Order a report with a given structure. Return JSON {"question":"...","sections":"...","audience":"...","length":"..."}. question — topic of the report; sections — list of sections in order (for example: summary, key findings, details, risks, recommendation); audience — who the report is for and its level (beginner/expert); length — desired volume (for example: output up to 5 lines, about one page in total).
Mechanics: without a given structure, the model writes linearly - as it thinks, it pours. The reader is drowning. When you define sections, the model puts what it finds on shelves and notices the gaps itself (“but I didn’t find any risks”). Levers. The order is “conclusion first”: first a summary, then a rationale. A busy person reads the first five lines and already makes a decision. Explicit sections: findings, details, risks, recommendation. Audience level: “explain for a beginner without jargon” changes both depth and language. Inside reception: the section “what remained unclear / what was missing data” is required. It turns a glossy report into an honest tool and tells you where to dig next. A model that is allowed to acknowledge the gap is less likely to invent. Second trick: set the volume in lines or paragraphs, and not “shorter”. The model understands the vague “briefly” as it suits her; “output up to 5 lines, just one page” gives a predictable result. A typical mistake: not setting the format and then complaining about the sheet. The structure is your responsibility, not the model's guess.
Unlock access to submit solutions for instant AI review.