20 November 2026
Give prompts a measurable purpose, a source and a way to check the response.
Getting started
Start with an actual task, such as summarising a policy for a new learner. State the audience and the decision the summary should support. Supply the relevant source and ask the model to flag missing information instead of guessing.
Putting it into practice
Ask for a specific format, then run the same prompt on two examples. Compare factual accuracy, readability and omitted details. Save the version that performs consistently and record where a human still needs to review the result.
Try it yourself
Use one approved document to test an open-ended prompt and a structured prompt. Record which changes led to fewer omissions.