, 9 November 2026)
- JBizNews — AI-Exposed Workers Take Longer to Find Jobs, Cleveland Fed Study Finds (10 November 2026)
- Federal Reserve Bank of Dallas — Young workers’ employment drops in occupations with high AI exposure
- Lenny Rachitsky — State of the tech job market: late 2026 (10 November 2026)
- Stats NZ — Labour market statistics: September 2026 quarter
— CJ Murden, editor of Singularity.Kiwi. Former digital technologies teacher, author of AI-focused books. Writing with a New Zealand focus.
