Ai referencing optimization

AI Referencing Optimization

This skill ensures a web codebase or application is strictly optimized for ingestion by 2026 SOTA AI web crawlers and LLM retrieval engines (e.g., Perplexity, ChatGPT Search, Gemini).

When to Use

• Use when setting up repository documentation routing (llms.txt).
• Use when defining application intents or manifest files for AI agent plugins.
• NOT for standard 2010s human-centric SEO (like keyword stuffing).

Core Process

Phase 1: The AI Router (llms.txt)

• The repository must serve an llms.txt at the root, directing AI agents to high-density context files.
• It must structurally separate [System Constraints] from [Implementation Guides].

Phase 2: Answer Engine Optimization (AEO)

• Human HTML semantics (<h1>, <article>) are not enough.
• You must inject explicit, dense JSON-LD microdata for every primary entity to feed knowledge graphs directly.
• Ensure critical data is not locked behind client-side React/Svelte hydration, as fast AI crawlers may not execute JS.

Phase 3: Agentic API Contracts

• AI Agents consume APIs via strict schema interpretation.
• You must generate strict OpenAPI 3.1 specs with exhaustive description fields for every parameter. Omitted descriptions cause AI hallucination.

Common Rationalizations

Rationalization	Reality
"I'll just add some meta tags for SEO."	Legacy meta tags are ignored by semantic answer engines. You must implement JSON-LD and structured entity graphs.
"The LLM will figure out the API from the parameter names."	Agents hallucinate heavily without explicit enum definitions and verbose property descriptions in OpenAPI specs.

Red Flags

• Repositories lacking an llms.txt or .well-known/ai-plugin.json.
• OpenAPI definitions missing explicit description blocks on parameters.

Verification

Before finalizing the AEO optimization:

• An llms.txt file exists and accurately maps the repository.
• The public API possesses an exhaustive OpenAPI spec.
• HTML outputs contain valid, dense JSON-LD graphs.