GPT 6 prompts for careful, repeatable work

Plan my week from [COMMITMENTS], [FIXED EVENTS], [ENERGY PATTERN], and [ONE PRIORITY]. Separate fixed events from flexible work. Protect meals, sleep, and transition time. Offer a first draft, then list the two tradeoffs that matter most. Do not invent availability.

Expected output: A day-by-day table, assumptions, and two questions before any schedule is treated as final.

• All fixed events appear once.
• The plan contains buffers.
• Unknown availability is labeled.

Help me decide between [OPTIONS] for [GOAL]. Use these criteria and weights: [CRITERIA]. Ask only for missing information that could change the ranking. Show evidence, assumptions, downside, and a reversible next step. Do not present a score as an objective truth.

Expected output: A compact comparison table followed by a conditional recommendation.

• Weights are visible.
• Evidence and inference are separated.
• The next step is reversible.

Explain [IDEA] to [AUDIENCE] who already knows [PRIOR KNOWLEDGE]. Start with a one-sentence answer, then use one concrete example and one boundary case. Define necessary terms. Mark uncertainty and avoid a claim stronger than the evidence.

Expected output: Answer, example, boundary case, and a short “what this does not mean” section.

• The audience is respected.
• Every technical term is defined.
• Limits are explicit.

Process these meeting notes: [NOTES]. Extract decisions, open questions, action items, owner, due date, and dependencies. Preserve uncertainty when the notes do not assign an owner or date. Quote a short source phrase for each decision.

Expected output: Decisions, action table, unresolved questions, and a follow-up message draft.

• No owner is invented.
• Dates are copied or marked unknown.
• Decisions are traceable to notes.

Rewrite [UPDATE] for [STAKEHOLDER TYPES]. Keep factual content stable. For each audience, lead with the decision or impact, then risks, required action, and timing. Keep unknowns visible and flag any sentence that needs confirmation.

Expected output: One version per audience plus a shared fact ledger.

• Facts do not drift between versions.
• Audience specific actions are clear.
• Unverified claims are tagged.

Convert [PROCESS DESCRIPTION] into a checklist for [ROLE]. Include prerequisites, ordered steps, decision points, evidence to retain, and a stop condition. Separate required controls from helpful suggestions. Do not fill gaps with assumptions.

Expected output: A numbered checklist, decision tree in prose, and evidence record fields.

• Prerequisites come first.
• Stop conditions are visible.
• Each control has evidence.

Build a source matrix for [QUESTION]. Use only the supplied sources: [SOURCES]. For each claim, record source, publication date, exact supporting passage in your own words, confidence, contradictions, and what remains unknown. Distinguish official fact, reported observation, inference, and anecdote.

Expected output: Claim matrix followed by a short evidence-bounded answer.

• Every claim has a source or is labeled inference.
• Dates are recorded.
• Conflicts are not silently resolved.

Synthesize these interviews: [TRANSCRIPTS]. Group observations by job, pain, workaround, desired outcome, and contradiction. Include representative paraphrases with interview IDs. Count only the supplied interviews and label the sample as directional, not representative.

Expected output: Theme table, counterexamples, opportunity hypotheses, and follow-up questions.

• Interview IDs are retained.
• No prevalence claim exceeds the sample.
• Counterevidence is included.

Compare [OPTIONS] for [USE CASE] using current primary sources where available. Record price date, eligibility, limits, region, and important exclusions. Keep observed facts separate from your recommendation. If a page is dynamic or inaccessible, say so and mark the field unverified.

Expected output: Comparison table, evidence notes, unknowns, and a conditional recommendation.

• Region and date are explicit.
• Price is not treated as permanent.
• Unverified fields remain visible.

Review this diff for [REPOSITORY CONTEXT]: [DIFF]. Prioritize user impact, correctness, security, data loss, compatibility, and test gaps. Cite file and line. Do not praise style unless it affects behavior. If behavior cannot be established from the diff, state what must be checked.

Expected output: Findings ordered by severity, questions, and a focused verification plan.

• Findings are actionable.
• Severity reflects impact and likelihood.
• Unknown behavior is not asserted.

Diagnose [BUG REPORT] in [STACK]. First restate the observable symptom and expected behavior. List competing hypotheses, the cheapest discriminating check for each, required logs, and a minimal reproduction. Do not implement a fix until the failing behavior is isolated.

Expected output: Symptom, hypotheses, ordered checks, reproduction, and evidence needed.

• Expected behavior is explicit.
• Checks distinguish hypotheses.
• No fix is claimed without evidence.

Draft a contract for [ENDPOINT] serving [CLIENTS]. Include request, response, errors, authentication assumptions, idempotency, pagination, limits, compatibility, and examples. Mark every decision that still requires product or security approval. Avoid inventing provider behavior.

Expected output: Contract table, examples, error semantics, and unresolved decisions.

• Success and failure shapes are complete.
• Security assumptions are stated.
• Open decisions are tagged.

Develop [NUMBER] distinct concepts for [PROJECT] aimed at [AUDIENCE]. Constraints: [CONSTRAINTS]. For each, give premise, emotional effect, visual or structural device, risk, and a small first test. Avoid near duplicates and avoid claiming that an untested concept will perform.

Expected output: Concept cards followed by a recommendation for the next test.

• Concepts differ in mechanism.
• Constraints are honored.
• Performance claims are hypotheses.

Outline a [DURATION] script about [IDEA] for [AUDIENCE]. Include hook, setup, escalation, turn, ending, visual or audio cues, and a single call to action. Keep the voice [VOICE]. Offer two alternate openings and explain the tradeoff in each.

Expected output: Beat sheet, draft script, alternate openings, and production notes.

• Timing is plausible.
• The turn follows from setup.
• The CTA is singular and clear.

Edit [DRAFT] against this brief: [BRIEF]. Return a clean revision and a change log grouped by clarity, rhythm, audience fit, factual risk, and voice. Keep lines that already work. Flag brief conflicts instead of silently choosing one interpretation.

Expected output: Revised draft, concise change log, and questions for the owner.

• Voice remains recognizable.
• Changes map to the brief.
• Factual risks are called out.