Remove ambiguity from the task
Specify the audience, purpose, source material and expected form. Ask one coherent question or break a complex task into stages. As in questionnaire design, vague terms and several questions combined can produce answers that are difficult to interpret. State what a useful output must contain and what should remain uncertain.
Avoid leading the answer
A prompt that assumes a conclusion can generate polished support for a weak premise. Ask for alternatives, missing evidence and conditions that would change the recommendation. For research work, preserve the distinction between participant evidence and the assistant's interpretation. Generated opinions are not substitutes for customer interviews.
Test with representative inputs
Compare outputs against a clear rubric and include difficult or incomplete examples. Record prompt changes and reviewer effort so improvements are observable. Nancy's featured writing connects survey design with prompt quality; apply that principle by making the task precise while retaining human judgment over the result.
Put the decision into practice
- State the decision and audience.
- Define the approved evidence.
- Avoid embedding an unsupported conclusion.
- Request uncertainty and alternatives.
- Evaluate representative outputs with a rubric.
Bring the business question, the evidence already available and the constraints that matter. Discuss your next step with Nancy Chou.