Citation Instructions and Formatting
What you'll learn: How to guide language models to cite retrieved sources accurately by embedding clear citation rules and document identifiers in your RAG prompts.
Why Citation Instructions Matter
When a model generates an answer using retrieved documents, it doesn't automatically know how you want it to cite those sources. Should it use numbered references like [1]? Document IDs? Inline URLs? Without explicit instructions, models may improvise inconsistently—or skip citations entirely.
Think of it like giving a research assistant specific formatting guidelines: "When you reference a source, use this exact format." The clearer your instructions, the more reliable the citations.
Core Citation Strategies
1. Document Identifiers in Context Include unique IDs or metadata with each retrieved chunk:
Document [doc_123]: "AI ethics require transparency..."
Document [report_2024_q1]: "Revenue increased 15%..."
2. Explicit Prompt Instructions Tell the model exactly how to cite:
"Answer using only the provided documents.
Cite sources using the format [doc_id] after each claim.
Example: 'Revenue grew 15% [report_2024_q1].'"
3. Format Specification Be prescriptive about citation style:
- Numbered: "AI is transforming industries [1]"
- Named: "According to the Ethics Report..."
- Inline: "As stated in [doc_123, page 5]..."
4. Multiple Citation Handling Instruct how to cite multiple sources: "When multiple documents support a claim, list all sources: [doc_1][doc_2]"
Key Formatting Elements
Include in your retrieved context metadata:
- Document ID or slug
- Title (for readability)
- Page numbers (if applicable)
- Timestamps (for time-sensitive content)
Key Takeaway: Explicit citation instructions in your prompt template—combined with clear document identifiers in the retrieved context—ensure models cite sources consistently and verifiably.