One prompt, tested against five inputs, shipped as a template

The situation

Take one thing you genuinely need an AI assistant to do repeatedly — reply to enquiries, summarise reports, draft product descriptions. You will build the prompt the way the course teaches: from vague to structured, with real examples, with a strict output format whose reliability you measure, split into steps, saved as a template, and tested against five inputs including two awkward ones.

What you'll be able to show

• Structure a prompt under the four headings and show where the value sits
• Use real examples to set a style, and compare against description
• Measure how often a strict format comes back clean
• Build a test set and score two prompt versions against it

What you need

• Any AI assistant
• Three real examples of the output you want (actual ones, not invented)
• A document for the template and the test results

Tasks

1. 1Vague, then complete
   Write your request badly on purpose and run it. Then list every question a competent colleague would have asked first, answer them all in the prompt, and run again. Keep both outputs.
   A correct result: Two prompts, two outputs, and the list of questions you had to answer.
2. 2The four headings
   Rebuild the prompt under the four headings from lesson 2, labelled. Run it. Delete only the context section and run again. Write two sentences on what the context was worth.
   A correct result: The structured prompt, and the with/without comparison with two sentences.
3. 3Examples versus description
   Put your three real examples in the prompt and ask for a fourth on a new input. Then remove the examples, describe the style in words instead, and compare which is closer to what you wanted.
   A correct result: Two outputs and a verdict on which approach won, with the reason.
4. 4Strict format, measured
   Write a version that must return only a three-column table with no surrounding text. Run it five times. Count how many came back clean. That number is your reliability.
   A correct result: Five runs, a count out of five, and one sentence on what that means for automation.
5. 5Split into steps
   Split the task into at least three prompts, correcting the output between each. Compare the final result with the single-prompt version.
   A correct result: The step sequence and a comparison verdict.
6. 6The template and its notes
   Turn your best prompt into a template with marked slots. Write the four notes: purpose, template, one good output, what it is not reliable for. Use it twice with different inputs.
   A correct result: A template with slots, four notes, and two uses.
7. 7The test set
   Write five realistic inputs including two awkward ones, and your success criteria first. Run two versions of your prompt against all five and score them. Record at least one change that helped one case and hurt another.
   A correct result: A 2×5 score table, the criteria written before running, and the trade-off you found.

What to hand in

The prompt evolution (tasks 1–5), the template with its four notes, and the test-set score table with the trade-off.

How it is marked