Write two or three real prompts of the task type you picked. Same shape, different content — two emails, not an email and a deck. Keep them identical across contestants.
Two or three prompts of the same type. Same shape, different content.
If the full battery feels heavy, start with one real prompt and two contestants labeled A and B. That already earns the rest of the course. Add variants when you want stronger signal.
Save outputs with letter labels only. Hide which Copilot option produced which. You will judge in a separate chat in Lesson 4.
Core principles
- Write the prompts before you run them. Do not improvise contestant B’s prompt.
- Label outputs A and B (and C if you have a third). No model names on the files.
- Run each contestant once before you start optimizing — premature iteration confuses the comparison.
- Confidential content stays in Copilot. Strip names if you will paste outputs into a judging chat.
Check yourself
Why run more than one prompt of the same type?
A single win can be luck or a lucky brief. Two or three of the same type tells you if the house style actually fits the job.
Do this in Copilot
Run at least one prompt on both contestants. Save the outputs as A and B with the mapping on a separate note you will not look at while scoring.
Fill the brackets with real work. Run A, then B, in separate Copilot chats.
Draft the battery prompts
I will taste-test Copilot on this task type: [TYPE]. Audience: [WHO]. What “great” looks like: [ONE LINE]. Draft 2–3 prompts of the same shape with different content I can actually run this week. Use [BRACKETS] for the parts I fill. Keep each prompt short enough to paste twice unchanged. After the prompts, remind me to run them identically on contestant A and contestant B and to save outputs with letter labels only.
- Mapping note (keep separate): A = [option], B = [option], web grounding on/off, reasoning depth if you set it.
Did you run this in Copilot? Mark complete when you have tried it.
RecordedNavigate: press j for next lesson,k for previous.