The Content of Enactment: Statutory Clauses and Policy Mechanisms in State AI Legislation

State legislatures have become the primary site of articial intelligence (AI) regulation in the United States, yet most bills never become law. Understanding what enacts AI legislation has immediate practical consequences for how humans and AI will coexist. Recent literature established that 67.7% of AI bills stall before a oor vote and proposed dedicated AI committees and adaptive legislation. This paper focuses on the content of the legislation itself to explore trends in AI legislation. This paper provides the first systematic clause- and mechanism-level measurement of U.S. state AI bills. We extract 24 statutory clauses and 12 cross-cutting policy mechanisms from the full text of 357 oor-vote AI bills across all 50 U.S. states (2017-2026). We contribute three new content taxonomies for state AI legislation (24 statutory clauses, 12 cross-cutting mechanisms, and an expanded 21-topic classier), operationalized via a dual-mode detector validated on a stratied 34-bill sample, and apply them to all 357 oor-vote bills with bivariate and adjusted logistic-regression analyses. We further show minor-protection bills bundling increasingly operational safeguards over time and, via a mechanism-level provenance case study of companion-chatbot regulation, that several mechanisms appear in state law before surfacing in federal bills. A coherent directional pattern emerges throughout: facilitative and consultative content (appropriations, procurement standards, consumer rights) accompanies passage, while regulatory and penal content (private right of action, consumer opt-out, penalties) accompanies failure.