The Difference Between Retrieval and Reasoning
Retrieval finds content that matches a query. A retrieval system, given the query "content governance best practices," returns content that contains those words or semantically similar terms. Retrieval is a search problem — it is solved by search indexes, embedding models, and ranking algorithms.
Reasoning understands content in relation to other content. A reasoning system, given the query "what governance approach should a regulated financial services firm use for AI-generated content?" does not just retrieve documents about governance — it understands the intersection of governance, regulation, financial services, and AI content; identifies the relevant guidance; and synthesises a response that reflects the specific context. Reasoning requires semantic structure — explicit encoding of what content is about, how it relates to other content, and what it means in specific contexts.
The Three Semantic Structure Investments
Entity tagging: Identifying and tagging the named entities — organisations, people, concepts, products, locations — referenced in content. Entity tags create the vocabulary for semantic search and relationship encoding.
Relationship encoding: Explicitly specifying how content items relate to each other — this guide extends that concept, this case study illustrates this principle, this regulation applies to this content type. Relationship encoding creates the connection layer that enables reasoning across the content library.
Schema.org markup: Structured data markup embedded in content pages that communicates content meaning to AI systems, search engines, and knowledge graphs in a machine-readable format.
Key Takeaways
1. Retrieval is a search problem; reasoning is a structure problem — semantic structure is what enables AI to reason about content rather than simply retrieve it.
2. The three semantic structure investments — entity tagging, relationship encoding, and Schema.org markup — build progressively toward a content library that AI can reason about.
3. Semantic structure is the highest-leverage IA investment for organisations deploying RAG systems, AI assistants, or knowledge management AI.