The short answer
The most useful AI prompts for product managers give the model your real evidence (interview notes, tickets, metrics), one clear task and a fixed output format, and ask it to flag assumptions. This page has 14 such prompts for discovery, PRDs, prioritisation, stakeholder updates and launches. They work in ChatGPT, Claude and Gemini; Claude Sonnet 5.5 is free on Claude, and Claude Opus 5.5 needs Claude Pro or above.
| Prompts on this page | 14, in five groups |
|---|---|
| Best tool for long specs | Claude (Sonnet 5.5 on Free, Opus 5.5 on Pro and above) |
| Best tool for feedback spreadsheets | ChatGPT (GPT-6 Luna on Free, GPT-6 Sol on Plus and above) |
| Best tool for research documents | Gemini Notebook, free Standard plan with 50 sources per notebook |
| Keep human | What to build, what to cut, what to promise |
How these product manager prompts are built
Most AI output for product work is generic because the prompt is generic. A product manager knows the customer, the constraints and the politics, and the model knows none of it. Every prompt on this page follows the same five-part pattern so the model starts from your reality instead of an average of the internet:
- Role and stakes. Tell it who it is helping and what a good result changes ("a B2B SaaS product manager deciding what goes into Q1").
- Source material. Paste or attach the real thing: interview notes, support tickets, the current spec, last quarter's metrics. The model is only as specific as what you give it.
- The task, in one verb. Cluster, draft, challenge, rank, rewrite. One job per prompt gives sharper answers than a list of five.
- Constraints and format. Word limits, a table with named columns, "no more than three options". This is what makes the output usable without editing.
- Honesty rules. Ask it to mark assumptions, quote evidence and say "not in the sources" when something is missing. This is the single best defence against confident invention, which we explain in what is hallucination.
Replace everything in [square brackets] before you send. If you only have time to fill in one placeholder properly, make it the source material.
Prompts for customer discovery and research
Discovery is where AI saves product managers the most hours, because synthesis is slow and the model reads fast. The three prompts below turn raw interview notes and tickets into themes you can defend, a sharper interview guide, and a competitor teardown that separates claims from evidence. The Cluster interview notes into themes prompt asks for direct quotes under every theme, which lets you check each one in seconds instead of trusting a neat-sounding label.
For a pile of research documents, a source-grounded tool works better than a chat assistant. Gemini Notebook (the tool Google renamed from NotebookLM on 16 July 2026) answers only from the sources you add and cites them, and its free Standard plan allows 100 notebooks with 50 sources each. See our Gemini Notebook guide and best AI for research.
Prompts for writing specs and PRDs
A first draft of a product requirements document is a good AI job; the decisions inside it are not. The Draft a one-page PRD prompt forces the model to separate what you told it from what it assumed, and to leave open questions open. The Write user stories with acceptance criteria prompt asks for edge cases and failure states, which are the parts busy teams skip. The Stress-test a spec prompt flips the model into a sceptical engineer and designer, which is a cheap way to find gaps before the review meeting does.
Claude is our first choice for long specs because it holds a long document well and follows formatting rules closely. Claude Sonnet 5.5 is on every plan including Free, and Claude Opus 5.5 needs Pro or above. Read the Claude guide for the plan details.
Prompts for prioritisation and roadmaps
AI cannot know your strategy, so never ask it "what should we build?". Ask it to apply a framework you chose to data you supplied, and to show its working. The Score a backlog with RICE prompt does exactly that and flags every estimate it had to guess. The Turn strategy into a now, next, later roadmap prompt keeps dates out on purpose, and the Argue against my top priority prompt gives you the strongest counter-case before your leadership team does.
Prompts for stakeholder communication
Product managers write all day: updates for executives, explanations for sales, apologies for delays. The prompts in this group are about tailoring one set of facts to several audiences. The Write the same update for three audiences prompt is the one most product managers reuse weekly. Save it as a project instruction in ChatGPT or Claude so you only paste the new facts each time. Both tools have projects; Claude Free allows up to 5 Projects, and ChatGPT sets project limits by plan.
Prompts for launches, metrics and retrospectives
The last group covers the end of the cycle: release notes people read, a metric plan before launch, and a blameless retrospective. The Define success metrics before launch prompt asks for a counter-metric for every goal, which is the habit that stops a team from shipping a feature that lifts clicks and hurts retention.
How to adapt these prompts to your product
- Add a standing context block. Write five lines about your product, customers, business model and current goals once, and paste it at the top, or save it in a project. Every prompt improves immediately.
- Change the format, not the logic. If your company writes PRDs in a specific template, paste the headings and say "use exactly these sections".
- Ask for questions first on big jobs. Add "Before you start, ask me up to five questions" to the PRD and roadmap prompts. The answers you give are usually the missing context.
- Iterate in the same chat. "Make the risks section more specific to enterprise buyers" beats starting again.
- Keep the judgement human. What to build, what to cut and what to promise a customer are your calls. Use the output as a draft and a sparring partner.
Which AI tool works best for product managers
For long documents and careful writing, use Claude. For spreadsheets of feedback, charts and quick analysis, ChatGPT is the easiest place to start: GPT-6 Luna is on the free plan and GPT-6 Sol on Plus and above. If your company runs on Google Workspace, Gemini in Docs can draft from your Drive files directly, covered in Gemini prompts for Gmail and Docs. If it runs on Microsoft 365, see Copilot prompts for Outlook. For a full role guide, read AI for product managers and best AI for product managers, or compare the two main assistants in ChatGPT vs Claude.
One rule before you paste anything: customer interviews and roadmaps are confidential. Use your company's work plan where one exists. On a personal ChatGPT account, the training switch is under Settings, Data controls, Improve the model for everyone. On Claude, it is under Settings, Privacy, Help Improve our AI models.
Prompts to copy
Cluster interview notes into themes
You are helping a product manager at [company type] make sense of customer discovery. Below are notes from [number] interviews with [customer segment]. Group the pain points into no more than six themes. For each theme give: a one-line name, how many interviews mention it, two direct quotes, and what the customer is trying to get done. Then list anything mentioned only once that could still matter. Use only the notes; if something is unclear, say so. Notes: [paste notes]
You get: Evidence-backed themes with quotes you can check
Sharpen an interview guide
I am interviewing [customer segment] to understand [problem area]. Here is my draft guide: [paste questions]. Rewrite it into 10 open questions that ask about past behaviour, not opinions or hypotheticals. Remove leading questions, add two follow-up probes for each, and explain in one line why you changed each original question.
You get: An interview guide that gets real stories, not guesses
Compare competitors on evidence
Compare [our product] with [competitor 1], [competitor 2] and [competitor 3] for [use case]. Build a table with rows for target customer, core workflow, pricing model, main strength and main weakness. Use only the material I paste below, mark any cell you could not support as "no evidence", and finish with three gaps we could own. Material: [paste pricing pages, reviews or notes]
You get: A competitor table that separates facts from guesses
Draft a one-page PRD
Draft a one-page PRD for [feature] using these sections: problem, who it is for, goals and non-goals, user journey, requirements (must, should, could), success metrics, risks, open questions. Context: [paste research summary, constraints and deadline]. Keep it under 600 words. Put anything you assumed rather than took from my context in a separate list called Assumptions, and do not resolve the open questions for me.
You get: A structured first draft with assumptions flagged
Write user stories with acceptance criteria
Turn this requirement into user stories for [persona]: [paste requirement]. Write each story as "As a..., I want..., so that...", then add acceptance criteria in Given, When, Then form. Include at least two edge cases and one failure state per story, such as empty data, no permission or a slow connection. Output as a numbered list ready to paste into [Jira or Linear].
You get: Ticket-ready stories that cover the awkward cases
Stress-test a spec
Act as a sceptical senior engineer and then as a senior designer reviewing this spec: [paste spec]. From each point of view list the five most serious gaps, ambiguities or risks, ranked by impact. For each one, quote the line that causes it and suggest the question I should answer before the review meeting.
You get: The hard questions before your review meeting
Score a backlog with RICE
Score these backlog items with RICE (reach, impact, confidence, effort). Our users: [number and segment]. Effort is in person-weeks. Items and what I know about each: [paste list]. Return a table sorted by score with the working for every number. Mark every estimate you had to guess with "guess", and tell me which three guesses would change the ranking most if they were wrong.
You get: A ranked backlog with the shaky estimates exposed
Turn strategy into a now, next, later roadmap
Our goal for [period] is [goal]. Our constraints are [team size, tech limits, commitments]. Here are the candidate initiatives: [paste list]. Arrange them into Now, Next and Later with no dates. For each initiative give the outcome it moves and the main dependency. Then list anything that does not support the goal and suggest dropping it, with one reason each.
You get: A date-free roadmap tied to outcomes
Argue against my top priority
My top priority for next quarter is [initiative] because [reasons]. Make the strongest honest case against it, as a well-informed executive would. Give three counter-arguments, the evidence that would prove each one right, and the cheapest test I could run in two weeks to find out. Do not soften it.
You get: The counter-case and a quick way to test it
Write the same update for three audiences
Here are this week's facts about [project]: [paste progress, metrics, risks, decisions needed]. Write three versions: an executive update in five bullets with the decision I need first, a note for sales and support on what changes for customers and when, and a Slack post for the engineering team that credits the work. Do not add facts that are not in my list.
You get: Three ready-to-send updates from one set of facts
Explain a delay without spin
Draft a message to [stakeholder] explaining that [feature] will ship on [new date] instead of [old date]. The real reasons are [reasons]. Be direct, take responsibility, give the new plan and what we are doing to protect the date, and offer [option] if they need something sooner. Under 150 words, no corporate filler.
You get: An honest delay message that keeps trust
Write release notes people read
Turn these shipped changes into release notes for [audience]: [paste ticket titles or changelog]. Lead with the change that matters most to customers. For each item write a bold one-line benefit and one sentence on how to use it. Group small fixes into one line at the end. Plain English, no internal ticket names.
You get: Customer-facing release notes in minutes
Define success metrics before launch
We are launching [feature] to [segment] to achieve [goal]. Propose one primary metric, two supporting metrics and one counter-metric that would warn us if we are causing harm. For each give the exact definition, the data source we would need, a realistic target range and the date we should review it. Flag any metric we probably cannot measure today.
You get: A measurement plan with a guardrail metric
Run a blameless retrospective
Here are notes from our retrospective on [project]: [paste notes]. Organise them into what went well, what did not, and root causes, without naming or blaming individuals. Then propose no more than three changes we can make next cycle, each with an owner role and a way to tell if it worked.
You get: A clear retro write-up with three actions