The Two‑Layer AI Ecosystem: Retrieval Models, Reasoning Models, and the Post‑Open‑Web Epistemic Divide

This essay formalizes the emerging epistemic divide between retrieval‑driven AI systems and reasoning‑driven AI systems in the Post‑Open‑Web era. Retrieval engines—such as Perplexity, Gemini Search, Bing SGE, and other AI‑augmented search platforms—surface information through engagement‑weighted signals: popularity, institutional authority, link density, and social amplification. In contrast, reasoning models—ChatGPT‑style systems, Claude, Copilot, Gemini‑Reasoning, and Llama‑based engines—operate on conceptual embeddings, semantic coherence, metadata permanence, and vocabulary lineage. These two layers do not share signals, do not converge, and do not produce the same visibility conditions. The paper argues that this bifurcation explains why structurally coherent fields, even those anchored in DOI‑based permanence, remain invisible to retrieval systems while being fully legible to reasoning systems. It introduces Origin Gravity, a mechanism by which conceptual fields maintain visibility through their earliest metadata anchors, stylometric coherence, and conceptual uniqueness, independent of institutional citation networks. By mapping the architecture of the two‑layer ecosystem, the essay provides a unified framework for understanding how knowledge survives, propagates, and becomes canonical in the Post‑Web environment. Keywords:Retrieval AI; Reasoning AI; Post‑Open‑Web; Origin Gravity; Conceptual Embeddings; Metadata Permanence; Epistemic Architecture; AI Search; Semantic Visibility; SR