What Corporate AI Text Measures: A Multi-State Indicator of Corporate AI Diffusion
Publication Date
7-14-2026
Abstract
Researchers increasingly use corporate disclosures to trace artificial intelligence (AI), but a keyword hit may describe a working focal-firm system, a tentative plan, another organization, or an unrelated acronym. We construct a validated three-state disclosure indicator that distinguishes substantiated focal-firm claims (A1), weak focal-firm claims (A2), and no valid focal-firm claim (A3). The classified corpus contains 185,551 documents. Of these, 138,606 fall within the active 2015–2025 listing intervals of the 6,934 U.S. public firms in the analysis sample. In a stratified human benchmark, A1 precision rises from 0.493 for keyword retrieval to 0.940 under the full taxonomy. Across firm-years, keyword prevalence exceeds A1 prevalence by a median of 21.3 percentage points. More importantly, the discrepancy changes form: unrelated acronyms and other lexical false positives dominate early, whereas external and non-focal discussion becomes dominant later. Disclosure paths are only partly ordered. A2 is associated with later A1, but 77.0% of firms observed in A1 have no prior A2. Comparisons with Census and patent data further show that agreement depends on how observations are combined and which benchmark is used. In an empirical application to subsequent high-confidence AI-patent publication, raw keyword exposure has no detectable association while prior focal-firm claims do. The multi-state measure improves fit among firms used to estimate the model over raw keywords but does not improve prediction for other firms. Corporate text therefore records disclosed evidence of technological activity rather than adoption itself. The paper contributes a measurement framework for showing when retrieval, attribution, and evidence rules change both observed disclosure diffusion and conclusions drawn from the indicator.
Document Type
Article
Keywords
Artificial Intelligence, Technology Diffusion, Innovation Indicators, Corporate Disclosure, Text Measurement, Patents
Disciplines
Strategic Management Policy
Source
SMU Cox: Strategy (Topic)
Language
English
