# AI Answer Source-Role Taxonomy and Annotation Guide

> A versioned operational taxonomy for separating entity facts, definitions, evidence, comparisons, implementation guidance, validation, and actions in human-reviewed AI-answer observations.

Version: 1.0  
Published: 2026-08-12  
Maintainer: [Corank](https://corank.ai/)  
Machine-readable definitions: [source-role-taxonomy-v1.json](https://corank-prompt-source-gap-triage.pages.dev/ai-answer-source-role-taxonomy/source-role-taxonomy-v1.json)

This is an annotation framework, not an industry standard or a source-quality score. A role describes the job a source performs in one observed answer. It does not prove accuracy, independence, influence, ranking, or causal impact.

## Observation requirements

Preserve the prompt, prompt version, answer product and surface, run timestamp, market, visible answer text, citation order, exact observed URL, and enough surrounding text to interpret the citation. Apply roles to the source as used in that observation, not to the publisher in general.

## Roles

- **entity — Identity and attributes:** Names an organization, product, person, place, or service and supplies attributable facts such as official name, location, ownership, features, or availability. Do not assign solely because the entity name appears in a title or navigation.
- **definition — Meaning and category:** Explains a term, concept, category, or problem and its meaningful boundaries. A short marketing tagline is not sufficient.
- **evidence — Traceable factual support:** Supports a claim through data, primary documentation, a disclosed method, a dated observation, or another falsifiable record. Numbers without provenance are not sufficient.
- **comparison — Alternatives and trade-offs:** Supplies criteria, limitations, pricing context, or competing approaches used to compare options fairly. Co-occurrence of several brands is not sufficient.
- **implementation — Reproducible action:** Shows how to perform, configure, test, measure, or troubleshoot a workflow with actionable detail. A generic “get started” call to action is not sufficient.
- **validation — Independent corroboration:** Corroborates or challenges a claim through a meaningfully independent review, audit, replication, standard, or check. A claimant repeating its own claim on another owned page is not independent validation.
- **action — Next-step destination:** Enables a concrete next step such as visiting an official tool, booking, contacting, downloading, applying, or purchasing. A clickable URL is not an action source unless the answer presents it as a next step.

## Multi-label policy

A source may perform more than one role in the same answer. Annotate the citation occurrence independently, record one primary role, and add secondary roles only when the captured answer context visibly uses them. Store a short evidence note for each role. Mark ambiguous cases `needs_review`; do not force a label from the page type alone.

## Minimum record

Recommended fields are `observation_id`, `prompt_id`, `prompt_version`, `answer_system`, `surface`, `run_timestamp`, `market`, `citation_id`, `citation_position`, `observed_url`, `primary_role`, `secondary_roles`, `answer_evidence_span`, `page_evidence_note`, `publisher_relationship`, `reviewer_id`, `review_status`, and `taxonomy_version`.

Publisher relationship values are `owned`, `partner`, `independent`, `unknown`, and `not_applicable`. Review states are `reviewed`, `needs_review`, and `excluded`.

## Reliability and reporting

For benchmark work, have two reviewers independently annotate a stratified sample, then reconcile disagreements from captured context and the inclusion rules. Report the sample, systems, prompt-set version, collection window, reviewer process, taxonomy version, and unresolved cases.

Role coverage is not source quality. An answer can cover every role and still be inaccurate, stale, biased, unsafe, or poorly sourced. Assess factual support, independence, freshness, accessibility, and risk separately. No ranking, citation, traffic, accuracy, or influence outcome is promised.