Semantic Search

What it means

Queries and documents live in the same embedding space; the system returns items closest in meaning to what the user asked.

Why designers should care

Semantic search needs UX for ambiguity: query refinement, faceted filters, and previews of why a result matched, especially in enterprise knowledge bases.

Example

A design system search for “make destructive actions obvious” surfaces Danger button, error banners, and deletion modals, even if none contain the word “destructive.”

Common mistakes

• • Single-result answers with no list to verify the right doc was found.
• • Ignoring metadata filters (team, product, locale) that embeddings alone cannot infer.
• • Over-trusting top result without snippet highlighting.

Related terms

• Embeddings: Text is converted to vectors in high-dimensional space; closer vectors mean more similar meaning, enabling “find like this” search beyond keywords.
• Vector Database: Specialized storage optimized for similarity search over millions of vectors, often with metadata filters (user, folder, date).
• Retrieval-Augmented Generation (RAG): RAG = search step + generation step: find chunks likely to contain the answer, inject them into the prompt, then produce a reply grounded in those sources.
• Context Window: Everything sent to the model must fit inside the context window; overflow is truncated, summarized, or rejected depending on product policy.
• Hallucination: The model generates plausible-sounding content that does not match facts, retrieved sources, or user-provided data.