# The Prompting Craft

By [DYLIT Chronicles](https://dylit.info/user/dylitmediabuzz)

[Everything AI - beyond the hype](https://dylit.info/pr/everything-ai-beyond-the-hype/6a9efac02e92664f4d50cf9d) > [The Prompting Craft](https://dylit.info/ch/the-prompting-craft/6ab358fa99638e119f592c7d)

Before and After, Example: Summarizing a Report The weak prompt "Summarize this report." The strong prompt "Summarize the attached report for a VP with only 90 seconds to spare. No more than five bullets. Open with the decision that is needed, then the supporting evidence. Call out anywhere the report's own data seems to contradict itself." What changed Four things. - The reader is named, and given a time budget. A length limit is set. The order of priority is specified, decision first. A critical read is requested, not just a condensation. Why each one matters The reader . "Summarize" has no natural audience, so the model writes for nobody in particular, which in practice means it writes for a patient generalist. A VP with ninety seconds is a completely different job. It changes what gets cut. The length. Five bullets forces ranking. Without a limit the model tends to preserve the document's own proportions, giving equal weight to the important section and the procedural one. A cap makes it choose. The order . Most summaries reproduce the source's structure, which usually buries the decision at the end where the original author put it. Saying "open with the decision that is needed" inverts that. Your reader gets the point in sentence one. The critical read . This is the addition that does the most work and gets used the least. You are asking the model to do something a summary does not normally do: look for internal inconsistency. Reports contradict themselves more often than people expect, usually because different sections were written by different people from different data pulls. The model is good at catching this, and will not do it unless asked. One variation to try Add a second pass on your own work: "Now list the three things a skeptical reader would push back on first." Run it on your own report before you circulate it. What comes back is a reasonable approximation of the questions you will actually get, and you have the chance to answer them in the document instead of in the room. The underlying move The weak version asks for a task. The strong version specifies a reader, a constraint, a priority order, and a standard. None of that is clever phrasing. It is just deciding what you want before you ask, which is most of what separates a usable AI output from a generic one.