The 19 percent everyone is quoting does not mean what the headlines say

Erik Brynjolfsson, Bharat Chandar and Ruyu Chen published the August 2026 edition of “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence,” built on ADP payroll records covering millions of US workers through June 2026. One number from it is travelling fast, and it is being quoted in a shape the paper does not use.

What the number actually says

The paper's second fact reads: employment of young workers aged 22 to 25 in AI-exposed occupations now stands 19 percent below where it would be had it kept pace with that of their less-exposed peers. Experienced workers show no comparable gap.

That is a counterfactual, not a headcount. It does not say young people in those jobs are 19 percent fewer than their peers. It says the gap between the two groups' growth paths has opened to 19 percent since the two were moving together.

Two measures, one comparison

Coverage of the update, including Ars Technica's, sets 19 percent against a figure of 13 percent from last year. The paper is explicit that these are different measures. Earlier versions headlined regression estimates adjusting for firm-level shocks: 13 percent at the July 2025 data vintage, 16 percent at the September 2025 vintage.

The authors now emphasize a simpler descriptive divergence that requires no modeling choices — and by that measure the shortfall was 15 percent at the July 2025 vintage and has widened to 19 percent as of June 2026. Fifteen to nineteen is the like-for-like move. Thirteen to nineteen mixes a modeled estimate with a descriptive one and makes the change look half again as large.

The other five facts point elsewhere

The 19 percent is one of six findings, and the rest are quieter than the headline. The paper finds no evidence of widespread, economy-wide job displacement. The divergence works primarily through reduced hiring of young workers, not increased separations — nobody is being pushed out, fewer are being let in.

Declines concentrate in occupations where AI usage substitutes for human tasks; where it complements workers, employment is flat or rising, especially for experienced workers. Adjustment shows up in employment rather than in base pay. And the divergence persists when technology firms and computer occupations are excluded, which rules out the easiest objection to it.

• No economy-wide displacement.
• Young workers in AI-exposed jobs: 19 percent below a kept-pace path; experienced workers unaffected.
• The gap has widened steadily since first documented in August 2025.
• It runs through hiring, not firing.
• Substitution hurts; complementarity does not.
• Employment adjusts, base compensation does not.