Fourteen Reusable Prompts Survive Changes in AI Products

These reusable prompt patterns describe the work, the evidence, and the human decision, not a product feature, so they survive changes in AI products. Paste your own context where a bracket asks for it, and use an assistant your organization has approved for confidential material.

Put Context Before the Task#

The Master Context Wrapper. Place this before a task-specific prompt when the AI lacks context:

You are helping with a consequential CTO hiring and management decision. Use only the information supplied. Separate facts, estimates, opinions, disagreements, and unknowns. Do not invent company, candidate, legal, regulatory, financial, security, or technical facts. Do not infer protected characteristics, personality, motive, or mental state. Explain assumptions. Provide alternatives and the conditions under which each is sensible. Identify evidence that would change the answer. The final decision belongs to [NAMED HUMAN OR GOVERNING BODY]. Company Context Pack: [PASTE]. Current decision: [STATE].

Separate Facts From Claims Before Debating#

The Evidence Separator. Use when a discussion has produced a persuasive narrative:

Convert the following material into five sections: directly observed facts, supplied metrics and their definitions, participant claims, interpretations, and unknowns. For each interpretation, identify the facts that support it and at least one alternative explanation. Do not resolve disagreement. List the next evidence that would most reduce uncertainty. Material: [PASTE APPROVED NOTES].

Best use. Role diagnosis, candidate debriefs, 90-day reviews, and difficult executive conflicts.

Failure to watch. The AI may classify a repeated opinion as fact because several people said it. Repetition is evidence of shared belief, not proof of the underlying claim.

List Every Assumption and Its Smallest Test#

The Assumption Ledger.

Create an assumption ledger for [DECISION]. For each assumption, state who appears to hold it, why it matters, current evidence, confidence stated as low/medium/high with explanation, date by which it must be tested, smallest useful test, and consequence if false. Separate assumptions about the market, business, technology, organization, candidate, authority, funding, and time. Do not turn confidence into a probability unless data supports one.

Best use. Defining a startup CTO, planning a transformation, choosing fractional versus full-time, and making the final hire.

Generate Three Genuinely Different Models#

The Three-Model Generator.

Produce three genuinely different role or operating models for [SITUATION]. Model A should minimize new executive structure. Model B should use the proposed CTO role. Model C should use a different distribution of authority. For each, show decisions, reporting, team, cost categories, benefits, failure modes, and conditions. Do not create cosmetic variations. End with evidence that would favor one model.

Best use. Resisting the assumption that a CTO is the only answer.

Replace Vague Responsibility Words With Decision Rights#

The Decision-Rights Translator.

Find every vague responsibility word in this text, including own, lead, drive, oversee, partner, support, collaborate, and be responsible for. For each, ask the decision question hidden underneath. Rewrite the text using recommend, decide, approve, fund, execute, verify, accept risk, or veto where the supplied context supports it. Leave a bracketed question where it does not.

Best use. Job descriptions, mandates, founder transitions, and CTO-VPE boundaries.

Argue Against Your Preferred Answer#

The Red Team.

Make the strongest evidence-based case against our preferred answer. Do not invent facts or use generic objections. Identify which assumptions carry the decision, what evidence is missing, who bears the downside, and what early signal would show the choice is wrong. Then make the strongest case for the answer. End with the smallest set of facts that distinguishes the two cases.

Best use. Final candidate choice, platform investment, reorganization, and one-year role review.

Failure to watch. A red team can produce symmetry where evidence is not symmetric. Ask it to state the strength of supplied evidence, not merely equal numbers of arguments.

See the Decision Through Each Executive’s Responsibilities#

The Responsibility Lens.

Analyze [DECISION] from the legitimate responsibilities of the CEO, CTO, product leader, engineering leader, CFO, CISO or risk leader, business owner, customer, and board. For each, state the outcome they protect, evidence they need, authority they hold, and risk they may transfer to others. Do not invent motive. Identify conflicts that require an executive decision rather than more discussion.

Best use. Capacity, customer commitments, security risk, enterprise standards, and transformation.

Assume the Hire Failed and Work Backward#

The Pre-Mortem.

Assume that [HIRE OR PLAN] has clearly failed by [DATE]. Construct eight plausible failure paths grounded in the supplied context. For each, show the sequence of events, earliest observable signal, person able to act, prevention, and recovery. Include failure caused by the person, role design, company behavior, and external change. Do not treat failure as proof of bad intent.

Best use. Before offer, before founder transition, and before a major mandate change.

Classify Choices by How Hard They Are to Reverse#

The Reversibility Test.

Classify the choices in [PLAN] as easily reversible, reversible with material cost, or difficult to reverse. For each, state lock-in source, decision deadline, information value of waiting, cost of delay, and a smaller reversible step. Highlight vendor, architecture, organization, customer, data, safety, and public-commitment lock-in.

Best use. First 90 days and any new CTO’s large program.

Name What Stops When Something Starts#

The Stop-Doing Test.

For every new outcome, program, meeting, platform, standard, or role in this plan, identify the capacity, attention, or money required. Name current work that must stop, narrow, or move. If nothing stops, mark the item unfunded. Identify who must decide the tradeoff and the business consequence.

Best use. Day-90 agenda, enterprise portfolios, SME transformations, and scale-platform programs.

Test Whether a Candidate’s Result Will Transfer#

The Transfer Test.

Analyze this candidate evidence. State the result, candidate contribution, authority, surrounding support, company conditions, and skill demonstrated. Compare those conditions with our role. Identify what is likely to transfer, what may not, and what interview or reference evidence would reduce uncertainty. Do not downgrade or upgrade evidence solely because of company brand, title, or size.

Best use. Candidates from famous companies, adjacent industries, or different stages.

Compare the CTO’s Calendar With the Mandate#

The Role-Drift Monitor.

Compare the current CTO calendar, decisions, and outputs with the mandate. Classify work as core mandate, temporary coverage, unplanned drift, or work that should belong elsewhere. Estimate which mandate outcomes are crowded out. For each drift item, propose stop, delegate, time-bound, or formally revise. Identify company behavior contributing to the drift.

Best use. Monthly CEO-CTO discussion and 90/180/365 reviews.

Turn a Technical Update Into a Board Brief#

The Board Translator.

Convert this technical update into a board decision or oversight brief. Use: business outcome, material change, evidence and uncertainty, options, management recommendation, risk accepted, decision or oversight requested, and next signal. Remove operational detail that does not affect the board’s role. Preserve bad news and uncertainty.

Best use. Incidents, platform investments, cybersecurity, AI programs, and transformations.

Check a Prompt Before You Trust It#

The Prompt Quality Check. Before using any important prompt, ask:

• Does it provide the relevant company context?
• Does it identify the decision and human owner?
• Does it distinguish fact, opinion, and unknown?
• Does it request alternatives and conditions?
• Does it ask for evidence that changes the answer?
• Does it protect confidential and personal information?
• Does it prevent the AI from making the employment decision?
• Does it request an artifact someone can review?

If several answers are no, improve the prompt before trusting the output.