This paper argues that the GEO and AEO category - as currently constituted and as represented by its leading platforms - is producing measurable improvements in the wrong evidence layer. The content it recommends and generates operates at the community and editorial retrieval layer of AI systems. The layer that determines purchase recommendation outcomes at the decision stage is the knowledge graph entity anchor layer: Wikidata entity definitions, Wikipedia category statements, trained-model entity representations, and the structured evidence architecture that AI models evaluate when applying criteria filters at Turn 3 of a buying sequence. These two layers are structurally independent. Improvements in the first layer do not propagate to the second. In documented cases, content produced by GEO programmes actively conflicts with legacy knowledge graph anchors, producing the pattern AIVO has documented consistently across 195+ brands: improved first-prompt visibility combined with unchanged or degraded decision-stage recommendation performance. This paper introduces the term layer mismatch to describe this phenomenon and documents it through five brand case studies, each with AIVO Meridian probe data. The five brands - DocuSign, Akamai Technologies, Clarins, Chanel N°5, and TUI - span B2B SaaS, cloud infrastructure, prestige beauty, luxury fragrance, and European travel. Two are named customers of the leading GEO platforms. Three are not. In every case, the displacement mechanism identified by the AIVO probe methodology operates at the knowledge graph layer that GEO content programmes do not address. The paper documents the three conflict mechanisms through which GEO content can make this problem worse rather than better, the structural reasons the category cannot address the layer mismatch, and the methodology required to do so.