Searchplex ID Retrieval terminology registry

General Concept

Retrieval-Augmented Generation

Canonical ID https://id.searchplex.net/rag/

A generation pattern in which retrieved external information is supplied to a generative model to support, condition, or ground its output.

Also called

  • RAG

Scope

RAG systems commonly combine indexing, retrieval, reranking, prompt or context construction, and generation. RAG uses retrieval, but it is not identical to retrieval, dense retrieval, semantic search, or grounding.

Not equivalent to

  • Retrieval-Augmented Generation is not Agentic RAG. RAG is the broader retrieval-augmented generation pattern; Agentic RAG specifically adds agentic control over retrieval or tool-use steps.
  • Retrieval-Augmented Generation is not Agentic Retrieval. RAG is a generation pattern using retrieved information; agentic retrieval is an agent-controlled retrieval pattern that may be used within RAG systems.
  • Retrieval-Augmented Generation is not Dense Retrieval. Dense retrieval is one possible retrieval component in RAG; RAG is the broader system pattern.
  • Retrieval-Augmented Generation is not Grounding. RAG is one way to provide external context for grounding; grounding is the broader goal or practice of anchoring output in source material.
  • Retrieval-Augmented Generation is not Retrieval. Retrieval is the selection of relevant items; RAG is a generation pattern that uses retrieved information as input to a generative model.
  • Retrieval-Augmented Generation is not Semantic Search. Semantic search is meaning-oriented retrieval; RAG is a retrieval-augmented generation pattern that may use semantic search.
  • Retrieval-Augmented Generation is not Chunking. Chunking is one design choice inside many RAG systems; RAG is the broader retrieval-augmented generation pattern.
  • Retrieval-Augmented Generation is not Conversational Search. Conversational search is a multi-turn search interaction pattern; RAG is a generation pattern using retrieved information.
  • Retrieval-Augmented Generation is not Deep Research. RAG is a retrieval-augmented generation pattern; deep research emphasizes multi-step search, browsing, evidence gathering, and cited synthesis.
  • Retrieval-Augmented Generation is not Multimodal Retrieval. Multimodal retrieval may be used inside multimodal RAG systems, but RAG is a retrieval-augmented generation pattern.
  • Retrieval-Augmented Generation is not Passage Retrieval. Passage retrieval is often used inside RAG systems, but RAG is the broader retrieval-augmented generation pattern.
  • Retrieval-Augmented Generation is not Question Answering. Question answering is a task; RAG is a retrieval-augmented generation pattern that can be used for question answering and other tasks.

Learn more

ID