Vector indexes vs. full-text search
Prism Fast · 2 messages · 200 tokens
When should I use a vector index instead of full-text search?
Prism Fast
They answer different questions. Full-text search matches the words people typed; a vector index matches the meaning behind them.
Use full-text search when:
- People search for exact names, SKUs, error codes or quotes.
- You need predictable ranking and easy debugging.
- Your corpus is small enough that keyword recall is already good.
Use a vector index when:
- Queries and documents use different words for the same idea ("refund" vs "money back") .
- You're building retrieval for an assistant, where the question is a full sentence.
Most production systems combine both, then re-rank the merged results . If you only pick one to start, start with full-text. It is cheaper, and you'll learn which queries fail.
Sources
- 1Semantic search explainedNorthwind engineering blog
- 2Hybrid retrieval patternsInternal wiki · Search
Prism can make mistakes. Check important information.