Foundations · 5 min read

Give AI the context to do useful work.

A model cannot reliably infer the operating rules inside your head. A useful brief makes those rules visible. The skill is choosing the information the task needs and keeping it current.

What you’ll take away

Build a reusable context brief that helps AI produce work you can evaluate.

Explain the situation and the desired result.

Say who the work is for, what is happening, and what the output should help someone do. “Summarize these emails” leaves the model to choose what matters. “Prepare the information our catering lead needs to confirm tomorrow’s event” gives it an audience and a purpose.

Specify the output format when it serves the work. A short table of requirements, owners, deadlines, and open questions may be more useful than a polished paragraph.

Include the business rules that affect the answer.

Give the relevant policies, constraints, definitions, and examples. If a customer reply depends on your refund policy, provide that policy. If a delivery estimate depends on supplier confirmation, say so.

Tell the model which sources are authoritative and when they were updated. A recent customer request and an old internal note can disagree. Ask the model to surface that conflict rather than quietly choosing one.

Goal and audience
Facts and source dates
Constraints and approved policies
A useful example
Output format and review criteria

Keep the context focused.

More text is not automatically more useful context. Long conversations can mix current instructions with abandoned ideas and stale information. Create a fresh brief when the task or operating assumptions change.

Ask AI to help prepare a compact handoff: confirmed facts, decisions, unresolved questions, and the next task. Review that handoff before reusing it. Keep important business records in a maintained document or system you control.

Ask for uncertainty to be visible.

Give the model a way to handle missing information: label it unknown, cite the source, or ask a question. “Make it complete” can encourage a convincing answer built on guesses.

Request a short account of assumptions and supporting evidence. You need enough explanation to review the result, not an exhaustive internal monologue. Check the cited source yourself when the detail matters.

Example: brief a catering handoff

Use approved event details and your current policy. Replace the bracketed fields before sharing.

Goal: prepare a handoff for our catering lead to confirm an event.
Audience: the person responsible for event delivery.
Sources: [customer request with date], [current event record], [current catering policy].

Create a table of guest count, menu, dietary requirements, delivery location, timing, responsible person, and open questions. For each fact, name its source.

Mark conflicts and unknowns. Do not infer a confirmed booking from an inquiry. Ask the three most important questions needed to finish the handoff.

Replace the bracketed fields before using this with AI.

Try it in your business.

Choose one recent AI task that needed corrections.
Write the five-part brief above for that same task.
Try it again and note which missing facts still caused trouble.

Use the matching template