Definition
Multi-step retrieval is a process where an AI system runs more than one round of search, using what it found in the first round to shape the next. Rather than retrieving once and answering, the model reads, identifies what is missing, and retrieves again until it has enough to respond. Each round is informed by the last.Why It Matters for AI Search
Harder questions are answered in passes. Multi-step retrieval lets an assistant follow a thread, pulling a definition, then the specifics, then a comparison. A business can enter the answer at any step, so being findable for the follow-up matters as much as being findable for the opening query. The pages that get cited are often the ones that answer the second and third questions a model asks itself, not only the first.Related Terms
Agentic search
See also
RAG
See also
Query decomposition
See also
Query fan-out
See also
Reranking
See also