What you'll learn: A scannable cheat sheet of 25 AI prompts organized by PM workflow. One tight, copy-paste prompt for each recurring product management job.
You don't need 100 prompts. You need one good prompt for each job you do every week.
This is the cheat sheet I wish I'd had: 25 prompts grouped by the five workflows that fill a PM's calendar. Discovery, strategy and planning, specs and docs, analysis and metrics, communication. Each one is short enough to skim, specific enough to produce real output, and built to refuse generic answers.
Bookmark this page. When you're staring at a blank doc on a Tuesday afternoon, scroll to the right group, grab the prompt, fill the brackets, and go. The prompts work in Claude or ChatGPT. The skills that make them effortless, with no setup and no re-explaining your product, run in Claude Code. Prompts are the fast start; skills are where it gets automatic.
The Cheat Sheet at a Glance
Five workflows, five prompts each. Here's what each group covers.
| Workflow | Prompts cover | When you reach for it |
|---|---|---|
| Discovery | Interview synthesis, problem validation, JTBD, feedback patterns, survey questions | You have raw signal and need themes |
| Strategy & Planning | OKRs, positioning, prioritization, roadmap framing, bet sizing | You're deciding what to do and why |
| Specs & Docs | PRDs, user stories, edge cases, release notes, acceptance criteria | You're handing work to engineering |
| Analysis & Metrics | Funnels, metrics trees, experiments, churn, RICE | You have numbers and need a decision |
| Communication | Status updates, exec summaries, alignment, announcements, board narrative | You're getting stakeholders on the same page |
The pattern that makes every prompt below better: tell the model what bad output looks like. "Be specific, no generic answers" does more work than any clever phrasing.
A prompt is only as good as the context behind it. Give the model your product, your users, and one clear example of what "too generic" looks like. Then it stops guessing.
Discovery Prompts (1-5)
1. Interview Synthesis
Here are notes from [N] interviews about [topic]:
[paste notes]
Pull out the top 3-5 themes with a direct quote for each, plus any
signal that contradicts what we assumed about [persona].
Be specific. Not "users want better onboarding" — give me
"3 of 5 abandoned setup at the integrations step."
2. Problem Validation
I think [persona] has this problem: [problem]. My evidence:
[paste tickets, notes, or survey data]
Tell me how strong this evidence really is, what could explain it
besides a real problem, and whether I'd bet on building for it.
Be blunt. "Your evidence is thin" beats "consider more research."
3. Jobs-to-Be-Done
From these interview notes, extract jobs-to-be-done in the format:
"When [situation], I want to [motivation], so I can [expected outcome]."
Notes: [paste]
For each, label functional/emotional/social, rate confidence
(observed vs. inferred), and note where today's solution falls short.
4. Feedback Pattern Analysis
Here is [N] pieces of feedback from [source]:
[paste]
Group by recurring theme and frequency, separate bugs from requests
from workflow complaints, and name the one pattern worth acting on
this sprint.
Skip the obvious. Surface what I'd miss reading these one by one.
5. Survey Question Generator
I'm surveying [audience] to decide [decision]. Generate 8-10 questions.
Mix quantitative and open-ended, avoid leading phrasing, and include
at least two that could disprove my assumption. End with a question
that flags good interview candidates.
No throwaway demographic filler unless it changes the decision.
Related — Want the deeper versions of these? Claude Prompts for Product Managers is the full 30-prompt playbook with richer discovery, strategy, and analysis prompts.
Strategy & Planning Prompts (6-10)
6. OKR Drafting
Draft OKRs for [team] for [quarter].
Company priorities: [list]
Our focus: [list]
Constraints: [team size, dependencies]
Give me 2-3 Objectives with 3-4 Key Results each. Every KR is a
number, set at roughly 70% confidence, and a leading indicator
where possible. Flag any two objectives that fight each other.
7. Competitive Positioning
My product: [one line]. Target user: [persona].
Main competitor: [name + one line].
Tell me where I win and where they win across value prop, depth vs.
breadth, pricing, and switching cost.
Be opinionated. "Both do project management" is useless. I want
"They win on enterprise permissions; you win on time-to-value."
8. Prioritization Gut Check
Here are the features competing for next quarter:
[list with one-line descriptions]
Rank them by impact-to-effort and tell me which one I'm probably
overrating and which I'm underrating, with reasoning.
Name your assumptions so I know what to validate before committing.
9. Roadmap Framing
Help me frame a [quarter/half] roadmap for [product].
Business goal: [outcome]
Themes I'm considering: [list]
Known constraints: [list]
Organize the themes into now / next / later, tie each one to the
business goal, and call out anything that looks like activity
without a clear outcome behind it.
10. Bet Sizing
I'm deciding whether [bet] is worth it for [audience]. What I know:
[paste signals or data]
Rate the value, usability, feasibility, and business risk as
HIGH / MEDIUM / LOW with specific reasoning for each.
Tell me the single riskiest assumption and the cheapest way to test it.
Specs & Docs Prompts (11-15)
11. PRD First Draft
Write a PRD for [feature].
Problem: [user problem] | User: [persona]
Success: [measurable outcome] | Constraints: [list]
Include: problem statement, user stories with acceptance criteria,
v1 scope (and what's explicitly out), success metrics, open questions.
Leave out implementation details that don't change the experience.
This is the PM-to-eng handoff.
12. User Story Expansion
Break this story into 4-6 smaller ones:
"As a [role], I want to [action] so that [outcome]."
For each sub-story: same format, 3-5 acceptance criteria in
Given/When/Then, flagged dependencies, and a Small/Medium/Large size.
No filler stories. Each one should be independently shippable.
13. Edge Case Generator
I'm speccing [feature]. Happy path: [describe it].
List 12-15 edge cases grouped by input (empty, max, special chars),
state (concurrent, offline, mid-flow), permissions, and integrations.
For each, one line on the expected behavior. Skip the trivial ones.
14. Acceptance Criteria
Here's a feature description: [paste].
Write acceptance criteria in Given/When/Then that engineering and QA
could test against without asking me follow-up questions.
Cover the success path and the obvious failure paths. Be concrete —
real values, not "a valid input."
15. Release Notes
We shipped these changes:
[paste changelog or bullets]
Audience: [customers / internal]. Tone: [pick one].
Write release notes that lead with the user benefit, group by impact
(major / minor / fixes), and skip internal refactors users won't feel.
One sentence per change, not a paragraph.
Related — On ChatGPT instead of Claude? ChatGPT Prompts for Product Managers covers the same workflows with prompts tuned for how ChatGPT handles context and formatting.
Analysis & Metrics Prompts (16-20)
16. Funnel Analysis
Here's our funnel: [paste stages with numbers].
Tell me the biggest absolute drop-off and the biggest proportional
one, the stage with the most revenue leverage, and three hypotheses
for the worst step — with what I'd measure to test each.
Lead with the one stage I should fix first.
17. Metrics Tree
Build a metrics hierarchy for [feature/area].
North Star: [metric] | Business goal: [goal]
Give me: the North Star, 3-5 input metrics that drive it, 3-5 health
guardrails, and 2-3 feature-specific metrics. For each: definition,
target, and data source. Flag anything needing new instrumentation.
18. Experiment Design
I want to test [hypothesis].
Current state: [now] | Change: [proposed] | Primary metric: [metric]
Traffic: [users on this surface]
Give me null/alternative hypotheses, rough sample size and runtime,
guardrail metrics, and the rule for ship vs. iterate vs. kill.
Tell me if this surface has too little traffic to ever reach significance.
19. Churn Analysis
We're at [churn rate] monthly churn. What I have:
[paste cohort data, cancel reasons, usage]
Segment the churn the most useful way, name the likeliest drivers
from the data, and rank three retention experiments by expected impact.
For each experiment, the one metric that proves it worked.
20. RICE Scoring
Score these with RICE:
[list 5-10 features]
Reach (users/quarter), Impact (3 to 0.25), Confidence (% ), Effort
(person-months). Show your reasoning per score and flag assumptions
to validate. Rank by score, then add a gut-check column for whatever
the math gets wrong.
Communication Prompts (21-25)
21. Weekly Status Update
This week on [project]:
[paste raw notes]
Write a stakeholder update: what shipped (with user impact), what's
in progress (with ETA), what's blocked (with the unblock), and any
decision needed.
Under 250 words. No "the team made great progress" filler.
22. Executive Summary
Summarize [doc] for [CEO / board / VP Eng]:
[paste or summarize source]
200-300 words. Open with the "so what" for this audience, give the
3-5 key findings, end with the ask. Use their language — revenue
for the CEO, capacity for eng, market for the board.
23. Stakeholder Alignment Email
I need to align [stakeholders] on [decision].
Context: [background] | My recommendation: [what] | Trade-offs: [list]
Write an email that states the decision in the first line, names each
stakeholder's likely concern, backs the recommendation with evidence,
and asks for feedback by [date]. Under 400 words.
24. Feature Announcement
We shipped [feature]. Write an internal announcement for [Slack/email].
What it does: [user-facing] | Why: [reason] | Who it affects: [segment]
How to use it: [link] | Known gaps: [list] | Credit: [team]
Lead with the benefit, keep it [Slack-length / email-length], and
make the call-to-action obvious.
25. Board Deck Narrative
Write the product section of our board deck.
Quarter: [Q] | Wins: [list] | Misses: [list]
Metrics: [actuals vs. targets] | Next quarter: [priorities]
400-600 words. Open with strategic context, celebrate wins without
inflating them, address misses with root cause and fix, and connect
the metrics to the story instead of just listing them.
Related — These five workflows are the full PM job. How to Use Claude as a Product Manager maps each one to the skills that run it without prompting.
From Cheat Sheet to System
A cheat sheet saves you the blank page. It doesn't save you the setup.
Every prompt above still asks you to re-explain who you are, what you're building, and who you're building it for. You paste the product context, specify the format, and fix the generic output every single time. That tax adds up.
The next step removes it. When your product, users, and competitors live in context files, the model already knows them. The cheat-sheet prompts become installable skills that read that context automatically, so "write a PRD for this feature" produces a draft grounded in your actual product, not a template you have to rewrite.
That's the difference between a prompt and a skill: a prompt is something you paste; a skill is something that runs, with your context already loaded. The 25 prompts here are the on-ramp. Using Claude as a product manager is where the on-ramp leads.
Browse the free PM skills → Every prompt in this cheat sheet has a skill version that reads your context and runs without setup. See the skills directory →
FAQ
What's the best AI prompt for product managers?
The best prompt is the one that refuses generic output. Across every workflow, the highest-leverage move is telling the model what bad looks like: "not 'users want better onboarding,' give me '3 of 5 dropped at the integrations step.'" That single instruction does more than any clever phrasing. Start with the PRD First Draft prompt if you want one that pays off immediately.
Do these work in ChatGPT and Claude?
Yes. Every prompt here is plain text and works in any AI chat tool. Output quality tracks how much context you give it, not which tool you use. The difference shows up at scale: in Claude Code with context files, the product context is always loaded, so you skip the re-explaining step entirely. If ChatGPT is your daily driver, the ChatGPT prompts for PMs are tuned for it.
How do I get more consistent output?
Two things. First, give the model context (your product, your users, your competitors) so it stops guessing. Second, give it a format and one example of what "too generic" looks like. Do both manually with each prompt, or set up context files once so the context is always there. The second path wins as soon as you're running the same prompt more than a few times.
Are there ready-made versions?
Yes. Most of these prompts have a skill version that loads your context and runs without setup: /prd-generator, /okr-coach, /roadmap-builder, /executive-update-generator, and more. Browse the free PM skills → to grab the ones that match your workflow.
About the Author
Ron Yang is the founder of mySecond — he builds and manages PM Operating Systems for product teams. Prior to mySecond, he led product at Aha! and is a product advisor to 25+ companies.