Erick Baumgartner

Bocconi University · Economics · 2026–27 job market

Fields

Labor, Development, Political Economy

Job market paper

Direct and Spillover Effects of Generative AI: Evidence from the Judiciary

Generative AI lowers the cost of producing text, but much of that text must be processed by others. I study this two-sided productivity shock in civil litigation, where lawyers write petitions and judges turn them into decisions. Using metadata and full text from airline small-claims cases in São Paulo, I detect AI use directly in legal documents from 2022–2025. Lawyers respond mainly on the extensive margin: after ChatGPT, adopting lawyers become 16 percentage points more likely to file an airline case in a given month and 9 points more likely to file outside the airline docket. The additional cases are not observably weaker, and petitions become less repetitive and denser, with no detectable effects on case winning rates and on proxies for procedural mistakes. Cases filed by adopting lawyers show downstream effects, leading to a 9 pp decrease in the likelihood that the judge reaches a decision within six months. Judges' own AI adoption produces no immediate average productivity gain, although output rises with experience and among early adopters. Adoption also increases cross-judge dispersion in awards for cases with similar pleaded facts. In a Priest–Klein framework, greater outcome dispersion feeds back into more litigation. The results show that adopter-side productivity gains can miss the margin that matters for system-wide performance.