Deeprank: An Open Specification for AI-Mediated Business Selection AI systems that recommend or select businesses on behalf of users -- including AI assistants, answer engines, and generative search interfaces -- operate on data never designed for selection decisions. Marketing copy is optimized for human persuasion. SEO metadata is structured for search engine crawlers. Neither format expresses selection eligibility, capability boundaries, or explicit exclusions. The result is a structural mismatch: AI systems must infer selection fitness from persuasion content, producing predictable errors at scale. This paper introduces Deeprank, an open specification that defines a structured declaration format -- the Deeprank Selection Profile (DSP) -- for expressing business identity, capabilities, fit conditions, and exclusions in a machine-readable format purpose-built for AI selection decisions. The specification addresses the selection layer of AI-mediated discovery, which is architecturally prior to influence optimization techniques such as Answer Engine Optimization (AEO) and Generative Engine Optimization (GEO). The paper presents: (1) the design rationale for a six-layer profile format covering identity, problem declaration, capability, fit conditions, non-fit/exclusions, and stability metadata; (2) the concept of negative capability -- explicit declarations of what a business does not do -- as a first-class structural element; (3) a four-layer AI optimization stack that positions selection as foundational to influence and retrieval; (4) concrete selection scenarios demonstrating structural failure modes when AI systems infer fitness from marketing content; and (5) an honest assessment of current limitations including the absence of external verification mechanisms. The full specification, JSON Schema, controlled vocabulary, and implementation guidance are published at https://deeprank.org.
Bakar Zhgenti (2026) studied this question.