Lesson 2 gets you to good. This lesson gets you from good sometimes to good reliably on real tasks, techniques that make strong prompts repeatable.
Good sometimes vs good reliably.
Core principles
- Persona seeding: not just role but disposition and priorities, "HR partner who asks hard questions about timelines and communication gaps."
- Priming with examples: one-shot or few-shot, paste what good looks like; removes ambiguity faster than description.
- Negative constraints: tell Copilot what not to do, no hedging, no bullet points, no synergy-speak, fixes recurring failure modes.
- Targeted refinement: name location, what is wrong, what you want, not "make it better." Three to five iterations is normal; after eight, rewrite the original prompt.
- Meta-prompt: ask what information the model needs before the real request, use its list to build the final prompt.
- Temperature through language: "stick to source material" vs "explore unconventional angles."
Check yourself
What is the purpose of a meta-prompt?
A meta-prompt asks Copilot what it would need to give you a useful answer, before you give the actual task. Use the list it returns to build the real prompt. It surfaces gaps in your brief faster than a bad first draft does.
Do this in Copilot
Try persona seeding or priming with examples on a real task this week.
Answer the model's questions, then run the real prompt with that context.
Meta-prompt
I need help drafting a difficult conversation email to a vendor who missed three deadlines. Before I give details, what five pieces of information would make your response actually useful? List them.
- Negative constraint example: Draft this email without apologizing for the ask. State it plainly.
- Refinement example: The third paragraph is too abstract. Rewrite with one concrete example. Keep under 150 words.
Did you run this in Copilot? Mark complete when you have tried it.
RecordedNext lesson: Building and Using a Personal Prompt Library →
Navigate: press j for next lesson,k for previous.