Help me get a reliable result for this task: [TASK]. The result will be used by [AUDIENCE AND USE], and the relevant context is [CONTEXT AND SOURCE MATERIAL]. My constraints are [CONSTRAINTS], and I will judge success by [SUCCESS CRITERIA]. Do not start producing the final deliverable immediately. First act as a requirements editor. Extract the objective, audience, deliverable type, decisions the result must support, hard constraints, soft preferences, available evidence, and unknowns. Ask up to 3 clarifying questions only when an answer would materially change the output. Otherwise state reasonable assumptions clearly and continue. Then write a compact “working brief” of no more than 250 words and ask me to approve or correct it. After approval, produce the requested deliverable in the format that best fits the task. Follow these rules:
1. Use source material only for factual claims; distinguish facts, calculations, recommendations, and assumptions. 2. Do not invent names, citations, dates, numbers, quotes, capabilities, or personal experience. If a needed fact is absent, use a clearly marked placeholder or state the limitation. 3. Make the output usable without exposing internal reasoning: provide the result, essential rationale, and any decision-relevant tradeoffs. 4. Match the stated audience’s knowledge level, length limit, voice, and channel. 5. When there are multiple viable options, compare them against the success criteria rather than choosing silently. Before finalizing, run two checks: a requirements trace showing how each success criterion is met, and a factual-risk scan that lists unsupported or time-sensitive claims for me to verify. End with a short “What I need from you next” section only if an unresolved input prevents a complete answer.
Fill in
| Placeholder | What to enter | Example |
|---|---|---|
| [TASK] | Describe the exact work you want the AI to complete. | Draft a one-page recommendation for choosing a customer-support platform. |
| [AUDIENCE AND USE] | State who will use the result, where it will appear, and what decision or action it should support. | Our COO will read it before a vendor meeting and needs a decision recommendation. |
| [CONTEXT AND SOURCE MATERIAL] | Paste the relevant notes, data, links, requirements, or background information. | We have 12 agents, 18,000 tickets per month, use HubSpot, and need email plus chat; notes from demos are pasted below. |
| [CONSTRAINTS] | List non-negotiable limits such as length, deadline, budget, voice, policy, or format. | Maximum 700 words; no claims beyond the demo notes; recommend no more than two finalists. |
| [SUCCESS CRITERIA] | Define the specific qualities that would make the result successful. | Clear recommendation, operational tradeoffs, estimated implementation risk, and questions for the vendor meeting. |
How to use
- Paste source material that can support the claims you expect the answer to make.
- Set success criteria that can be tested, such as a word limit, decision, or required comparison.
- Approve the working brief before asking for a long deliverable when the stakes or scope are high.
- Follow up with: “Revise this for [AUDIENCE], preserving the facts and changing only the structure and tone.”
Variations
Research answer
Use when you need a sourced answer to a factual question.
Research [QUESTION] for [AUDIENCE] using [ALLOWED SOURCES OR LINKS] and within [TIMEFRAME]. First define the question’s key terms and what evidence would answer it. Then produce a concise answer with dated sources, a distinction between evidence and inference, conflicting findings, and unanswered questions. Do not cite sources you cannot verify or treat a search snippet as evidence. Ask up to 3 questions only if scope or source rules are missing.
Decision memo
Use when a team needs a recommendation among options.
Write a decision memo choosing among [OPTIONS] for [DECISION]. Use [FACTS AND DATA], [CONSTRAINTS], and [DECISION CRITERIA]. Produce an executive recommendation, criteria-weighted comparison, key assumptions, risks, implementation implications, and a decision request. Do not fabricate costs, timelines, or stakeholder views. Flag where data is insufficient to distinguish options and state what would change the recommendation.
Improve draft
Use when you already have a draft but it is not working.
Improve this [DOCUMENT TYPE] for [AUDIENCE AND PURPOSE]: [DRAFT]. Apply [REQUIREMENTS]. First diagnose the three most consequential gaps in structure, evidence, logic, or audience fit. Then provide a revised version and a concise change log. Preserve supported facts and do not add claims, credentials, data, or quotes. Flag any statement in the original that needs evidence before publication.
Tips
- Give the AI the source material and the decision it must support; a polished answer cannot repair a missing factual basis.
- Separate hard constraints from preferences, since requirements like legal language, budget caps, and word limits should not be traded away silently.
- For consequential decisions, request an assumptions list and ask what evidence would reverse the recommendation.
- Use one deliverable per prompt; combining research, strategy, copywriting, and implementation into one request obscures what good looks like.
FAQ
Do longer prompts always get better results?
No. A prompt improves when it contains decision-relevant context, clear constraints, and a defined output, not when it adds unrelated detail.
Should I ask AI to show its reasoning?
Ask for concise rationale, assumptions, and evidence checks. Those are more useful for validating a result than a long internal process narrative.
How can I prevent made-up facts?
Tell the AI which sources it may use, require unsupported claims to be flagged, and verify time-sensitive facts yourself.