Long-Context Ordering

Definition

For large / data-rich inputs (20k+ tokens), place longform documents at the top of the prompt and put the query, instructions, and examples last — improving response quality across all models.

Key points

• Documents first, query last. Placing the query at the end can improve response quality by up to 30% in tests, especially with complex, multi-document inputs.
• Structure each document with XML: wrap in <document> tags with <document_content> and <source> (plus other metadata) subtags; the canonical form is <documents> → <document index="1"> → <source> + <document_content> — see XML Structuring.
• Ground responses in quotes: ask Claude to first quote relevant parts of the documents (e.g. into <quotes> tags) before performing the task, to cut through noise.
• Relates to the "lost in the middle" effect — attention is strongest at the start and end, so the query at the end stays salient.

Why it matters for the exam

• "Documents first / query last, up to 30%" is a near-guaranteed, verbatim D4 fact.
• Structured Data Extraction and multi-document RAG scenarios test exactly this ordering.

Common gotchas

• Putting the query before the documents is the wrong answer for long context.
• The 30% figure is tied to long, multi-document inputs — not a blanket claim for short prompts.

See also

• XML Structuring
• Prompt Clarity And Motivation

Sources

• docs-prompting
• netlight-study-guide