Semantic Search

Search that matches by meaning rather than by exact keywords. Instead of looking for documents that contain the words in your query, semantic search converts both the query and the stored content into embeddings—numerical representations of meaning—and finds the content whose meaning is closest to what you asked for. Searching "how to reset a password" can surface a document titled "Account Recovery Steps" even though the two share zero words in common.

The implementation typically involves similarity search under the hood (cosine similarity, nearest-neighbor lookups), but the distinction matters: similarity search is the mechanism (find vectors that are mathematically close), while semantic search is the capability (find content that means the same thing). One is a math operation; the other is what the user experiences. You can have similarity search without semantic search (e.g., searching image feature vectors), and you can approximate semantic search without vector math (e.g., synonym expansion), but in modern AI systems the two are almost always paired.

Why it matters for writers: Semantic search is why document structure matters for retrieval. When search understands meaning, the organization of your content—headings, metadata, document type—becomes part of that meaning. A well-structured document with clear frontmatter and consistent metadata gives semantic search systems more signal to work with. A flat wall of text forces the system to guess what's important. This is the core argument behind FractalRecall: structural context doesn't just help humans find things, it helps machines find them too.

Related terms: Similarity Search · Embedding · Retrieval-Augmented Generation