Intent

A topic tells the model what to talk about; intent tells it what the output is for. Naming the outcome — the decision to be supported, the artefact to be produced, the shape it must arrive in — is usually the single highest-leverage line in a prompt.

Name the outcome

• The job to be done. "Help me decide whether to migrate" produces a different answer from "Explain database migration", even though the topic is identical.
• The artefact. An email, a table, a checklist, a patch, a JSON object — say which.
• The format. Sections, length, ordering, schema. If the output feeds a program, give the exact schema and say that nothing else should be emitted.
• The success test. Tell the model how you will judge the answer, and it will aim at that target.

Techniques

• Show an example. One short sample of the desired output shape conveys more than a paragraph describing it.
• Assign a role only when it changes the content. "You are a tax accountant" shifts vocabulary and priorities; "You are a helpful assistant" changes nothing.
• Say what happens next. "This goes straight into a slide" implies brevity without you having to specify it.

Example

Tell me about our churn numbers.

GOAL: help me decide whether to fund a retention project this quarter.
FORMAT: a 5-row table (segment, churn %, trend, likely cause, suggested action), then one paragraph with your recommendation.
DATA: attached CSV of monthly churn by segment.