Similarity Is Not Relationship

Vector search finds semantically similar content. But expertise often involves connecting things that aren’t similar, they’re related in ways similarity search can’t capture.

A drug and a side effect aren’t semantically similar. A regulation and the business process it affects aren’t semantically similar. A customer complaint and the root cause in your supply chain aren’t semantically similar.

But the relationships between them are exactly what matters.

This distinction explains why pure vector RAG struggles with certain query types: when the answer requires traversing explicit relationships rather than finding semantically adjacent content. Knowledge graphs make relationships first-class objects, enabling reasoning that similarity-based retrieval cannot perform.

The most capable knowledge systems use both: vector search for discovery when you don’t know what you’re looking for; graph traversal when you need to reason through connections.

Related: 05-atom—rag-core-equation, 05-molecule—rag-architecture-taxonomy