Vector search · Semantic search
Vector search is search by semantic closeness between embeddings, rather than keyword matching. It finds a passage that answers the question even without the same words, and is the basis of retrieval in RAG systems.
Standalone definition · 35 words
How does vector search work?
Passages are encoded into vectors and stored in an index. The query is encoded in turn, then the system looks for the closest vectors. At large volumes, it uses approximate nearest-neighbor search, for example with HNSW graphs (Malkov and Yashunin).
How does it differ from keyword search?
Lexical search finds exact terms and handles proper nouns well. Vector search captures meaning, paraphrases and related languages, but can miss exact references. Hybrid systems combine both and merge the rankings.
Impact on writing
Semantic closeness depends on the clarity of the passage. Text centered on a subject, with its reference vocabulary and common rephrasings, is easier to retrieve. Passages that mix several subjects produce less sharp vectors.
Frequently asked questions
What is the difference with dense retrieval?
Dense retrieval is the retrieval that relies on these embeddings. Vector search describes the mechanism.
Do web engines use vector search?
Antitrust trial documents mention embedding-based retrieval at Google.
Do I still need keywords?
Yes, especially for exact terms.
Related terms
Further reading
- Checklist GEO and LLM / AI 2026
- 2026 GEO Tools Comparison: The Decision Guide
- ← Back to the GEO glossary (114 terms)