Describe the AI, what report to collect. Return JSON {"data":"...","audience":"...","sections":"..."}: data — context; audience — for whom is the report and what is important to this person; sections - what short sections it consists of (for example: main conclusion, 3 key numbers, what to improve). Keep sections short, without a wall of text.
A report is not “all the numbers”, but an answer to the reader’s question. AI structures exactly the way you set the sections; If you didn’t ask, you’ll get a wall of text where the conclusion is lost. Mechanics: the model fills your frame, so the frame is more important than the content. Levers: name the audience (the boss, not the analyst - he needs solutions, not methodology), ask for 3-4 short sections, ask for the main conclusion FIRST. Insider: the rule of the inverted pyramid - first the conclusion and recommendation, then the evidence numbers. The busy reader gets to the end of the first line, and it should answer the “so what?” Second technique: ask to accompany each number with one word-evaluation - “good / bad / normal.” The raw “average rating of 3.8” says nothing, but “3.8 is below the target of 4.2, alarming” says everything and pushes to action. A typical mistake: asking “make a beautiful detailed report” without sections and addressee - you will get a long, streamlined and useless solution.
Unlock access to submit solutions for instant AI review.