Research update · 17 September 2026
Texas just broke the “no AI hiring signal” story. It did not prove layoffs.
A Dallas Fed analysis finds that Texas online job postings fell more in occupations with more tasks that generative AI can automate. That is a sharper hiring warning than a forecast. It is still a Texas posting result built around exposure, not a national count of jobs lost to AI.
What the study measured
The authors link millions of Texas Lightcast job postings to an occupation-level task score based on tasks Claude users actually perform. They compare more- and less-automatable occupations within the same industry after ChatGPT’s release. That is an observed demand signal in ads, not a record of which Texas employer deployed a particular AI system.
The pullback is large enough to matter
By the first quarter of 2025, postings for more-exposed positions were about 8% lower relative to less-exposed positions. In a balanced panel, incumbent firms with more pre-ChatGPT exposure posted about 8–9% fewer openings by early 2026. The authors’ aggregate calculation implies 2.6% fewer Texas Lightcast postings in 2025 because of generative-AI automation exposure.
That can squeeze people trying to enter or switch fields before it appears as a layoff. It does not show that a given worker was replaced, or that every occupation with an exposed task is shrinking.
Why this is a split, not the final verdict
An earlier Federal Reserve analysis of broad U.S. postings found little indication of a distinct AI-driven decline. A separate Stanford working paper also reports no statistically significant exposed-occupation response in its posting and layoff tests. The measures, places, models and windows differ. Texas may be an early signal, a regional pattern or a method-sensitive result; these sources cannot yet tell us which.
What the Texas result cannot answer
Lightcast captures online ads and underrepresents some kinds of work. The exposure score comes from Claude usage, not verified Texas firm adoption. A comparison after ChatGPT’s release can still pick up another change that hit more-automatable work. The result is a serious association with labor demand, not a causal estimate of AI’s national employment effect.
The claim to demand before acting
Ask where the signal appeared, whether it measures postings, hires or employment, how AI was measured and what the comparison was. “AI cut Texas online postings in more-automatable work” is a defensible warning. “AI eliminated jobs everywhere” is not.