From AI Diffusion to Resident Prosperity: A Framework for an Australian AI Dividend Account

Artificial intelligence may raise Australian output without producing an equivalent gain in the real income or welfare of Australian residents. This paper develops a framework for an Australian AI Dividend Account: a signed measurement architecture for tracing a credibly identified AI-related economic increment from AI treatment or exposure through real production, nominal domestic income, resident income and household incidence, while keeping consumer welfare and asset revaluation as separate, non-additive domains. The contribution is not a new national-account identity or a causal estimate of Australia’s AI dividend. It is a claim-auditing protocol. A Gate-0 treatment record defines what “AI” means in the relevant source; a four-dimensional Evidence Register records availability, construction, inferential status and uncertainty; and a Linked-Estimand and Bridge Test requires each downstream outcome to be separately defined and connected to the preceding layer by an identified accounting or empirical bridge, or by an explicitly simulated bridge. Where the necessary bridge is not identified, the AI-attributable claim stops.Applied to Australian public evidence, Stage 1 finds that many relevant quantities are measurable but a national causal AI dividend is not. The Productivity Commission’s A$116 billion scenario remains a simulated production-side quantity unless additional evidence identifies the subsequent income bridges. A second application tests the stronger ownership proposition that foreign control of scarce AI assets can create a wedge between Australian production gains and resident income: the mechanism is economically plausible, but its magnitude cannot be inferred from an AI import bill or ownership label alone. The paper closes with a bounded Stage-2 design and a companion source-and-calculation package specifying what would be required for stronger claims.