A public-health team asks Claude to answer questions about a 60-page communicable-disease reporting manual. The current prompt states the question first, then pastes the manual after it. Answers are serviceable but frequently miss details that are plainly in the text. What ordering change should the team make first?

Prompt ordering is a real variable for long-context work rather than a cosmetic choice. When a question arrives before sixty pages of source, the model holds the task in mind while ingesting the document; when the document comes first, the instruction lands with the entire manual already in context and can be oriented against a complete source.

Anthropic's long-context guidance is to place lengthy documents near the top of the prompt and the query at the end, a change reported to improve response quality substantially on document-grounded tasks. For the public-health team it is a one-line edit to the template, costs no additional tokens, and targets the reported symptom directly, which is answers that miss details plainly present in the text.

Chunking into one request per paragraph severs the cross-references that make a reporting manual coherent and multiplies request volume without improving recall; asking questions against a summary discards precisely the specific details being missed; and interleaving the question after every page duplicates text while giving the model several competing places to anchor the task.

Exam caveat: ordering helps but does not substitute for retrieval once the corpus outgrows the context window, and grounding improves further when the model is asked to quote the relevant passages before answering. Operational check: hold a fixed set of manual questions with known answers, run the template both ways, and score the two orderings against that answer key.