Query Rewriting
Also called Query Reformulation
Rephrasing a messy or context-dependent question into a cleaner one before searching.
Think of it like
A good assistant turning "what about the other one?" into "what are the specs of the 15-inch model?" before acting.
Example
In a chat, "and its price?" gets rewritten to "what is the price of the MacBook Air M3?" using earlier context, so retrieval actually works.
How it actually works
Follow-up questions and vague phrasing wreck retrieval because the search sees only the latest words. Query rewriting uses the conversation or an LLM to produce a self-contained, well-formed query. Done well it dramatically improves multi-turn search; done carelessly it can inject assumptions the user never made and quietly change what they asked.
For product teams
The unsung fix that makes chat-style search actually work across turns.
For engineers
LLM- or rule-based reformulation to a standalone query; conditioning on history resolves coreference and ellipsis.
Related
- Query Expansion — Related to query expansion.
- Retrieval Pipeline — Often precedes retrieval in the pipeline.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome