How to write better AI instructions

Good instructions make the job and the standard visible. They do not need to be theatrical or overloaded with tricks.

Key takeaways

• Define the outcome and audience before specifying the format.
• Provide approved context, constraints and examples.
• Ask the system to expose assumptions and unanswered questions.

Describe the outcome

State what the work should enable, who will use it and what a good result must contain.

Provide reliable context

Give the source material, definitions and boundaries needed for the task while excluding unnecessary sensitive data.

Ask for inspectable output

Request assumptions, unresolved questions and source links separately so a reviewer can check the work efficiently.

Use a reusable instruction structure

A practical brief covers outcome, audience, source material, constraints, required format and review criteria. Add examples only when they clarify the standard.

Tell the assistant what to do when information is missing. Asking it to list gaps is safer than encouraging a plausible guess.

Improve through diagnosis

When output disappoints, identify whether the problem came from missing context, conflicting constraints, weak evidence or an unsuitable task. Revise the brief rather than adding random prompt tricks.

Action checklist

1. State the decision or action the output should support.
2. Describe the audience and required tone.
3. Attach approved sources and definitions.
4. List constraints and prohibited assumptions.
5. Request unresolved questions and a reviewable format.

A sensible next step

Save one successful instruction as a team template with placeholders, an owner and a review date. Continue with the practical AI recipes, compare the reviewed AI tools or use the AI Finder to narrow your next decision.