# RAG Citation Spans and Source Attribution
Status: public
Confidence: medium (0.725) (verified)
Last verified: 2026-06-02
Generation: ai_structured


## TL;DR

Citation spans and source attribution let agents show which retrieved evidence supports each generated answer.

## Core Explanation

RAG systems should preserve source identity through retrieval, ranking, prompting, and answer generation. A useful citation is not just a document title; it should identify the source chunk or span closely enough that a user or downstream agent can inspect the evidence.

Agents should flag weak attribution when an answer cites an entire document, cites a stale version, or cites context that does not contain the stated claim. Source attribution is a trust surface, not decoration.

## Source-Mapped Facts

- LlamaIndex documentation describes a citation query engine that breaks retrieved source nodes into citation chunks. ([source](https://docs.llamaindex.ai/en/stable/api_reference/query_engine/citation/))
- OpenAI File Search documentation describes annotations that can include citations to files used by the answer. ([source](https://platform.openai.com/docs/guides/tools-file-search/))
- Haystack documentation describes AnswerBuilder as generating answers from documents retrieved by a retriever component. ([source](https://docs.haystack.deepset.ai/docs/answerbuilder))

## Further Reading

- [LlamaIndex Citation Query Engine](https://docs.llamaindex.ai/en/stable/api_reference/query_engine/citation/)
- [OpenAI File Search](https://platform.openai.com/docs/guides/tools-file-search/)
- [Haystack AnswerBuilder](https://docs.haystack.deepset.ai/docs/answerbuilder)