Five-Layer Framework for Measuring LLM and Agent Interaction with Web Sources

Web publishers can increasingly observe automated requests from crawlers, agents, search systems, and model-related services, but server-side telemetry does not by itself reveal whether a resource was admitted into a system, retrieved for a task, cited, or materially influential. This working paper separates five evidence layers: Access, Eligibility, Retrieval, Citation and Influence, and defines for each the unit of analysis, admissible evidence, candidate metrics, inferential limits and recommended claim language. Version 0.2 updates the implementation context (edge/origin architecture, edge telemetry caveats, corroboration levels for client classification, sealed measurement with versioned criteria, machine-readable reservation as context) and adds measurement hygiene to threats to validity. Layer definitions are unchanged from version 0.1 (2026-07-29). Includes an authorised Spanish translation; the English text prevails.