TL;DR: Here are 30 AI prompts for product managers organized across five core tasks: writing PRDs, generating user stories and acceptance criteria, prioritizing with RICE and ICE, running competitive analysis, and synthesizing customer interview notes. Every prompt uses [bracketed variables]. Fill in your product context once and reuse each template across sprints without rebuilding from scratch.
What are the best AI prompts for product managers in 2026?
The best AI prompts for product managers are chained templates tied to specific PM outputs — a PRD section, a user story with acceptance criteria, a RICE score table — rather than open-ended questions. When you ask an AI to "help me with my roadmap," it guesses your product context, your constraints, and your format. When you run a structured prompt with role, context, task, and output format, it produces an artifact you can paste into Confluence with a five-minute edit.
AI prompts for product managers are most valuable where PMs lose time but not judgment: synthesizing interview notes, drafting the first pass of a PRD, structuring acceptance criteria the team can actually test. These are cognitive tasks with a clear output format. The model handles the structure; you supply the product knowledge and make the calls the model cannot.
This guide gives you 30 copy-paste templates across five task groups. For the workflow that chains these prompts together — raw meeting notes through finished spec — see our guide to the PM AI workflow from notes to spec. For keeping product context persistent across sessions so you never re-explain your product from scratch, see Contexts for PMs. All 30 prompts work best saved in a prompt library for product managers with your variables pre-filled.
What do most PM prompt lists miss?
Most AI prompt lists for product managers stop at the artifact level: "write a PRD," "generate user stories," "create a roadmap." That solves the blank-page problem but not the chaining problem. A PRD without a prior problem statement prompt produces vague goals. User stories without acceptance criteria are untestable. A prioritization exercise without scored dimensions is just opinions.
The gap this post fills is the chain. These 30 prompts are designed to flow into each other. You run the problem statement prompt before the PRD prompt. You run the user story prompt before the acceptance criteria prompt. You run the interview synthesis prompt before the competitive analysis. Each prompt consumes the output of the one before it, so the final artifacts share the same ICP, the same success metric, the same product framing throughout. That internal consistency is what makes AI-assisted PM work feel like a workflow rather than a collection of one-off tasks.
How do I write a PRD with AI?
AI cuts the time to draft a Product Requirements Document from several hours to about 30 minutes by handling structure and prose while you supply the product judgment. These six prompts cover the PRD lifecycle from problem statement through open questions. Run them in order — the output of each feeds the next.
1. Problem statement
Role: You are a senior product manager at a [B2B SaaS / consumer app / marketplace] company.
Product: [product name — one sentence on what it does].
Target user: [user role and context].
Problem we are solving: [describe in 2-4 sentences — what breaks today, what the user does instead].
Write a 100-word problem statement for a PRD. Structure: current behavior, pain points,
impact, and what a good solution looks like — no solution framing yet.
Flag any assumption that needs validation before we commit.
2. Goals and success metrics
Problem statement: [paste output from prompt 1].
Write the Goals and Success Metrics section of a PRD.
Format: 2-3 goals (user outcomes, not feature outputs), 1-2 measurable KPIs per goal,
one counter-metric per goal to guard against gaming.
Each metric must be observable and verifiable by the engineering team.
3. Full PRD first draft
Role: Senior PM at [company]. Product: [product name].
Audience: [ICP: role, company size, primary use case].
Problem: [2-sentence summary].
Proposed solution: [2-sentence description of the approach].
Write a full PRD. Sections: Executive Summary (100 words), Problem Statement,
Goals & Metrics, User Stories (5-7 placeholder list), Technical Constraints: [list any],
Open Questions (5 items).
Flag any section where you are making an assumption about implementation.
4. Technical constraints section
Product: [product name].
Engineering context: [stack, platform, known limitations].
Regulatory context: [GDPR / SOC2 / HIPAA / none].
Write the Technical Constraints section of a PRD.
Format: numbered list, each constraint as one sentence with a rationale.
Flag constraints that may change during scoping.
5. Open questions list
PRD draft: [paste full PRD or key sections].
Generate 8-12 open questions for this PRD that a skeptical engineering lead or designer
would raise in review. Group by category: scope, feasibility, measurement, edge cases.
For each question: note who owns the answer (PM / Eng / Design / Legal).
6. Executive summary for leadership
PRD: [paste full PRD].
Audience: [C-suite / VP of Product / non-technical stakeholders].
Rewrite the executive summary for this audience. Max 150 words.
Lead with business impact. Omit technical detail. Include one metric that defines success.
Tone: direct, no jargon.
The discipline that makes these six prompts work is running them in order. Founders and PMs who jump straight to prompt 3 often find the goals section vague because the problem statement was never sharpened first. The problem statement is not a ceremony — it is the constraint that keeps every downstream section honest.
How do I generate user stories and acceptance criteria with AI?
User stories are only useful if the acceptance criteria are specific enough to be testable. Most AI-generated user stories stop at "As a [user], I want [feature], so that [benefit]" — a format, not a complete artifact. These six prompts take you from user need to criteria a QA engineer can check off.
7. User story set from a feature description
Feature: [1-2 sentence description of the feature].
Users: [list up to 3 user types who will use this feature].
Write 5-8 user stories in the format:
"As a [user type], I want [specific action], so that [specific outcome]."
Each story must be independently testable and independently shippable.
No compound stories (avoid "and"). Flag any story that assumes another feature exists.
8. Acceptance criteria in Given-When-Then format
User story: [paste one story from prompt 7].
Write 5-8 acceptance criteria in Given-When-Then format.
Each criterion: one behavior, one clear pass condition, one fail condition.
Include: happy path, at least one error state, one edge case.
Flag any criterion that requires data the engineering team does not currently have.
9. Edge case generator
Feature: [feature description].
Primary flow: [describe the happy path in 2-3 sentences].
Generate 10 edge cases a QA engineer or beta user might encounter.
For each: describe the scenario, the expected behavior, and the risk level (low / medium / high).
10. User story refinement
Original user story: [paste].
Acceptance criteria: [paste].
Identify: any ambiguity in the story, any criterion that is not independently testable,
and any implicit assumption about the current system state.
Rewrite the story and criteria to resolve these issues.
11. Epic-to-story decomposition
Epic: [title and 1-paragraph description].
Engineering capacity this sprint: [X story points or N engineers for N weeks].
Decompose this epic into user stories that each fit within [2 / 3 / 5] story points.
Output: numbered list with estimated point values and dependencies flagged.
12. Localization and accessibility criteria
Feature: [feature description].
Markets: [list of target markets and languages].
Accessibility standard: [WCAG 2.1 AA / Section 508 / none].
Write the localization and accessibility acceptance criteria for this feature.
Format: numbered list. Flag any criterion requiring specialist review before sign-off.
The single most common mistake in AI-generated user stories is accepting the first output without running prompt 10. The refinement pass is where you catch stories that assume another feature exists, criteria that are not actually testable, and ambiguities that will produce a four-way disagreement in sprint planning.
How do I use AI for roadmap prioritization with RICE and ICE scoring?
Prioritization debates without a shared scoring framework consume sprint planning hours. These six prompts produce scored tables your team can review, debate, and update — rather than starting from opinions every cycle.
13. RICE score table
Product context: [product, audience, stage — early / growth / scale].
Feature backlog: [list 6-10 feature ideas, one line each].
Score each on RICE: Reach (users affected per quarter), Impact (0.25 / 0.5 / 1 / 2 / 3 scale),
Confidence (percentage), Effort (person-weeks).
RICE score = (R × I × C) / E.
Output: markdown table sorted by RICE score descending.
Flag any score where Confidence is below 50%.
14. ICE score table
Feature backlog: [list 6-10 features].
Score each on ICE: Impact (1-10), Confidence (1-10), Ease (1-10).
ICE score = I × C × E.
Output: markdown table sorted descending.
Note at the bottom: ICE is a quick triage model — flag the top 3 for deeper RICE scoring.
15. Prioritization rationale memo
Top 5 features by RICE score: [paste table from prompt 13].
Company goal this quarter: [one sentence].
Write a 200-word prioritization rationale memo for the engineering team.
Explain why features 1-2 are in-sprint, why 3-4 are next quarter,
and why feature 5 was scored but deprioritized. No fluffy language.
16. Stakeholder objection prep
Prioritized roadmap: [paste top 3 items].
Stakeholders who may push back: [Sales / Marketing / Support / Executive].
For each stakeholder group: generate their 2 most likely objections to this prioritization.
For each objection: draft a 50-word data-backed response a PM can use in a review meeting.
17. Opportunity sizing
Feature idea: [feature description].
User base: [total addressable users, active users, relevant segment].
Estimate: the percentage of users affected per month,
the likely lift in [retention / activation / conversion / revenue],
and the engineering effort in person-weeks for an MVP.
Flag every estimate that is an assumption vs. a known data point.
18. Trade-off analysis
Option A: [feature or approach description].
Option B: [alternative].
Evaluate on 5 dimensions: user impact, technical risk, time to ship,
strategic fit, reversibility.
Output: comparison table. Add a recommendation row with a one-line rationale.
How do I run competitive analysis with AI prompts?
Competitive analysis done manually produces inconsistent output because the format changes every cycle. These six prompts produce a consistent, comparable matrix every time — so you can diff last quarter's analysis against this quarter's with a single pass.
19. Competitor feature matrix
Our product: [product name and primary use case].
Competitors: [list 3-5 names, one sentence each on what they do].
Dimensions: [list 5-7 — e.g., onboarding, API access, mobile support, pricing model, integrations].
Build a comparison matrix. Rows = competitors + us. Columns = dimensions.
Cells: Yes / No / Partial plus a one-line note where relevant.
Flag any cell where you are uncertain and would need a trial to verify.
20. Positioning gap finder
Competitor matrix: [paste output from prompt 19].
Identify 3-5 gaps where our product leads or could credibly lead.
For each gap: name the dimension, describe the gap in one sentence,
and suggest a positioning claim we could make if we shipped or emphasized X.
Flag any claim that is aspirational rather than currently true.
21. Competitor messaging teardown
Competitor: [name].
Their homepage headline: [paste].
Their top 3 stated features: [paste].
Analyze: the primary positioning claim, the audience they are signaling to,
what they avoid saying, and the top objection their messaging leaves unanswered.
22. Win/loss analysis
Deal outcome: [won / lost]. Competitor involved: [name].
Notes from sales call or customer interview: [paste].
Extract: 3 deciding factors, the product gap (if any) that influenced the outcome,
and one action the PM team could take to improve future win rate against this competitor.
Output: table with columns — Factor | Insight | PM Action.
23. SWOT for a new market entry
Our product: [product name].
Target market: [segment, geography, or vertical].
Generate a SWOT analysis for entering this market.
Format: four-quadrant table, 3 bullets per cell.
Flag any strength that is also a dependency on a third party.
24. Analyst report extraction
Source text: [paste up to 500 words from an industry report or analyst post].
Extract: 3 findings relevant to a PM building in [your space],
the assumption behind each finding, and one roadmap implication.
Do not fabricate data from outside this source. Flag any extrapolation you make.
How do I synthesize customer interview notes with AI?
The insight is in the transcript. These six prompts turn raw interview notes into structured findings a team can act on — and flag the assumptions the data does not yet address.
25. Single interview synthesis
Interview transcript or notes: [paste — raw is fine].
Extract:
1. Top 3 pain points with a direct quote for each.
2. Top 3 desired outcomes (what they want to achieve, not what feature they requested).
3. Their current workaround or alternative.
4. One insight that contradicts a product assumption.
Output: structured table.
26. Cross-interview theme finder
Interview summaries: [paste 3-5 summaries, one per customer].
Identify the 5 most common themes across all interviews.
For each: frequency (how many mentioned it), a representative quote,
and the product implication.
Sort by frequency descending.
27. Jobs-to-be-done extraction
Interview notes: [paste].
Using the Jobs-to-be-Done framework, identify:
1. The functional job (what they are trying to accomplish).
2. The emotional job (how they want to feel).
3. The social job (how they want to be perceived).
For each: one JTBD statement in the format "When [situation], I want to [motivation],
so I can [expected outcome]."
28. Assumption validation mapper
Current product assumptions: [list 4-6 assumptions your team holds].
Interview findings: [paste summary from prompt 26].
For each assumption: rate validation status (Confirmed / Contradicted / Inconclusive)
and cite the specific evidence.
Flag any assumption that remains unvalidated after this interview set.
29. Feature request to underlying need
Feature requests from interviews: [paste list].
For each request: reframe it as an underlying user need or job-to-be-done.
Suggest 2-3 ways to address that need — one of which should NOT be
the literal feature the user requested.
30. Insight readout outline
Interview findings: [paste theme table from prompt 26].
Audience: product team, engineering lead, VP of Product.
Outline a 6-slide readout: (1) context and method, (2) top 3 themes,
(3) user quote highlight reel, (4) jobs-to-be-done, (5) assumption validation status,
(6) recommended next steps.
Each slide: headline (8 words max), 3 supporting bullets, who presents it.
What prompt structure produces the best PM artifacts?
The four-layer PM prompt structure that consistently produces usable artifacts is: Role, Context, Task, Format. Skip any layer and the model fills it with a statistical average — which is almost always wrong for your specific product, stage, and audience.
| Layer | What to include | PM example |
|---|---|---|
| Role | Expertise lens | "You are a senior PM at a B2B SaaS company, 10 years experience." |
| Context | Product, audience, stage, constraints | "Product: [name], early-stage, mobile-first, enterprise sales motion." |
| Task | Specific artifact and quantity | "Write the Goals and Success Metrics section. 2-3 goals." |
| Format | Output structure and shape | "Markdown table. Each metric: observable, testable, with a counter-metric." |
Two habits make every prompt in this list compound over time.
- Fill every [variable] before running. A prompt that still says
[product name]produces artifacts about a product that does not exist. The variables are the signal; the template is the container. - Save each prompt after a strong output. The PRD prompt that produced a first draft you sent with minimal edits is worth saving with your product context pre-filled. Next sprint, you update two fields instead of rebuilding from scratch.
For teams, a shared ICP card adds a third habit: pin a 100-word description of your primary user at the top of every prompt — role, company size, the problem they have, how they measure success. Every downstream artifact becomes more internally consistent because the audience context is already embedded.
How Prompt Architects fits this workflow
All 30 prompts above work in any AI tool you already use — ChatGPT, Claude, Gemini. Prompt Architects adds the infrastructure that makes them reusable: a prompt library where you save templates with [variables] intact, Global Variables that inject your product context automatically, and a Chrome extension that puts your saved PM prompts one click away inside whichever AI tool you have open.
Half of our 2,170 customers had no prompt management system before signing up — not Notion, not docs, nothing (our customer data, July 2026). For product managers who run the same PRD structure every sprint, the library is the first-order value: the prompts you build this week should not be rebuilt next sprint from memory.
For teams, Teams lets your whole squad share the same prompt library — the user story template the PM lead wrote lives alongside the interview synthesis prompt the junior PM refined. The prompt playbook becomes a team asset, not something buried in someone's browser history.
"The prompt library is genius — I save structured prompts by category and reuse them. Clean UI, no bloat. Just does the thing." — info.webefo, Verified AppSumo review
Prompt Architects is free to start, no credit card required. For the full workflow that chains these prompts into a repeatable PM system, see Contexts for PMs in every AI chat and the stakeholder update and release notes templates.
Pick the five prompts that match your current sprint bottleneck, save them with your product variables filled in, and run them this week. The time you recover on the first PRD draft or user story set pays for the setup inside a single sprint.