Recursive retrieval iteratively fetches documents, drilling into them or following references for deeper exploration. Pattern: Initial retrieval → analyze results → identify citations/references → retrieve those → continue until sufficient depth. Use cases: Research with citations (follow references), hierarchical content (summary → details), multi-part questions, complex reasoning chains. Implementation: Retrieval loop with early stopping based on: information sufficiency, maximum iterations, relevance threshold. Types: Drill-down: Start with high-level, retrieve more specific chunks. Citation following: Extract references from retrieved docs, fetch those. Entity expansion: Identify entities, retrieve more about them. Tree exploration: Build knowledge tree through iterative retrieval. Agentic approach: LLM decides when more retrieval needed and what to retrieve. Challenges: May diverge from original topic, computational expense, determining stop criteria. Integration with RAG: Self-RAG pattern where model evaluates if more retrieval needed. Best practices: Set maximum depth, maintain relevance scoring, cache intermediate results.