A Prompt Pattern for Turning Tool Features Into a Usable Checklist

When a new AI tool appears, it is tempting to ask for a complete review in one prompt. The result usually lists features without helping anyone decide what to test. A better prompt gives the model a narrow job and leaves the decision with the person using the tool.

Start with the real task

Write the input, desired output and audience in plain language. “Choose a transcription service for short interviews” is more useful than “compare these tools”. Add the conditions that would make a result unusable.

Ask for evidence-shaped notes

Separate known from untested

Tell the model to label claims as documented, observed or unknown. It should not turn a feature page into proof that a particular file will work.

Request a small test plan

Ask for three representative samples and the exact checks to run. Include one awkward case, not only the easiest demo.

Keep the output editable

Use headings such as Input, Test, Observation and Next step. A short table can be more useful than a persuasive paragraph.

Close the loop with a human result

A guide to comparing AI drafts shows the same principle: define criteria before judging the output. Apply it to your own samples and record where the prompt needs revision.

For responsible use, NIST’s AI risk framework is a solid reference for thinking about oversight and uncertainty. The best prompt does not make a tool sound certain; it makes the next check easier to perform.