From Information Retrieval to Agentic Action: a Framework for Brand Visibility in AI-Mediated Markets

Brand Erasure: Why Most Companies Will Disappear from AI Answers Before They Notice In March 2025, Pew Research tracked nearly 69,000 Google searches across 900 U.S. adults. When a search produced an AI Overview, users clicked a traditional result 8% of the time, against 15% without one. They clicked a link inside the AI summary 1% of the time. That is not a softer click curve. It is a phase change in how attention reaches brands. This paper is the first unified treatment of that phase change. It integrates three optimization disciplines emerging in response to the shift — Answer Engine Optimization (AEO), Generative Engine Optimization (GEO), and Agentic Optimization (AgO) — under a single theoretical lens: delegated consumer–AI agency. Rather than reaching for classical Jensen–Meckling agency theory, where the metaphor breaks down at first contact, the framework is grounded in the marketing literature on consumer–AI delegation (Puntoni et al., 2021; Davenport et al., 2020) and the embedding-bias tradition that lets us measure how brands are represented in the substrate AI assistants think in. The central contribution is the construct of brand erasure — the systematic elimination of brand identity from synthesized answers and completed agentic actions. The paper develops six testable propositions, an audit methodology that detects erasure across foundation models over time, and a managerial KPI shift table that maps the click-economy metrics onto their AI-mediated counterparts: share-of-citation, embedding proximity, agent execution rate, hallucination rate, semantic equity. Three reasons to read it now: The shift is happening, and erasure compounds. Brands lose attribution gradually as paraphrase chains accumulate; the period during which the substrate is still malleable is short. The framework is operational, not descriptive. The audit toolkit — open-sourced at github.com/marcosfigueira/brand-erasure-framework — lets practitioners measure semantic equity, share-of-citation, and erasure rates today. Most existing work treats SEO, brand, and AI as separate disciplines. The paper argues they have collapsed into a single design problem with three sequential layers, and specifies what to build at each. If your category includes anything a consumer might ask an AI assistant — and most do — the question is no longer whether to compete in the AI-mediated layer. It is whether the way you compete leaves your name attached when the answer arrives.