This guide gives you a reusable starting pattern. It is designed to help you see the work more clearly; it is not a substitute for judgment, source checking, or responsibility for the result.
Prepare the inputs before you ask for output.
The model only sees what you give it. Spend a few minutes naming the reader, desired result, and uncertain information. This makes a first draft easier to assess and reduces the need for decorative rewriting later.
Give the task a useful brief.
Help me turn these project notes into a one-page decision memo. Decision to make: [state one decision question] Decision owner and audience: [who will decide and who will read] Source notes: [paste notes with source labels and dates] Known constraints: [time, budget, policy, technical, or relationship limits] First, separate the notes into evidence, assumptions, options, and unresolved questions. Then draft a memo with: (1) decision question, (2) relevant context, (3) evidence with source labels, (4) options and trade-offs, (5) recommendation only where the evidence supports it, (6) risks and unanswered questions, and (7) the next confirmation needed from the decision owner. Do not invent facts, source links, costs, approvals, owners, or dates. Label any inference as an inference.
Replace every bracketed field with your real context. Read the output before reuse.
Four steps that keep the result usable.
- 1
Name one decision that an accountable person can actually make.
- 2
Paste notes with their source labels and dates, rather than a polished summary.
- 3
Ask AI to separate what is known from assumptions before it recommends anything.
- 4
Have the decision owner verify the evidence, trade-offs, and next step before sending.
Begin with the decision, not the document
A memo becomes vague when it starts with a broad project recap. Write the exact choice the reader needs to make first. That gives every later section a test: does this fact, option, or question help the owner decide? If it does not, move it to an appendix or leave it out.
Give each claim a trail back to the note
Use short labels such as [Budget call, 12 Aug] or [Customer interview 04]. The model can then retain a trace beside important claims instead of converting several notes into a confident paragraph with no visible origin. This makes the memo much faster for a decision owner to verify.
Keep options and recommendations separate
An option describes a possible path and its stated trade-offs. A recommendation states which path appears best and why. The difference matters when the notes are incomplete: AI can organize options, but it should not turn a partial preference into an approved recommendation.
End with the question that unlocks the next move
A useful memo does not pretend every uncertainty is solved. Close with the smallest confirmation, evidence check, or owner decision that would allow work to move. A visible unresolved question is more useful than a made-up deadline or approval.
A worked example: build or buy
A product team deciding whether to build or buy a reporting component has notes from a spike, a vendor trial, and a finance conversation. The memo keeps the decision question at the top, groups the spike evidence under “options”, and leaves the recommendation conditional: “buy if the trial data confirms the query limit”. The owner can then act on one check instead of re-reading the notes.
Keep the memo a decision aid, not a record
The memo exists to make one decision easier. Every paragraph should survive the test “does this help the owner decide?”. Facts that only provide background belong in an appendix or the source notes. If the evidence is thin, say so and name the check that would settle it rather than forcing a confident recommendation.
Run a human check.
- Can the reader identify one decision owner and one decision question?
- Does every material claim retain a source label, date, or clear “needs verification” marker?
- Are options described separately from any recommendation?
- Did the draft avoid inventing a cost, date, approval, owner, or source?
- Is the next confirmation specific enough for the decision owner to act on?
AI is strongest here when it makes missing information, structure, and options easier to see. The moment an output becomes a claim, commitment, or decision, bring a person back into the loop.