Build Your Own Prompt Test Sheet | Follow-Along Exercise

What you will make

A text deliverable that fits the task, a fact-check sheet and a reusable prompt.

You will need

• A chat, image or video tool suited to the task (PromptKite does not provide model credits)
• Your own input materials and a folder for saving results

Steps

1. Define the inputs for this exercise

   The demo inputs for this exercise are: Template: Generate a mug product image: a white ceramic mug on a light gray surface, soft side lighting, 4:5, white space above, no text. The real product's appearance should follow the provided reference images; Task: generate product images. First confirm whether these conditions fit your situation; replace “paste your material” in the example with real material. Write the goal as a deliverable you can check against.

2. Run it once for your scenario

   Replace the placeholder material in the example with real material you can provide, then give the complete prompt to a chat tool. Redact personal or business information as needed first; if the model asks about missing facts, supply the material rather than asking it to make things up. Focus of this task: define success criteria for the task and design three types of input: normal, missing information and conflicting requirements.

3. Check against the criteria and revise

   Check the output against the input: “Do the tests include failure scenarios, and do the conclusions go beyond the sample? Verify names, numbers and qualifiers in the input.” Tag each conclusion with its supporting material; move anything unsupported to a “to be confirmed” section. Pick one wording or structural problem to fix and compare the old and new versions, rather than just asking for it to be “more professional.”

4. Save and review

   Save the final result, the prompt, the source of your input material and one reason for a change. Name files with the date plus the task name; save reusable versions to your PromptKite Prompt Pocket. Next time, change one condition and try again, noting whether the result still holds.

Common pitfalls

• The input is still placeholder text, so the output looks empty — Replace placeholders such as “paste your material” in the example with real material; state clearly that missing facts are unknown instead of asking the model to fill them in.
• The result doesn't meet the standard for this scenario — Check the output against the input: “Do the tests include failure scenarios, and do the conclusions go beyond the sample? Verify names, numbers and qualifiers in the input.” Tag each conclusion with its supporting material; move anything unsupported to a “to be confirmed” section. Pick one wording or structural problem to fix and compare the old and new versions, rather than just asking for it to be “more professional.”

Practice template

Complete the task using the following inputs: the template is “{subject}”; the task is to generate product images. Define success criteria for the task and design three types of input: normal, missing information and conflicting requirements. Provide a per-run log: input, model name, date, output issues, changes and conclusions. Change only one main factor per round, and don't treat a single success as reliable capability. Output in this order: understanding and assumptions, ready-to-use text, information I need to confirm, and next steps. Use clear, short sentences and avoid empty slogans; if key inputs are missing, ask at most three specific questions instead of filling in facts with guesses. Finally, check item by item whether it meets my purpose, and state clearly which results still need human review.