Embeddings power the retrieval step in RAG: documents are embedded and stored in a vector database, the query is embedded, and the nearest passages are returned. Embedding quality is a major determinant of whether the right passage is findable at all.
Glossary
Embedding
An embedding is a numeric vector representation of text (or other data) that captures meaning, so similar items sit close together in vector space. Embeddings let a system retrieve by semantic similarity rather than exact keywords.