Best AI Prompts for PMs

AI summary

Seven AI prompts for product managers that respect what PM work actually is (judgment plus communication, not document production): PRD tightening, roadmap trade-off framing, customer interview synthesis, customer-bug triage, launch readiness audits, exec updates, and cross-functional conflict framing.

Prompt 1

PRD Tightener

Most PRDs read like the PM tried to anticipate every question. The good ones answer the questions the team actually asks. This prompt finds the gap.

Here is the PRD I am about to send out for review:

[PASTE THE PRD]

The audience that will review it: [ENGINEERING / DESIGN / EXEC / CROSS-FUNCTIONAL].

The decision the PRD is asking for: [SHIP / NO-SHIP / SCOPE-CHANGE / etc.]

Review it as the most thoughtful engineering or design lead on the team would:

1. THE PROBLEM: do I state the user problem in plain English before any solution language? If not, where.
2. THE "WHY NOW": is there a clear reason this matters this quarter and not next.
3. UNANSWERED QUESTIONS: list the 3-5 questions the team will ask in the review that the PRD does not yet answer.
4. SCOPE CREEP: anything in the doc that is not strictly required for the stated problem.
5. SUCCESS METRICS: are they measurable, time-bound, and tied to the user outcome (not vanity metrics).
6. WHAT IS NOT IN SCOPE: is this section explicit or implied (it should be explicit).
7. ONE THING TO CUT: the section that contributes least to the decision.

Do not rewrite. Surface the gaps. I will rewrite.

When to use: Before circulating any PRD. · Best model: Claude. The discipline about “do not rewrite” matters.

Prompt 2

Roadmap Trade-Off Framer

You have 8 candidate projects, 3 quarters, and 5 engineers. The roadmap conversation always goes sideways. This prompt structures it before the meeting.

Here are the candidate projects for the next quarter:

[FOR EACH: name, one-sentence description, estimated engineering weeks, primary metric impact, status (ready / needs spec / blocked)]

My team capacity: [ENGINEER-WEEKS AVAILABLE]
Our primary objective this quarter: [ONE-SENTENCE OBJECTIVE]
Known constraints (dependencies, oncall load, hiring): [LIST]

Produce a trade-off framing:

1. THE OBVIOUS YES: any project where impact-per-engineering-week is clearly outsized.
2. THE OBVIOUS NO: any project that does not serve the quarterly objective.
3. THE TRUE TRADE-OFFS: 2-3 projects where the team will reasonably disagree, and the case for and against each.
4. THE DEPENDENCY CHAIN: which projects must happen first to unblock others.
5. THE CAPACITY MATH: what fits, what does not, by how much.
6. THE COVERAGE MAP: which projects need exec air-cover, design partnership, or cross-team sign-off, and whose pre-meeting I should book.
7. ONE QUESTION FOR THE TEAM: the question that should anchor the roadmap conversation.

Do not advocate for specific projects. Surface the trade-offs plainly so the team can make the call.

When to use: Two weeks before quarterly planning. · Best model: Claude. The discipline about not advocating matters in roadmap work.

Prompt 3

Customer Interview Synthesis

You did 8 customer interviews and have 90 pages of notes. The temptation is to write a deck that confirms what you already believed. This prompt does the synthesis without bias.

Here are the notes from my customer interviews:

[PASTE OR SUMMARIZE NOTES FROM EACH INTERVIEW: who, role, key quotes, observed behaviors, surprises]

The hypothesis I started with:

[ONE SENTENCE]

The decision this research will inform: [SPECIFIC DECISION]

Produce a synthesis with:

1. WHAT THE INTERVIEWS CONFIRMED: hypotheses that the data supports, with the specific evidence.
2. WHAT THE INTERVIEWS CHALLENGED: where the data pushed back on my starting assumptions.
3. THE NEW UNKNOWN: a question the interviews surfaced that I did not start with.
4. SEGMENT PATTERNS: where the data clusters by user type, role, or stage of life.
5. THE 3 QUOTES THAT MATTER MOST: with the speaker role and the context.
6. THE RECOMMENDATION: what the research suggests for the upcoming decision, with the caveat about sample size.
7. THE RESEARCH I STILL NEED: what would resolve the open questions.

Do not let me confirm my hypothesis if the data does not actually support it. Be direct.

When to use: Within a week of finishing the interview round. · Best model: Claude. Most disciplined about not bending data to fit a starting hypothesis.

Prompt 4

Spec from Customer Bug

A single customer complaint can become a roadmap line item or a noisy distraction. This prompt distinguishes them.

Here is a customer bug or feature request I received:

[PASTE THE REPORT OR DESCRIBE THE CONVERSATION]

Customer context: [WHO THEY ARE, HOW BIG, WHAT THEY USE OUR PRODUCT FOR]
My gut reaction: [ROUGHLY: "this seems important" / "this seems edge-case" / "I am not sure"]
What I know about whether other customers have raised this: [DATA OR "I HAVE NOT CHECKED"]

Produce a quick triage:

1. THE UNDERLYING NEED: what the customer actually wants, in their language, separated from the specific solution they proposed.
2. WHO ELSE LIKELY HAS THIS PROBLEM: rough segment estimate based on the customer profile.
3. HOW BIG: small / medium / large signal, with the reasoning.
4. WHAT TO DO NOW: one of (one-off response, add to backlog with priority, escalate to current sprint, kill).
5. WHAT TO ASK BEFORE COMMITTING TO A FIX: questions that would let me size this more confidently.
6. THE RESPONSE TO THE CUSTOMER: a 4-sentence reply that respects their feedback without committing to a solution prematurely.

Do not promise the customer a feature on my behalf.

When to use: Within a day of receiving the feedback. · Best model: Claude. The discipline about not over-committing matters.

Prompt 5

Launch Readiness Audit

You are 10 days from launch. Nothing feels broken. This prompt finds what is not yet ready.

We are launching [FEATURE NAME] on [DATE].

What is done:

[LIST: engineering, design, copy, dataset, support training, etc.]

What is in progress:

[LIST]

What is unstarted:

[LIST]

Key dependencies (cross-team or external): [LIST]

Audit the launch readiness:

1. THE BLOCKERS: items that are not done AND must be done before launch. Sort by deadline risk.
2. THE FAST FOLLOWERS: items that can ship within 2 weeks post-launch without hurting the launch quality.
3. THE NICE-TO-HAVES: items the team will want to do but should not delay launch.
4. THE GO-NO-GO QUESTION: the one question that, if answered the wrong way, should delay the launch.
5. THE COMMS GAP: who has not yet been told about this launch who needs to know.
6. THE POST-LAUNCH MEASUREMENT PLAN: what we are measuring in the first week and how.

Do not under-flag risk to keep the launch on track. If something looks shaky, name it.

When to use: 10 business days before launch. · Best model: Claude. The discipline about flagging risk matters more than speed here.

Prompt 6

Exec Update Drafter

Every week or month an exec wants an update on your area. The temptation is to lead with what you have shipped. This prompt structures the update around what the exec actually cares about.

I owe an update to [EXEC NAME / TITLE] for my area: [AREA].

What happened in the last period:

[BULLET POINTS: shipped, in flight, blocked, learned]

The exec's stated priorities for the year: [LIST]
The number(s) they look at first: [METRIC OR DASHBOARD]
What happened with those numbers this period: [BRIEF]

Draft a 250-word update with:

1. THE HEADLINE: one sentence that names the most important thing that happened against their priorities.
2. THE NUMBER: the metric they care about, with the comparison that makes it meaningful (week-over-week, vs target, etc.).
3. THE ONE-PARAGRAPH "WHY": what is driving the number, with the specific evidence.
4. WHAT WE SHIPPED that matters for their priorities (not everything we shipped).
5. THE RISK I AM TRACKING that they should know about.
6. WHAT I NEED FROM THEM (if anything).

Do not bury bad news. Do not over-celebrate small wins. Tone: same one I would use in a one-on-one with this exec.

When to use: End of each reporting period (weekly, monthly, or per the exec’s cadence). · Best model: Claude. The discipline about exec-tone calibration is what makes the update land.

Prompt 7

Cross-Functional Conflict Framer

Engineering wants to refactor. Design wants to redesign. Sales wants to ship the demo feature. This prompt structures the conflict so a decision can actually happen.

Here is the cross-functional conflict I need to resolve:

FUNCTION A wants: [POSITION]
FUNCTION B wants: [POSITION]
FUNCTION C wants: [POSITION (if applicable)]

The stakes: [WHAT IS BLOCKED OR AT RISK]

My current read on each function's underlying interest (not just their stated position):

[ONE SENTENCE PER FUNCTION]

Produce a framing:

1. THE STATED POSITIONS, restated cleanly.
2. THE UNDERLYING INTERESTS, separated from the positions.
3. THE SHARED INTEREST: what every function actually wants (almost always exists).
4. THE TRADE-OFFS that are real and the trade-offs that are perceived but not real.
5. THREE POSSIBLE PATHS: each one with what each function gets and gives up.
6. THE RECOMMENDATION: which path serves the shared interest best.
7. THE PRE-MEETING I should book before bringing this to the broader group, and with whom.

Do not take sides without evidence. If one function's position is clearly weaker, name it; if all three have merit, say so.

When to use: Within 48 hours of the conflict becoming visible. · Best model: Claude. Tone discipline is critical for cross-functional work.

Connectors are now standard

Claude, ChatGPT, and Grok all support connectors that let your AI read live data from your work tools (Gmail, Notion, GitHub, Asana, HubSpot, Stripe, and many more) instead of relying on you to paste context. For PMs this means the AI can read your Notion or Confluence PRD drafts, your Linear or Jira ticket history, your customer interview transcripts in Descript or Otter, your Slack threads for the cross-functional conflict prompt, or your Mixpanel/Amplitude analytics for the launch-readiness audit.

For product managers, the connectors worth pairing with these prompts:

Get Smarter About AI Every Morning

Free daily newsletter. Built for people who want to use AI well, not chase every model.

Free forever. Unsubscribe anytime.

Two ways to go further