Attribution of post-ChatGPT state-level unemployment risk increases

Ascertain whether post-ChatGPT increases in unemployment risk for computer and mathematical occupations observed in California, Washington, and Alaska are attributable to diffusion of large language models or to other macroeconomic or sectoral factors.

Background

Nationally, unemployment risk in highly exposed occupations began rising in early 2022—prior to ChatGPT—and generally stabilized thereafter. However, the authors observe that a small number of states show post-launch increases in computer and math occupation unemployment risk.

They explicitly state that timing alone cannot rule out a contribution from LLM diffusion in these states, highlighting an unresolved attribution question that requires further evidence to determine the underlying drivers.

References

While these patterns hold nationally, a small number of states show post-launch increases in computer and math occupation unemployment risk (e.g., CA, WA, and AK; Figs.~\ref{fig:allUiRiskTimeSeries_CA}–\ref{fig:allUiRiskTimeSeries_AL}). In these cases, timing alone cannot rule out a contribution from LLM diffusion.

— AI-exposed jobs deteriorated before ChatGPT  (2601.02554 - Frank et al., 5 Jan 2026) in Subsection “Unemployment risk for AI-exposed occupations”