writing effective instructions
Key facts
- Purpose
- Reusable one-page PRD, full PRD, AI drafting, red-team, and PRD-to-spec templates.
- Template rule
- Start with the one-page PRD. Expand only when the missing detail creates review risk, build risk, launch risk, or measurement risk.
- Best next step
- Understand what each template field is for.
Template
One-page PRD
Use this when the decision is narrow, reversible, and familiar. Keep it to one page unless a reviewer finds a real decision gap.
Reusable block
One-page PRD template
- Problem
- What user, business, or operational problem exists today? Include the current workaround.
- Primary user
- Who experiences the problem and who must act differently after launch?
- Goal
- What outcome changes if this succeeds?
- Non-goals
- What is explicitly out of scope for this release?
- Proposed solution
- What is the product shape, in plain language?
- Requirements
- Three to seven user-visible requirements with acceptance criteria.
- Metrics
- Primary metric, diagnostic metrics, guardrails, baseline, target, and measurement window.
- Risks and open questions
- What could make this fail, who owns each answer, and by when?
Template
Full PRD
Use this when the work is cross-functional, high-risk, or expensive to reverse. The full PRD should still be skimmable: lead with decisions, push evidence and details into structured sections.
Reusable block
Full PRD template
- Executive summary
- One paragraph on the product bet, customer, outcome, and launch decision needed.
- Context and evidence
- Research, data, customer quotes, support signals, sales input, market or operational facts, each with a source.
- Problem statement
- The problem without solution language.
- Users and stakeholders
- Primary users, secondary stakeholders, excluded personas, and jobs-to-be-done.
- Goals and non-goals
- Few measurable goals and explicit boundaries.
- User journeys
- Current workflow, target workflow, and failure or exception paths.
- Requirements
- Functional requirements, acceptance criteria, edge cases, permissions, accessibility, localization, and content needs.
- Success metrics
- Outcome, diagnostic, guardrail, baseline, target, instrumentation owner, and review date.
- Launch and rollout
- Pilot cohort, rollout phases, communications, support readiness, documentation, migration, and rollback.
- Risks and decisions
- Known risks, unresolved questions, owners, dates, and decision log.
Prompt
AI drafting prompt
Paste your notes first. Then paste this prompt. The key instruction is that the model must mark missing information instead of filling it with plausible guesses.
Reusable block
AI PRD prompt
- Input rule
- Use only the notes above. If evidence is missing, write UNKNOWN and explain what source would resolve it.
- Task
- Draft a PRD with problem, users, goals, non-goals, success metrics, requirements, edge cases, risks, launch plan, and open questions.
- Acceptance rule
- Every requirement must include Given/When/Then acceptance criteria.
- Metric rule
- Every metric must include baseline, target, owner, or UNKNOWN.
- Tone
- Write like a practical product manager. Be specific, concise, and honest about uncertainty.
Prompt
AI red-team prompt
Run this after the first draft. It is intentionally adversarial because PRDs often fail through confident omissions.
Reusable block
Red-team prompt
- Claim audit
- List every factual claim that needs a source or owner.
- Ambiguity audit
- Find phrases that different teams could interpret differently.
- Scope audit
- Identify hidden goals, weak non-goals, and accidental commitments.
- Engineering audit
- List missing acceptance criteria, edge cases, dependencies, migration needs, and observability requirements.
- Launch audit
- Find missing support, documentation, rollout, permissions, and rollback decisions.
Template
PRD-to-spec handoff
Use this at the end of the PRD when the team is ready to write a technical spec or hand work to an AI coding agent.
Reusable block
Handoff checklist
- Requirement map
- Each PRD requirement has an ID and a trace link to design, spec, ticket, or test plan.
- Open decisions
- Every open question has an owner and deadline.
- Telemetry
- Events, properties, dashboards, and owners are named.
- Release gates
- Pilot, launch, rollback, and success-review gates are explicit.
- Agent-ready detail
- If an AI agent will implement the work, examples, non-goals, constraints, and done criteria are unambiguous.