Product management is full of documents that look like decisions: roadmaps, PRDs, briefs, research summaries, and launch plans. AI can produce all of them quickly. That is precisely the risk. A fluent document can hide weak evidence just as easily as it can clarify strong evidence.
The prompts below use AI as a decision assistant. They require sources, preserve uncertainty, and end with a human decision. Pair them with the broader customer-research prompt patterns and project-planning patterns.
Product Discovery
1. Synthesize interview evidence
Synthesize these customer interviews for the decision: [DECISION]. Group
observations into needs, current workarounds, triggers, barriers, and desired
outcomes. Every theme must cite interview IDs and include a count. Preserve
contradictory evidence and minority views. Do not infer market prevalence from
this sample.
Interview notes: [ID-LABELED, APPROVED NOTES]
2. Build a problem brief
Create a one-page problem brief from the evidence below. Include target user,
situation, current behavior, pain, consequence, evidence strength, assumptions,
and what we still need to learn. Do not propose features.
Evidence: [SOURCE-LABELED EVIDENCE]
3. Separate requests from underlying needs
Review these feature requests. For each, distinguish the requested solution
from the likely job, constraint, and desired outcome. Cite the original request,
label the inferred need as a hypothesis, and suggest one validation question.
Requests: [REQUESTS WITH SOURCE IDS]
4. Create a research discussion guide
Design a 30-minute discovery interview about [PROBLEM AREA]. Start with recent
behavior, then explore workflow, alternatives, consequences, and decision
criteria. Avoid leading questions, hypotheticals, and pitching our solution.
Add the purpose of each question and two neutral follow-ups.
Strategy and Prioritization
5. Frame strategic options
Turn this situation into three genuinely different strategic options. For each,
show target segment, value proposition, capability required, evidence supporting
it, evidence against it, major risk, reversible first step, and a kill criterion.
Do not rank the options.
Context and evidence: [CONTEXT]
6. Pressure-test a roadmap item
Act as a skeptical product review panel. Pressure-test this roadmap item against
customer value, strategic fit, evidence quality, opportunity cost, dependencies,
adoption risk, and measurable outcome. Return the five questions that must be
answered before commitment.
Roadmap item: [ITEM]
Strategy: [STRATEGY]
Evidence: [EVIDENCE]
7. Prepare a prioritization packet
Apply this supplied prioritization framework to the candidate items. Show the
input behind every score, flag missing inputs, and run a sensitivity check on
the two most uncertain estimates. Do not replace missing numbers with guesses.
Framework: [RICE, WSJF, OR TEAM METHOD]
Candidates: [DATA]
8. Write a decision memo
Draft a decision memo for [DECISION]. Include recommendation, alternatives,
decision criteria, supporting evidence, counterevidence, assumptions, risks,
reversibility, and the date or signal that triggers review. Keep facts and
judgment visibly separate.
Requirements and Delivery
9. Draft an outcome-based PRD
Draft a lean PRD from this approved problem brief. Include problem, target user,
desired outcome, non-goals, user scenarios, constraints, acceptance criteria,
instrumentation, risks, and open questions. Do not invent technical architecture
or commit to a solution not present in the brief.
Problem brief: [BRIEF]
10. Find ambiguity in requirements
Review these requirements for ambiguous actors, undefined terms, hidden states,
missing error paths, conflicting rules, and criteria that cannot be tested.
Quote each problematic line and propose a clarifying question before suggesting
a rewrite.
Requirements: [TEXT]
11. Generate acceptance scenarios
Convert these verified requirements into Given/When/Then scenarios. Cover the
happy path, permissions, empty states, invalid input, retries, partial failure,
and recovery. Trace every scenario to a requirement ID; flag untestable items.
12. Prepare a scope negotiation
Create three scope options for this deadline: minimum viable outcome, balanced,
and full. For each, show user value preserved, work removed, new risk, dependency,
and what evidence we will lose. Do not disguise a date commitment as an estimate.
Outcome: [OUTCOME]
Current scope: [SCOPE]
Constraints: [CONSTRAINTS]
Experiments and Launch
13. Design an experiment
Design the smallest ethical experiment that tests [HYPOTHESIS]. Specify segment,
treatment, comparison, primary metric, guardrail metrics, minimum duration logic,
instrumentation, decision thresholds, and confounders. Identify what the result
would not prove.
14. Pre-mortem a launch
Assume this launch missed its outcome six weeks after release. Generate failure
modes across value, usability, reliability, discoverability, enablement, support,
and measurement. For each, name an early signal, prevention, contingency, and
owner. Prioritize by impact and plausibility, not drama.
15. Build a launch readiness review
Create a launch readiness checklist for [PRODUCT/CHANGE]. Cover product quality,
security/privacy, analytics, support, documentation, sales or internal enablement,
rollback, ownership, and communications. Every item needs evidence, an owner,
and a pass/block status.
16. Draft launch communications by audience
Turn the verified launch brief into messages for customers, support, sales,
executives, and engineering. Keep the core facts consistent while changing
detail and emphasis. Do not add claims, dates, or capabilities absent from the
brief. List any claim that needs legal or product verification.
Launch brief: [BRIEF]
Learning After Launch
17. Diagnose a metric movement
Analyze this metric change without assuming causation. Separate observations,
possible explanations, evidence for and against each explanation, segmentation
checks, instrumentation checks, and the next analysis with highest information
value.
Metric definition: [DEFINITION]
Data summary: [DATA]
Recent changes: [CHANGES]
18. Run a product review
Create a monthly product review from these verified inputs. Cover outcome versus
target, segment differences, qualitative evidence, reliability and support
signals, what we learned, decisions needed, and next bets. Cite a source for
every number and keep activity metrics separate from user outcomes.
The Rule: Ask AI to Improve the Decision Surface
Good product work makes the decision surface clearer: what is known, what is assumed, what conflicts, what options exist, and what would change the choice. AI helps when a prompt demands that structure. It hurts when the prompt is simply “write a roadmap” or “tell me what to build.”
Before sharing any output, trace claims to evidence, verify customer data was handled correctly, and put the accountable person’s name next to the actual decision. For launch-specific material, continue with AI prompts for product launches.
