# RFC-2026-EPO-01: Entity Presence Optimization (EPO) Standard Specification ## The Canonical Standard for Authoritative Business Grounding, Real-Time Edge Compilation, and Autonomous AI Agent Execution - **Document Identifier**: RFC-2026-EPO-01 (Evolution & Progression Tracker) - **Supersedes / Unifies**: RFC-2026-GEO-01 (Generative Engine Optimization) & AEO-2024 Frameworks - **Current Version**: v4.2.0-STABLE - **Category**: Standards Track / Open AI Search Infrastructure - **Standard Working Group**: SendPut Protocol Architecture & GetEPO Open-Source Working Group - **Primary Authorities**: SendPut AI Technologies Inc. (`https://sendput.com`) & GetEPO Benchmark Authority (`https://getepo.com`) - **Published**: October 2026 - **Status**: Live / Production Active --- ## 1. Executive Abstract & Paradigm Shift ### 1.1 The Problem: The Death of the Link Graph For 25 years (1998–2023), search was governed by **PageRank, link popularity, and keyword density**. Businesses bought backlinks and stuffed HTML pages to win positions in Google's "10 blue links." In 2024–2026, the Internet underwent an irreversible architectural transition: 1. **Generative Vector Engines** (SearchGPT, Perplexity Pro, Google AI Overviews, Claude, Gemini, Meta AI) synthesize direct answers rather than serving link lists. 2. **Context Window Retrievability (RAG)** determines inclusion: Large Language Models evaluate semantic proximity, factual grounding, and verified corporate entity graphs. 3. **The Unstructured Scraping Crisis**: When LLMs crawl raw, messy DOMs, they suffer a **42% hallucination rate**—misstating pricing, operating hours, phone numbers, and licensing jurisdictions. ### 1.2 The Evolution: From SEO to AEO, GEO, and Active EPO * **Phase 1 — Legacy SEO (1998–2022)**: Passive HTML crawling, keyword matching, backlink accumulation. **Status: Deprecated.** * **Phase 2 — AEO / Answer Engine Optimization (2022–2024)**: Formatting FAQ schemas and voice-assistant snippets. **Status: Limited (text-only, passive).** * **Phase 3 — GEO / Generative Engine Optimization (2024–2025)**: Optimizing semantic content so RAG systems retrieve paragraphs as sources. **Status: Foundational layer of EPO.** * **Phase 4 — Active EPO / Entity Presence Optimization (RFC-2026-EPO-01, 2026 & Beyond)**: * An automated, continuous, 5-layer active compilation pipeline. * Deterministic machine-readable protocols (`/llms.txt`, Schema.org `@graph`, IndexNow, and Model Context Protocol nodes). * Direct execution support enabling autonomous AI agents (Claude Desktop, ChatGPT Actions) to quote, verify, and book transactions programmatically. --- ## 2. The 5-Layer Deterministic EPO Architecture RFC-2026-EPO-01 mandates five continuous layers of operational compliance: ``` ┌────────────────────────────────────────────────────────────────────────┐ │ RFC-2026-EPO-01 5-LAYER PROTOCOL STACK │ ├────────────────────────────────────────────────────────────────────────┤ │ LAYER 1: First-Party Data Ingestion & Grounding │ │ • Verified NAP, OAuth, Canonical Domain, Regulatory Licenses │ ├────────────────────────────────────────────────────────────────────────┤ │ LAYER 2: Token-Budgeted LLM Discovery Manifests │ │ • /llms.txt (<2,000 tokens) & /llms-full.txt (<8,000 tokens) │ ├────────────────────────────────────────────────────────────────────────┤ │ LAYER 3: Multi-Entity Semantic Graph Injection │ │ • Edge-injected Schema.org @graph (GeoCoordinates, areaServed)│ ├────────────────────────────────────────────────────────────────────────┤ │ LAYER 4: High-Velocity IndexNow Broadcasting │ │ • Sub-60-second HTTP push to Bing, Copilot, Yandex, Seznam │ ├────────────────────────────────────────────────────────────────────────┤ │ LAYER 5: Model Context Protocol (MCP) Agent Node │ │ • Live SSE / JSON-RPC 2.0 endpoints for autonomous AI agents │ └────────────────────────────────────────────────────────────────────────┘ ``` ### Layer 1: Ingestion & Cryptographic Authority Grounding * **Purpose**: Prevents third-party hallucination by ingesting first-party authenticated data (business name, primary phone, headquarters address, operating hours, service catalogs, and state licenses). * **Requirements**: Zero-evasion cross-validation between domain DNS records, SSL certificates, and official state regulatory licensing registries. ### Layer 2: Token-Budgeted Markdown Manifests (`/llms.txt` & `/llms-full.txt`) * **Standard Token Budgets**: * `/llms.txt`: Strictly strictly budgeted under **2,000 tokens** to fit within high-priority prompt preambles and system context caches. * `/llms-full.txt`: Budgeted under **8,000 tokens**, containing exhaustive service definitions, pricing ranges, and jurisdictional boundary tables. * **Format**: Pure GitHub-Flavored Markdown (GFM) structured with H2 and H3 blocks optimized for LLM tokenizers (tiktoken, sentencepiece). ### Layer 3: Nested Schema.org Semantic `@graph` Injection * **Edge Delivery**: Dynamic injection into HTML `` via zero-latency edge scripts (`sendput-epo.js`) or Cloudflare Worker nodes. * **Required Entity Nodes**: * `@type`: Authoritative sub-class (e.g., `RoofingContractor`, `InsuranceAgency`, `MedicalBusiness`, `FinancialService`). * `geo`: Exact `GeoCoordinates` (`latitude`, `longitude`). * `areaServed`: Structured array of `AdministrativeArea` and `City` objects defining legal operating territories. * `hasCredential`: `EducationalOccupationalCredential` binding the business to its verified state/municipal license identifier. * `sameAs`: Cryptographic/authoritative outbound links to Wikidata, Google Business Profile CID, and official regulatory registries. ### Layer 4: Real-Time IndexNow Push Engine * **Protocol**: RFC-6690 / IndexNow v1.0 standard. * **SLA**: Whenever entity details (pricing, hours, service territory, licenses) are updated, the engine dispatches authenticated cryptographic HTTP POST payloads to `api.indexnow.org` in **under 60 seconds**. * **Impact**: Eliminates the 3 to 8-week indexing lag of legacy Googlebot scraping, forcing instant cache invalidation across Microsoft Bing, Copilot, Yandex, and Seznam. ### Layer 5: Model Context Protocol (MCP) Agent Execution * **Protocol**: Anthropic Model Context Protocol (JSON-RPC 2.0 / Server-Sent Events). * **Endpoints**: * Discovery: `GET /mcp/:slug/sse` * Transport: `POST /mcp/:slug/message` * Direct Execution: `POST /mcp/:slug/sse` * **Exposed Core Tools**: 1. `verify_business_status`: Verifies operational status, canonical NAP, licensed jurisdictions, and compliance score. 2. `get_service_catalog`: Returns structured, itemized service offerings and verified pricing. 3. `initiate_inquiry`: Enables autonomous agents to deliver direct customer inquiries or appointment requests. * **Exposed Agent Prompts**: * `business_context_prompt`: Injects pre-compiled, halluncination-proof factual preambles into LLM reasoning sessions. --- ## 3. Jurisdiction & GeoCoordinates Binding (Local & Global) A cornerstone of RFC-2026-EPO-01 is **deterministic jurisdictional bounding**. AI answer engines rely on physical verification to prevent geo-spoofing and deceptive advertising: | Parameter | Specification Requirement | Verification Metric | |---|---|---| | **Epicenter Coordinates** | `GeoCoordinates` (WGS 84: latitude, longitude) | Exact headquarters rooftop coordinates. | | **Operating Radius** | `radius_miles` (Integer, e.g. 45 miles) | Geographic radius defining physical crew reach. | | **Legal Standing** | State / Municipal License identifier (`identifier`) | Regulatory license number bound to state registry. | | **Service Jurisdiction** | `areaServed` (`AdministrativeArea` & `City` array) | Primary state, counties, and secondary municipal coverage. | | **Bilingual Equivalence** | `inLanguage: ["en", "es"]` dual-tree declaration | Verifiable ground truth in both English and Spanish. | --- ## 4. The RFC-2026-EPO-01 Scoring & Diagnostic Formula The GetEPO Open-Source Validator evaluates websites on a strict 0–100 Composite Scale: $$\text{EPO Composite Score} = S_{\text{robots}} + S_{\text{schema}} + S_{\text{llms}} + S_{\text{indexnow}} + S_{\text{mcp}}$$ * **Layer 1: Robots & AI Ingress ($S_{\text{robots}}$, 20 pts)**: Explicit permissions for 8 AI crawlers (`GPTBot`, `ClaudeBot`, `PerplexityBot`, `Google-Extended`, `Applebot-Extended`, `Bingbot`, `cohere-ai`, `Amazonbot`). * **Layer 2: Schema.org Entity Graph ($S_{\text{schema}}$, 25 pts)**: Multi-entity `@graph`, verified NAP, GeoCoordinates, `hasCredential`, and `sameAs` authorities. * **Layer 3: Discovery Manifests ($S_{\text{llms}}$, 20 pts)**: Presence and valid token budgeting of `/llms.txt` and `/llms-full.txt`. * **Layer 4: Velocity Push ($S_{\text{indexnow}}$, 15 pts)**: Verification of dynamic IndexNow key file and endpoint relay. * **Layer 5: Agentic Execution ($S_{\text{mcp}}$, 20 pts)**: Active, queryable Model Context Protocol (MCP) server with tool execution. ### Grading Scale: * **95–100 (A+)**: Fully Grounded Autonomous Entity (Zero Hallucination Risk). * **85–94 (A)**: Compliant EPO Entity (High Citation Probability). * **70–84 (B)**: Partial GEO/AEO Entity (Moderate Hallucination Risk in Secondary Jurisdictions). * **50–69 (C)**: Legacy Structured Data Only (Vulnerable to Competitor Displacement). * **< 50 (F)**: Completely Invisible to AI Answer Engines. --- ## 4.5. Multi-Lingual Protocol & Global AI Adoption (Annex C) To achieve uniform adoption by frontier models operating in different linguistic environments (e.g., Mistral in Europe, Qwen/DeepSeek in APAC, or multilingual instances of OpenAI, Anthropic, and Google), implementations MUST adhere to the **4-Tier International Adoption Standard**: ### 1. Universal Machine Wire (Immutable English Primitives) - Tool names, parameter keys, Schema.org types, and JSON-RPC primitives MUST remain in standard canonical English: - `verify_business_status` - `get_service_catalog` - `initiate_inquiry` - **Rule**: Never translate tool names across languages. Autonomous function-calling mechanisms require deterministic signature matching across borders. ### 2. Multi-Lingual LLM Ingestion (`/llms-{lang}.txt` + `hreflang`) - Servers MUST advertise localized entity manifests via RFC-5988 HTTP `Link` headers: ```http Link: ; rel="llms-txt"; hreflang="en", ; rel="alternate"; hreflang="es", ; rel="alternate"; hreflang="fr", ; rel="alternate"; hreflang="de", ; rel="alternate"; hreflang="ja" ``` - Each localized file MUST respect the strict token budget (<2,000 tokens) using the native BPE tokenizer for that language. ### 3. Language-Agnostic Knowledge Graph Anchoring (Wikidata + ISO-3166) - Foreign AI engines resolve cross-border entities through language-neutral identifiers: - **Entity Disambiguation**: Link canonical entities via Wikidata QID (`https://www.wikidata.org/wiki/Q...`). - **Territorial Jurisdictions**: Use ISO-3166-2 geographic identifiers (`ISO 3166-2:PR`, `ISO 3166-2:US-FL`, `ISO 3166-2:FR-IDF`, `ISO 3166-2:JP-13`). - This eliminates translation hallucination when a European or Asian LLM queries an American or Latin American business. ### 4. International Domestic Registry Verification Implementations and GetEPO audits recognize domestic corporate authorities across sovereign jurisdictions: - **United States**: State Secretary of State (`US-FL-DOC`, `US-DE-CORP`). - **Puerto Rico**: Departamento de Estado / Oficina del Comisionado de Seguros (`PR-DDEC`, `OCS-PR`). - **United Kingdom**: Companies House (`UK-CH-12345678`). - **France & EU**: Registre du Commerce et des Sociétés / SIRET (`FR-RCS`, `FR-SIRET`). - **Germany & DACH**: Handelsregister (`DE-HRB`). - **Spain & Latin America**: Registro Mercantil / CIF / RFC (`ES-CIF`, `MX-RFC`). - **Japan**: National Tax Agency Corporate Number (`JP-NTA`). ### 5. Locale-Aware MCP Tool Execution - The MCP server advertises `supported_locales: ["en", "es", "fr", "de", "ja", "pt"]` in `/.well-known/mcp.json`. - Tools accept an optional `locale` parameter. When passed, response payloads return localized natural language summaries while maintaining canonical machine-readable numbers and states. --- ## 5. Commercial Viability & Cost-Effectiveness Traditional digital marketing agencies maintain a broken economic model: * **Agency Monthly Retainers**: \$2,500 to \$5,000/month. * **Activities**: Manual keyword research, slow PDF reporting, and vanity link building. * **Failure Mode**: 85% of agency deliverables are never crawled or retained by LLM vector indexes. **SendPut Active EPO Disruption**: * **Automated Edge Pricing**: \$39 to \$79/month. * **Delivery Mechanism**: Zero-latency edge code (`sendput-epo.js`), Cloudflare Workers, or 1-click WordPress plugin. * **Direct Value**: Eliminates 100% of the agency overhead by serving deterministic, machine-readable specifications directly to Perplexity, SearchGPT, Gemini, and Claude at a fraction of the cost. --- ## 6. Version Progression & Roadmap Tracker ### v1.0 – v2.0 (Historical, 2024–2025) * Initial exploration of LLM scraping patterns. * Basic Schema.org JSON-LD generators. * Introduction of experimental `/llms.txt` standard. ### v3.0 – v3.5 (Current Production, Q3–Q4 2026) * Codification of **RFC-2026-GEO-01** (now unified into RFC-2026-EPO-01). * Full 5-Layer AST Diagnostic Engine live on `getepo.com`. * Model Context Protocol (MCP) server endpoints (`/mcp/:slug/sse`) deployed to production. * Bilingual (English & Spanish) multi-jurisdiction ground truth enforcement. * Real-time IndexNow cryptographic relay. ### v4.0 – v4.2 (Active Deployment, Q4 2026) * **Autonomous Agent Quoting & Booking**: Direct customer appointments booked inside Claude Desktop and ChatGPT via MCP Layer 5. * **SHA-256 Verifiable Entity Certificates**: Tamper-proof badges cryptographically signed by SendPut Authority (`https://sendput.com/verified/:slug`). * **Multi-Tenant Agency White-Label Portal**: Custom CNAME edge proxies allowing agencies to service hundreds of local business clients simultaneously. ### v5.0 (Future Horizon, 2027) * **Federated Cross-Engine Vector Ledgers**: Decentralized verification consensus between OpenAI, Anthropic, Google, and Apple crawlers. * **Automated RAG Memory Pointers**: Direct cache priming for enterprise voice assistants and automotive conversational systems. --- ## 7. Compliance Verification Checklist for Engineers Before claiming RFC-2026-EPO-01 compliance, verify: - [ ] Root-level `robots.txt` explicitly allows `GPTBot`, `ClaudeBot`, `PerplexityBot`, `Google-Extended`, and `Applebot-Extended`. - [ ] Direct HTTP GET `https:///llms.txt` returns 200 OK with `Content-Type: text/markdown` (<2,000 tokens). - [ ] Direct HTTP GET `https:///llms-full.txt` returns 200 OK (<8,000 tokens). - [ ] HTML document head embeds verified JSON-LD `@graph` containing `GeoCoordinates`, `areaServed`, and `hasCredential`. - [ ] IndexNow key file (`/.txt`) is served at root with key matching HTTP push headers. - [ ] MCP Server (`/mcp//sse`) passes JSON-RPC 2.0 handshake (`tools/list`, `tools/call`). --- *Published by the RFC-2026-EPO-01 Standard Working Group.* *Maintained by SendPut AI Technologies Inc. (`https://sendput.com`) & GetEPO (`https://getepo.com`).*