Claude is useful for the paper side of product management: turning interview notes into evidence you can point at, finding ambiguity in a PRD before engineering does, auditing a prioritisation score, and writing user stories with testable acceptance criteria. It is weak where the job is human. It has never met your users, does not know your codebase, and will hand you a confident spec that hides every decision you have not made. Work in four passes (evidence, spec, priority, stories), make it quote your own notes, and treat every number it produces as an input to challenge.
Nesta página
- What do you give it before it touches a spec?
- How do you turn interview notes into evidence, not opinions?
- How do you find the holes in a PRD before engineering does?
- How do you stop prioritisation scores from being fiction?
- What does a user story with real acceptance criteria look like?
- What changes when the feature is AI?
- Where does Claude fail at this?
- Which product skill fits which job?
- How do you install a product management skill in Claude?
- All-in-one: every skill, one price
- Related guides
- Frequently asked questions
Prompts for Product Managers
Get Prompts for Product Managers →What do you give it before it touches a spec?
Four inputs change every result more than prompt wording does. The decision the work feeds ("should we build bulk edit this quarter" beats "help with bulk edit"). Raw evidence, not your summary: a summary already contains your conclusion and Claude will agree with it. Constraints you already know, such as team size, deadline, platform limits and things already tried, because it cannot see your architecture. And what a wrong answer costs, since a reversible UI tweak and a pricing change need different caution. For a wider catalogue of prompts by task, see our roundup of Claude prompts for product managers; this guide runs four workflows in order.
How do you turn interview notes into evidence, not opinions?
The classic discovery mistake is remembering the quotes that agree with you. Ask a model for "the main themes" and it returns fluent themes whether or not your notes support them. Force a split between what people did and said and what you infer, with a quote on every claim. (Synthesis is only as good as the interviews: ask about specific past behaviour, not "would you use this", and keep your solution for the end.)
Check two or three quotes against your notes word for word, since tidied or fabricated quotes are the model's most dangerous habit here. Section 6 will often say a handful of interviews is not statistics. Take that as the point: this is direction, not proof.
How do you find the holes in a PRD before engineering does?
Drafting the first PRD is the least valuable thing Claude does; reviewing one is the most. Give it the role of a sceptical senior engineer and ask for questions, not a rewrite.
Section 5 earns its keep: "fast" in a requirement is a promise nobody can verify, and models love to supply "under 200 ms" with no basis. Send the top five questions to the people who can answer them and you have likely saved a planning session.
How do you stop prioritisation scores from being fiction?
RICE (Reach, Impact, Confidence, Effort) looks rigorous because it outputs numbers, but asked to score items a model supplies plausible inputs with no basis. Use it as a critic of your scores instead.
The sensitivity re-rank is the persuasive part: if your top item wins only when the guesses go your way, gather one more data point first. The pack above contains a RICE prompt that likewise asks which inputs are guesses.
What does a user story with real acceptance criteria look like?
"As an admin I want to export data" passes review and fails testing, because nothing in it can be tested. Good criteria describe observable behaviour, include unhappy paths, and are specific enough for a tester who skipped the meeting. Given/When/Then (the Gherkin convention many teams use) forces that.
An illustration of the format (invented feature):
The bracketed AC4 is what you want: a visible hole for a human, not a number quietly made up.
What changes when the feature is AI?
The workflows hold, but model output is probabilistic: the question becomes how often it is wrong and what a wrong answer costs the user. Put three things in the spec before building: today's baseline for the metric you want to move, the quality bar to clear before shipping and how you will measure it, and whether you need a custom model or an existing service with good interface design. Irreversible actions deserve a human confirmation step. A one-line executive request is a scoping job first.
AI & Data Product Manager AI Skill - Cassius
Get AI & Data Product Manager AI Skill - Cassius →Where does Claude fail at this?
It invents the user. Ask for "key user needs" without evidence and you get a list from general knowledge. Every need should trace to a participant or ticket; if it cannot, strike it.
It agrees with your hypothesis. State a favoured solution and it finds reasons it works. Give a kill condition up front and ask it to argue the opposite.
It treats a small sample as proof. Eight interviews find problems and poorly size them. Make it state the sample size next to each claim.
It writes specs that look finished. Fluent prose hides undecided questions. A PRD with no open questions is a bad sign.
It does not know your system. It cannot tell that a change touches a legacy module or that two engineers are available, so effort estimates are guesses.
It tests the happy path. Require validation, permission and empty-state criteria explicitly.
It cannot manage stakeholders. Who objects, and what was promised last review, is not in the chat.
Which product skill fits which job?
| Skill | Best for | Price |
|---|---|---|
| Prompts for Product Managers | The weekly cycle: discovery, PRD, stories, prioritisation, roadmap, launch (24 prompts) | $9 |
| AI & Data Product Manager AI Skill - Cassius | Scoping an AI or data initiative: brief, metrics, build-versus-buy | $19 |
| All-in-one | Both, plus every other skill in the catalog | $12/mo or $69/yr |
How do you install a product management skill in Claude?
With an All-in-one subscription there is nothing to download. Open Customize → Connectors, click +, choose Add custom connector and paste https://mcp.kissmyskills.com/mcp. Sign in with your email, then ask for a skill by name, for example "load Cassius". The same connector works in ChatGPT, Claude Code, Codex and Cursor; setup steps are on the connector page. Signing in is free, and your first skill call starts a 7-day free trial with every skill.
Bought a single file? The skill is a .skill file, a ZIP archive with SKILL.md inside. In Claude, turn on code execution under Settings → Capabilities, then Customize → Skills → + → Upload a skill, and toggle it on. The prompt pack is a plain text file to paste into a chat or Claude Project. In ChatGPT or Gemini, paste the SKILL.md contents into custom instructions.
All-in-one: every skill, one price
All-in-one is $12 a month or $69 a year, with 7 days free and every skill included, and a 30-day money-back guarantee. Read why a subscription, see the All-in-one plan, or open the connector setup.
Related guides
- Claude and ChatGPT Prompts for Product Managers
- How to Use Claude as a Business Analyst
- How to Use Claude for Market Research
- How to Use Claude as a UX Writer
Frequently asked questions
Is there a Claude skill for product managers?
Yes. Prompts for Product Managers is a pack of 24 prompts covering discovery, PRDs, user stories and acceptance criteria, prioritisation, roadmap communication and launch metrics. For AI and data initiatives, AI & Data Product Manager AI Skill - Cassius ($19) scopes the brief and routes work to the right specialist. Both are included in the $12/month All-in-one plan.
Can Claude write a PRD?
It can draft one, but its better use is reviewing one. Ask it to act as a sceptical engineer and list ambiguous requirements, undefined limits, missing error states and untestable words like fast or intuitive, each with the question that resolves it. A draft with no open questions is a warning sign.
Can Claude write user stories with acceptance criteria?
Yes, if you require specifics: Given/When/Then criteria covering a happy path, a validation failure, a permission failure and an empty or maximum case, plus a line stating what the story does not cover. Ask it to mark anything that depends on an unanswered question rather than invent a limit.
Can Claude analyse user interviews?
Yes, from transcripts you provide. Make it separate observations (what people did and said, with a verbatim quote and participant ID) from interpretations, then check a few quotes against your notes. It cannot run the interview, and a small sample points to problems rather than sizing them.
What is Claude bad at for product management?
It invents user needs when it has no evidence, agrees with the hypothesis you state, treats small samples as proof, writes finished-looking specs that hide undecided questions, and cannot see your codebase, team capacity or stakeholder politics. Its effort estimates and prioritisation inputs are guesses until the people who build and measure confirm them.
Prompts for Product Managers
Or get it with every other skill in All-in-one: $12/month, 7 days free, every skill.