AI fundamentals · 5 min read

A better brief. A more useful answer.

A vague request leaves the model to guess what matters. A good brief makes the task explicit and gives you a way to judge the result. You can use this structure in a chat, an agent instruction or a handover to a colleague.

By Matrix Academy · Published 28 October 2026

Start with the decision, not the document

“Summarize these notes” describes an activity. “Help the product team decide which onboarding problem to investigate next” explains the purpose. A purpose helps you choose what to include and what to leave out.

Name the audience and the action they will take. If you cannot explain how someone will use the output, make the task smaller before you ask AI to do it.

Give only the context the task needs

Supply the relevant source material, definitions and examples. Label each source so the output can point back to it. Keep facts separate from your interpretation.

Use fictional or public material while practicing. For work information, follow your organization’s actual policy and approved tools. More context is not automatically better if it adds irrelevant or sensitive information.

Make the boundaries concrete

Specify the format, length and things the model must not invent. Ask it to flag missing information rather than filling the gap. Say when it should ask a question or stop.

Here is a fictional example you can adapt:

Goal: help a product lead choose one onboarding question to investigate.
Sources: the five fictional interview notes below, labeled A–E.
Output: a one-page brief with recurring problems, supporting excerpts, open questions and a suggested next interview.
Boundaries: do not invent customer counts, quotes or causes. Label any interpretation as a hypothesis.
Done means: every important claim points to a source note, and the uncertainty is visible.

Judge one answer before scaling the task

Compare the result with the brief. Can you trace the claims? Does it answer the actual decision? Is it missing an important counterexample?

Change one instruction at a time and compare the outputs. Keep the version that improves the result, along with a note about where the brief still fails. A reusable prompt is useful only if you understand its limits.