Test a Prompt Before You Trust It
A reusable prompt is a small product. Test it across realistic inputs, different users, and more than one run before you depend on it.
Seven Checks for a Reusable Prompt
Goal
Does the output accomplish the intended job, or merely sound plausible?
Repeatability
Run it several times. Note which parts remain dependable and which drift.
People
Give it to someone who did not write it. Watch for assumptions they cannot see.
Conversation
Continue for several turns and check whether the model keeps the original objective.
Models
Try another model or version when portability matters. Compare behavior, not just polish.
Edges
Use incomplete, contradictory, unusually long, and out-of-scope inputs.
Verification
Confirm facts, calculations, citations, and required fields outside the model.
Act as a skeptical reviewer of this prompt. Identify hidden assumptions, ambiguous instructions, missing inputs, likely edge cases, and ways a user could misread it. Then propose the smallest revisions that would make the prompt more reliable. Do not run the prompt yet.
Record the evidence: save the prompt version, model, date, test input, result, and pass/fail reason. A polished output is not the same as a passed test.
Original Source and License
Adapted from the Wharton Generative AI Labs Prompt Library, including its guidance on analyzing, testing, creating, and refining prompts. Prompt materials are credited to Ethan Mollick and Lilach Mollick and offered there under CC BY 4.0.