Firm Size, Skills, and the Conversion of Generative AI into Productivity: Evidence for Spanish Manufacturing — pre-specified design lodged before data access

This deposit lodges the empirical design of a study of generative-AI adoption and total factor productivity in Spanish manufacturing, before the author has accessed the microdata the study requires. The paper distinguishes the task-level potential of generative AI from its firm-level realisation — the conversion problem — and asks under what organisational conditions a firm converts adoption into measured productivity. It pre-specifies the estimating equation, a shift-share instrument built from pre-period occupational composition, a regression discontinuity at the fifty-employee threshold of Article 63 of the Spanish Workers' Statute, the mediation test for skills and managerial capacity, a named rival hypothesis based on market power, and the inference and power analysis. The microdata are those of the Encuesta sobre Estrategias Empresariales (ESEE), distributed by Fundación SEPI. At the date of deposit an access application is pending and no ESEE microdata have been received, inspected or analysed. Section 4 and the results paragraph of Section 1 are deliberately left open. The deposit includes the occupational exposure measure used to build the instrument, frozen in advance, so that the choice of exposure index cannot vary with the results. The README states what is pre-specified, what is explicitly left open, and how any departure will be reported. This is a working paper. It carries no editorial endorsement and no journal has reviewed the design.