AI-Exposed Jobs Fell 0.5% in September; Gap Widest in Four Months

Employment in America's most AI-exposed occupations fell 0.5% in September 2026 from a year earlier, while employment in the least-exposed occupations grew 1.0%. [Fact] The gap between the two groups, 1.5 points, is the widest in the four months we have followed the Canaries Dashboard. [Estimate]

For workers aged 22 to 25, high-exposure employment fell 3.7%, the 36th consecutive year-over-year decline. Their employment in the least-exposed jobs rose 0.2%, after falling 2.0% in August. [Fact]

The September figures

The Canaries Dashboard is run by the Stanford Digital Economy Lab and ADP Research. It tracks payroll records of millions of workers in more than 730 occupations at tens of thousands of private U.S. employers and sorts occupations by an established measure of AI exposure. The September note, published October 21, 2026 by ADP chief economist Nela Richardson, reports seven numbers. [Fact]

All ages: high exposure -0.5%, least exposed +1.0%. Ages 22 to 25: overall -2.2% (August: -3.1%), high exposure -3.7%, least exposed +0.2%. Ages 26 to 30: high exposure -2.8%, least exposed +0.9%. ADP dates the young-worker streak to October 2023, twelve months after ChatGPT's public release. [Fact]

Four months in a row

Placed beside the June, July and August notes, the gaps between the high- and least-exposed groups run as follows. [Estimate — our subtraction of ADP's published rates]

• All ages: 0.8, 1.0, 0.8, then 1.5 points.
• Ages 22 to 25: 4.3, 3.0, 2.4, then 3.9 points.
• Ages 26 to 30: 3.5, 3.9, 3.2, then 3.7 points.

In August we set a test for this month. If least-exposed employment for 22-to-25-year-olds returned to flat while high-exposure employment stayed near -4%, August was a broad-market dip on top of a persistent AI-exposure penalty. If both stayed negative, young workers faced a general hiring freeze. September matched the first case: least-exposed +0.2%, high-exposure -3.7%. [Estimate]

The narrowing we described in August came from the comparison group falling. In September that group recovered and the exposed group did not, so the gap reopened by 1.5 points for the youngest band. [Estimate]

The exposed side did ease slightly from August in every band: by 0.1 points for all ages, 0.7 for ages 22 to 25 and 0.3 for ages 26 to 30. Across all four months, high-exposure employment for 22-to-25-year-olds has stayed between -3.0% and -4.4%. [Estimate]

What this means for exposed occupations

Richardson's June note placed software development and customer service at the high-exposure end and home health care at the low end; no later note names occupations. [Fact] The task profiles are on our pages for software developers, customer service representatives and home health aides. Matching ADP's exposure bands to those pages is our mapping. [Claim]

For a 22-to-25-year-old, September restores the contrast June showed. Employment for this age group is roughly flat in low-exposure occupations and 3.7% lower than a year ago in high-exposure ones. [Fact]

• Choosing a first job: in an exposed field, look for roles that carry client contact, sign-off on decisions or a licensed or physical setting, the parts of the task bundle a model does not do unsupervised.
• Ages 26 to 30 in an exposed role: your band has been 2.1% to 3.1% down for four months. Make the work your manager sees the part a model does not do.
• Reading next month's note: check which group moved before reading the gap.

Limits

The note publishes rates only: no counts, occupations, sectors, confidence intervals or revision history, so small month-to-month moves of a few tenths may be noise. The dashboard measures occupations rated as exposed and cannot show which employers adopted AI, and ADP's client base is not a random sample of U.S. employers.

Sources

• ADP Research, Nela Richardson, "Canaries Dashboard: Employment in AI-exposed occupations shrank in September," published October 21, 2026: https://www.adpresearch.com/research/canaries-dashboard-employment-in-ai-exposed-occupations-shrank-in-september
• ADP Research, "Canaries Dashboard: Employment in AI-exposed occupations slowed in August," published September 23, 2026: https://www.adpresearch.com/research/canaries-dashboard-employment-in-ai-exposed-occupations-slowed-in-august
• June and July figures as quoted in our earlier posts linked above.

About this analysis

This article was written with AI assistance from the full text of ADP's September note. Gaps and month-to-month changes are our subtraction of ADP's published rates and are marked [Estimate]. The mapping to our occupation pages is ours, not ADP's. ADP Research material is quoted briefly; no charts are reproduced.