What Deeprank Is
Specification
Deeprank is an open specification for AI-mediated business selection. It defines how businesses express selection eligibility and exclusion logic through structured identity declarations.
The core primitive is the Deeprank Selection Profile (DSP) — a machine-readable format that captures what a business does, what it does not do, and under what conditions it is a correct match. Deeprank is a specification, not a product. This documentation is the complete reference model.
Scope
The specification covers: the structure of the Deeprank Profile, a controlled vocabulary for standardized terms, the semantics of fit determination, and implementation guidance. It does not specify how AI systems should be built or which systems should adopt it.
Deeprank defines eligibility semantics. Everything else — ranking, scoring, presentation, trust verification — can vary by implementation.
Problem Statement
The Gap This Specification Addresses
AI systems are already making selection decisions on behalf of users. When someone asks an AI assistant for a recommendation, the system must determine which businesses fit the request. Its only data source is the business's existing web presence — marketing copy written to persuade human visitors, and SEO metadata structured for search engine crawlers. Neither was designed to express selection eligibility.
This is an architectural gap, not a quality problem. Marketing copy is optimized to be broad and inclusive. Selection decisions require narrow precision. These goals are in direct conflict. Better writing and more thorough SEO do not resolve it. Improving model capabilities does not resolve it either — the source data is designed for persuasion, and inference from persuasion content has a hard ceiling.
The result is predictable: businesses are selected for work they cannot do because their marketing implies broader capability than they possess, and excluded from work they are qualified for because their marketing emphasizes a primary audience and omits secondary capabilities. The AI is not making errors. It is doing the best it can with data that was designed for a different purpose. The structured data layer for selection declarations does not exist. Deeprank defines it.
The Deeprank Approach
Exclusions as a Feature
What a business does not do is as important as what it does. Exclusions prevent bad matches. A plumber who declares "no commercial HVAC" is more useful to a selection system than one who lists only what they do. The ability to say no is a core capability of the specification. This treatment of exclusions as a first-class structural element — not an afterthought — is the specification's primary conceptual contribution relative to existing structured data standards.
Declaration Over Inference
Explicit declaration replaces AI inference as the primary mechanism. If a business can state something directly — its capabilities, constraints, geographic scope, customer type — it should. Inference remains a fallback for businesses that have not yet adopted structured declarations.
Binary Fit
Deeprank models selection as binary: for any given query, a business either fits or does not. This simplification reduces ambiguity and enables clearer selection logic. AI systems may layer additional ranking signals on top, but the fit determination itself is discrete.
Specificity Enables Matching
Broad descriptions prevent accurate matching. Specific declarations enable it. Explicit capability boundaries — declared in structured form — give AI systems the precision they need to match businesses to appropriate queries.
What Deeprank Is Not
- Not SEO, AEO, or GEO. Deeprank does not optimize visibility, answer presence, or search rankings. It defines selection eligibility. It is complementary to these disciplines.
- Not a ranking system. Deeprank does not rank businesses or produce ordered lists. It provides structured data for selection eligibility. The selection logic is implemented by AI systems.
- Not a replacement for Schema.org. Schema.org defines what an entity is. Deeprank defines when that entity should or should not be selected. They are complementary.
- Not a ratified standard. Deeprank is an independent public specification, not ratified by W3C, IETF, or ISO.
- Not a product or service. There is no software to buy or platform to sign up for. The specification is publicly available documentation.
Architecture and Current Status
Definition Layer: deeprank.org
This site is the definition layer. It publishes the specification, controlled vocabulary, JSON Schema, and reference model for the Deeprank Profile. It defines the semantics of selection eligibility and exclusion logic. It does not generate, validate, or deploy profiles. The specification is public and open — the source is available on GitHub. Any AI product team, platform, or toolchain may implement it independently. For known implementations, see the Ecosystem page.
Current State vs. Future State
Deeprank Profiles work today as structured data that AI systems can consume via web scraping, JSON-LD embedding, and context injection. This requires no platform adoption. Native consumption by AI platforms — where systems natively read and integrate Deeprank declarations without scraping — is a future-state goal, not current reality. The specification is designed to be useful in both modes.
Specification Status
Version 1.0. The core structure (six-layer Deeprank Profile) is stable. The controlled vocabulary may be extended based on implementation feedback. Changes to the specification are additive only — new optional fields and extended vocabulary, not removed or renamed fields. The specification acknowledges that self-declared data requires verification mechanisms not yet defined. Trust architecture is an identified area for future development.
Publication & Citation
Academic Paper
The Deeprank specification is described in a published paper: The Selection Layer: Structured Eligibility Declarations for AI-Mediated Business Discovery. The paper presents the design rationale, the four-layer AI optimization stack, selection scenarios demonstrating structural failure modes, and an assessment of current limitations.
DOI: 10.5281/zenodo.18641029 — Full details and citation formats
Cite This Work
Zhgenti, B. (2026). The Selection Layer: Structured Eligibility Declarations for AI-Mediated Business Discovery. Zenodo. https://doi.org/10.5281/zenodo.18641029