The nine-part prompt I hand every new PM
A framework, not a template. Here is what each part is for, and what breaks when you skip it.
I got tired of reviewing prompts that were really wishes. The nine parts exist because each one maps to a failure I watched happen on real client work.
Role and context first, because a model with no context invents one, and the one it invents is generic. Task and constraints next — the constraints are the part people skip, and they are the part that makes output usable instead of merely correct.
Format matters more than people expect. If the output has to land in a Confluence page your team already reads, say so in the prompt, not in a cleanup pass afterwards.
The last three parts — evaluation, escalation, follow-up — are the ones that separate a prompt from a process. How will you know this answer is good? What should happen when the model is unsure? What is the next question you will almost certainly ask?
We benchmark the same nine-part prompt across four models quarterly. The rankings move. The framework does not.