Analytic Report Document

Frame a prompt as an analytical report — introduction through conclusion — when the task is objective analysis rather than conversation.

Models are trained on vast numbers of human-written analytical reports (business, literature, science, law — though legal defense is best left to professionals), which makes the report document archetype easy to leverage per the Little Red Riding Hood principle. Reports follow a familiar structure — introduction, leading to a conclusion, often with a recap — into which gathered information slots naturally as discussion or background sections. Reports favor objective analysis, which lightens the model's cognitive load by not requiring it to simulate social interaction the way an advice conversation document does; this format also suits chain-of-thought prompting well, since analysis-then-conclusion is exactly a report's native shape.

Two techniques specific to this archetype:

A Scope section stating boundaries up front (e.g. "This report focuses solely on novels, excluding self-help books") rather than negotiating exclusions dialogically — models respect such boundaries more consistently when stated in report form than when negotiated turn by turn in conversation.
A clear transition into decision-making mode where analysis ends and conclusion begins, since analysis typically precedes conclusion in this format; without one, the response risks meandering and needing extra parsing to extract the actual verdict. See prompt transition for the general technique.

Write report prompts consistently in Markdown. It's near-universal in training data, so models know it well; it's simple and lightweight to write and interpret; headings define a hierarchy that lets you organize, rearrange, or omit sections while preserving structure; indentation generally doesn't matter outside triple-backtick code blocks, which handle indentation-sensitive content like source code; it renders directly for users; and hyperlinks let the model include easily-parsed, verifiable links.

A table of contents at the start of a long Markdown report prompt helps the model orient, the same way it helps a human reader, and controls the completion in two ways:

Scratchpad sections — adding # Ideas or # Analysis sections before # Conclusion in the table of contents guides the model toward a more informed conclusion, while letting the application ignore or discard the earlier sections when parsing the output.
Signaling the response's end — adding a # Appendix or # Further Reading section after the conclusion, then setting that heading as a stop sequence, ensures the model finishes without wasting compute generating content past the conclusion.

Sources

Prompt Engineering for LLMs (John Berryman), ch. 6