Every few weeks a new dataset lands in the AI-jobs debate, and the honest answer about what it shows is usually “some of both”. This week’s entry is one of the strongest counterweights yet to the displacement headlines — and it comes from a national statistics agency, not a think tank.
Statistics Canada, the country’s official statistical agency, published a study last week (The Daily, 28 October) titled plainly: “Hiring and job separation rates in the era of generative artificial intelligence: Early evidence”. Its finding, as Canada’s The Canadian Press reported on November 3: “no broad evidence” that the rise of AI has led to substantial decreases in hiring or increases in job separation rates since ChatGPT launched in November 2022 — including for younger workers and for jobs the agency classifies as most amenable to AI substitution, such as data-entry clerks and coding-intensive occupations.
🔍 THE BOTTOM LINE: Youth hiring (ages 15-29) did soften between 2022 and 2026 — from 4.0% to 3.3% in low-exposure jobs and from 3.2% to 3.0% in high-exposure ones — but the declines appeared before ChatGPT’s release and in jobs AI least touches, so the study concludes AI is not the driver. Coding-intensive hiring fell more (2.5% to 1.9%), which the co-author attributes partly to pandemic over-hiring unwinding, not a new AI effect.
What the study actually measured
The Statistics Canada team, led by senior research economist Tahsin Mehdi, used labour force survey data for workers aged 15 to 64 covering January 2006 through August 2026 — roughly two decades. Rather than counting jobs, it tracked hiring rates and job-separation rates (which capture both layoffs and resignations), on the reasoning that those move earlier than employment totals: as the agency’s summary put it, the rates “may provide early signs of labour market shifts not captured by employment trends”. Each occupation was classified by two dimensions: exposure to AI and complementarity with it — whether AI enhances the worker’s output or substitutes for the tasks.
The results, three ways:
- Hiring and separations sit inside historical ranges. From the six months before ChatGPT’s launch to the six months before August 2026, rates were “generally within historical ranges or represented a continuation of pre-existing trends”.
- Job separation rates have stayed stable for five years across every age group and exposure category — AI-exposed workers are not being pushed out faster.
- Youth hiring softened everywhere. Young workers’ hiring rates declined regardless of occupational AI exposure, beginning before mass availability of ChatGPT.
“AI is no doubt reshaping workflows and redefining jobs right now,” Mehdi told The Canadian Press. “It’s just, it may not be eliminating entire jobs right now.”
Why the same data reads differently in the US
The Statistics Canada finding sits in tension with the American dataset this site has covered most: Stanford’s Digital Economy Lab “Canaries” analysis of ADP payroll data, which as of June 2026 pegged employment for 22-to-25-year-olds in the most AI-exposed occupations at about 19% below counterfactual — up from a 15% gap a year earlier. A Fortune opinion column by Mark Minevich, published November 3, cited the same figures (via Fortune) in arguing corporate America is “using it to stop hiring young people”, alongside Challenger’s tally that AI was cited in 120,136 of the 573,195 US job cuts announced through September — the year’s most-cited reason.
Three things reconcile the two, without either being wrong:
- Different labour markets. Canada’s economy carries different exposure weights — less employment in AI-heavy software and professional services, more in resource and construction work — so a US effect need not appear in Canadian data at the same size.
- Different measures. Stanford measures a counterfactual gap for one young cohort in one country’s payroll data; Statistics Canada measures rates against historical ranges for its whole market. A 19% relative gap can hide inside a 3.0% hiring rate that still looks normal by 2019 standards.
- Different windows. Statistics Canada’s series runs to August 2026 but benchmarks against 2006-2022 norms; the effect may still be building. The authors flag their work as early evidence and call for continued monitoring as adoption accelerates.
Viet Vu, manager of economic research at The Dais think tank at Toronto Metropolitan University, told The Canadian Press the findings match his own: “you just can’t detect AI’s impact on the Canadian labour market just yet”. His warning was about the road ahead, not the road behind — that AI threatens the “middle steps” of the school-to-work social contract, and statistics agencies by design measure what has happened, not what is about to.
The New Zealand read
New Zealand’s official statistics point the same direction as Canada’s — a soft youth market, no confirmed AI signature. Stats NZ reported a 5.6% unemployment rate in the June 2026 quarter, with youth unemployment and the 13.8% NEET rate drawing concern from NZIER, who described young people facing an AI “perfect storm” at the entry level. Meanwhile SEEK NZ’s job-ads data shows overall demand up 7.8% year on year with AI skill mentions at a record 4.2% of ads — demand growing and job titles shifting, exactly the pattern the Canadian study describes.
What Canada offers NZ is a method, more than a conclusion. A national statistics agency formally classifying every occupation by AI exposure and complementarity — and committing to re-run it as adoption deepens — is the template for answering the same question here with actual evidence. Stats NZ’s quarterly labour market releases are the natural place for that lens to land, and nothing equivalent exists yet in published NZ analysis. The honest state of the evidence in early November 2026: the measured effects so far are smaller than the announced fears, on the data available — and both governments and employers are making decisions ahead of that data, which is where the real risk sits.
❓ FAQ
Did Statistics Canada find AI is not affecting jobs at all? No. The study found no broad evidence of AI-driven falls in hiring or rises in separations through August 2026. It also found AI is reshaping workflows — co-author Tahsin Mehdi was clear that redefining jobs is happening — and called for continued monitoring as adoption accelerates.
Are young people still having a hard time getting hired? Yes — the study records declining hiring rates for workers aged 15-29 in both high- and low-exposure occupations. Its argument is about the cause: the declines predate ChatGPT’s release and appear in jobs AI least touches, pointing to economy-wide forces.
Why does this differ from the Stanford number showing a 19% decline? Different country, method and baseline: Stanford measures a counterfactual gap in US payroll data for 22-25-year-olds in exposed occupations; Statistics Canada measures hiring rates against 20-year historical ranges in Canada. The two are not directly comparable, and both teams note limits in their own evidence.
Is New Zealand likely to run the same analysis? No equivalent published analysis exists yet. Stats NZ publishes quarterly labour market statistics with occupation detail, which could support a similar exposure-classified view, but no such NZ study has been released.
— CJ Murden, editor of Singularity.Kiwi. Former digital technologies teacher, author of AI-focused books. Writing with a New Zealand focus.
