Best AI prompts for product managers

Last verified: 23 October 2026. The prompts below are original to this page and model-agnostic; the prompting advice is checked against the vendors' current documentation.

A good product prompt reads like a brief to a smart colleague on their first day. Anthropic's Claude prompting best practices says the same thing: "Think of Claude as a brilliant but new employee who lacks context on your norms and workflows." The same docs recommend 3 to 5 examples per prompt for best results, wrapped in tags so the model can tell them apart from instructions.

OpenAI's prompt engineering guide notes the caveat every PM should keep in mind: because model output is non-deterministic, "prompting to get your desired output is a mix of art and science." A prompt that worked once is a draft. A prompt that works every week, on new inputs, is a template.

How were these prompts chosen?

Each prompt had to pass four tests. It maps to a task a product manager does at least monthly. It has slots for the context only you have, marked in square brackets. It asks for an output you can check line by line, usually a table or a fixed list. And it tells the model what to do when the input does not support an answer, because a model with no instruction for missing evidence fills the gap with something plausible.

The prompts use the six-part structure from the prompt template for product managers in a shortened form. Paste your material where the brackets are and remove anything confidential first.

Which AI prompts should product managers save?

Prompt	Best for	What you paste in	What you get back
1. Interview synthesis	Discovery	Anonymised interview notes	Themes with quotes and counts
2. Problem framing	Turning requests into problems	A stakeholder request	A problem statement and open questions
3. PRD red team	Spec review	Your PRD draft	Gaps, risks and questions by section
4. Prioritization check	Roadmap trade-offs	A backlog with notes	Scores with the assumption behind each
5. Stakeholder update	Weekly comms	Raw status notes	A five-line update by audience
6. Metric diagnosis	Analytics	A metric change and context	Ranked hypotheses and the data to check
7. Competitor teardown	Positioning	Public pages you collected	A feature and message comparison
8. Interview guide	Discovery	A hypothesis	A behavioral interview script
9. User stories	Delivery	A feature description	Stories with acceptance criteria
10. Survey coding	Feedback	Open-text answers	A coded table with counts
11. Pre-mortem	Launch planning	A launch plan	Failure modes with early signals
12. Release notes	Shipping	Merged change titles	Customer-facing notes by type

Which prompts help most with product discovery?

Discovery prompts work when the model is forced to cite your material for every claim. These three cover the loop from a request to a synthesis.

1. Interview synthesis

Role: You are a product researcher helping me synthesise customer interviews.
Context: Product: [one sentence]. Research question: [what we are trying to learn].
Below are notes from [N] interviews, separated by "###". Names are removed.
Task: Group what participants said into themes.
Format: A table with columns: theme, number of participants who raised it,
two verbatim quotes with participant number, what it suggests we test next.
Rules: Only use what is in the notes. If a theme comes from one participant,
say so. List anything that contradicts the research question separately.
Notes:
[paste notes]

2. Problem framing from a stakeholder request

Context: A stakeholder asked for: "[paste the request]".
What I know about the users and the situation: [paste context].
Task: Separate the request from the problem underneath it.
Format:
1. The problem in one sentence: who has it, what gets in their way, why it matters.
2. The evidence we have, and the evidence we are missing.
3. Three questions I should ask the stakeholder before agreeing to any solution.
Rules: Do not propose a solution. If the request is already a well-defined
problem, say so and explain why.

8. Interview guide from a hypothesis

Context: Hypothesis: [who] struggles with [what] because [why].
What would change my mind: [evidence].
Task: Write a 30-minute discovery interview guide.
Format: a two-line intro, 6 questions about past behaviour (no hypotheticals,
no leading questions), one follow-up probe per question, and a closing question.
Rules: Flag any question I could not ask without revealing the hypothesis.

For the full interview workflow, from recruiting to synthesis, see AI for customer interviews.

Which prompts help with specs and prioritization?

Spec prompts are most useful as critics, not authors. A model that drafts your PRD writes a confident document; a model that attacks your draft finds the gaps you stopped seeing.

3. PRD red team

Role: You are a skeptical senior engineer and a skeptical designer reviewing a PRD.
Task: Review the PRD below before it goes to the team.
Format: For each section, list (a) what is ambiguous, (b) what is missing,
(c) the question each reviewer would ask in the meeting. Then list the
three changes that would most reduce rework.
Rules: Do not rewrite the PRD. Quote the exact sentence you are questioning.
PRD:
[paste PRD]

4. Prioritization check

Context: Our goal this quarter: [outcome and metric]. Team capacity: [rough].
Backlog items with notes:
[paste items]
Task: Score each item on reach, impact, confidence and effort.
Format: A table with the four scores, the total, and one line stating
the assumption behind the impact score.
Rules: Where my notes give no basis for a score, write "unknown" instead
of guessing, and list what data would fill the gap.

9. User stories with acceptance criteria

Context: Feature: [description]. Users: [who]. Constraints: [platform, data, policy].
Task: Split the feature into user stories small enough for one sprint.
Format: "As a [user], I want [goal], so that [reason]", then 3 to 5
acceptance criteria per story written as observable conditions.
Rules: Mark any story that fails the INVEST checklist and say which letter.

The scoring logic behind prompt 4 is covered in AI for prioritization, and AI prompts for writing user stories extends prompt 9 into a seven-step workflow.

Which prompts save the most time for a senior PM or product leader?

Leadership prompts handle communication and risk, where the cost of a vague sentence is a misaligned team.

5. Stakeholder update

Context: Audience: [exec team / engineering / sales]. What they care about: [one line].
Raw notes from this week:
[paste notes]
Task: Write this week's update for that audience.
Format: Five lines: status in one word (on track, at risk, off track),
what shipped, what changed and why, the decision I need from them, next week.
Rules: Under 120 words. No adjectives about effort. If the notes do not say
whether we are on track, ask me instead of guessing.

11. Pre-mortem

Context: We launch [feature] on [date] to [audience]. Plan: [paste plan].
Task: Assume it is three months after launch and the launch failed.
Format: The 8 most likely reasons, each with the earliest signal we would see,
where we would see it, and one action that reduces the risk now.
Rules: Order by likelihood, not severity. Include at least two reasons
outside engineering (pricing, support, sales, legal, timing).

6. Metric diagnosis

Context: [Metric] moved from [value] to [value] between [dates].
Recent changes: [releases, campaigns, pricing, seasonality].
Task: List hypotheses that could explain the change.
Format: A ranked table: hypothesis, why it fits, what would disprove it,
the exact cut of data to check (segment, platform, cohort).
Rules: Include measurement problems (tracking, definitions) as hypotheses.

AI for stakeholder updates covers tailoring prompt 5 to each audience.

Which prompts work for market and customer feedback?

Feedback prompts need a coding scheme you control, or the model invents categories that change every run.

7. Competitor teardown

Context: Our product: [one sentence]. Competitors and the public pages
I collected for each (pricing, homepage, docs):
[paste text]
Task: Compare how each one positions itself and what it offers.
Format: A table: competitor, target customer as they state it, headline
promise, three features they emphasise, pricing model, a source line for each cell.
Rules: Use only the pasted text. Write "not stated" where a page does not say.

10. Survey coding

Context: Survey question: "[question]". [N] open-text answers below, one per line.
Codes to use: [list 5 to 8 codes, plus "other"].
Task: Assign each answer one primary code.
Format: A table of answer number and code, then a count per code,
then up to 5 answers coded "other" that may need a new code.
Rules: Do not merge or rename my codes.

12. Release notes

Context: Merged change titles and linked issue titles for this release:
[paste list]
Task: Write customer-facing release notes.
Format: Group under Added, Changed, Fixed, Removed. One plain sentence per item,
describing what the customer can now do.
Rules: Leave out internal refactors. List any item you could not explain
in customer terms so I can check it.

AI for competitive analysis covers sourcing the pages for prompt 7.

When do AI prompts give product managers bad answers?

Prompts fail in predictable ways. Thin context produces generic output, so a prompt without your product, users and constraints returns advice that fits any company. Missing format rules produce prose you cannot check. And a prompt with no instruction for absent evidence invites the model to fill the gap with a confident guess, which is the failure most likely to reach a stakeholder.

The fix is the same each time: paste the source material, ask for a table, require a quote or data row for each claim, and tell the model to write "unknown". Few-shot prompting helps when the format still drifts; add two or three good outputs from earlier runs as examples.

Where can you learn to write these prompts yourself?

The prompts above are starting points. The best AI prompting resources for product managers collects the reading list behind them, from vendor documentation to product-specific guides, and AI prompts for product ideas applies the same brief-style structure to ideation.

Builders Camp's AI Prompting for Product bootcamp teaches the prompting patterns product people use to speed up discovery, decision-making and execution. It runs one week, with 2 live sessions and 19 self-paced microlessons, and its practical challenge is rated beginner difficulty. See the AI Prompting for Product bootcamp.