To write a PRD using Claude Code, start by giving Claude your product context (company, users, goals). Then ask it to research the problem space. Next, draft each PRD section: problem statement, proposed solution, success metrics, and scope. Iterate with specific feedback. A complete PRD takes 20 to 30 minutes with Claude Code, versus 4 to 8 hours by hand.
Each step includes the exact Claude Code prompt you can copy and use. Swap in your own details and run them in order.
Give Claude Code your product context before starting. It performs dramatically better when it knows your company, product, audience, and what you already know about the feature.
Claude Code Prompt
Tip: The more specific your context, the less editing you do later. Include your current sprint focus, key stakeholders, and any constraints.
Have Claude Code research competitors, user pain points, and existing solutions. AI-powered research surfaces patterns across sources faster than manual scanning.
Claude Code Prompt
Tip: Point Claude at specific competitors or research files you already have. If you have user interview transcripts, tell it to read those first.
Get Claude to synthesize research into a clear, bounded problem statement. A strong problem statement makes every other PRD section easier to write.
Claude Code Prompt
Tip: Ask for specificity — numbers, user segments, metrics. If Claude gives you "users are frustrated," push back with "what percentage of users? Which segment? What do they do instead?"
Collaborative solution design is where you add the most product judgment. Let Claude draft, then challenge it.
Claude Code Prompt
Tip: Push back on Claude's first draft. Ask "what could go wrong?" and "what are we missing?" The second draft is almost always stronger.
Define how you will measure success. Claude tends to suggest vanity metrics by default, redirect it toward outcomes.
Claude Code Prompt
Tip: Claude often suggests "increase user satisfaction" or "reduce time spent." Redirect: "give me a metric I can measure in Mixpanel within 30 days of launch."
Scope the work into phases. The key discipline: Phase 1 should be the smallest version that validates your core hypothesis.
Claude Code Prompt
Tip: Always ask Claude to make Phase 1 smaller. Whatever it suggests first, say "cut this in half — what is the absolute minimum we need to learn if this idea works?"
The review step is where Claude Code adds the most value. It catches gaps that are invisible when you have been staring at your own document.
Claude Code Prompt
Tip: Run the review prompt twice: once for content gaps, once for "what would a skeptical engineering lead push back on?" Two passes with different lenses catch more issues than one.
After running through the 7 steps, your PRD should follow this structure. Use it as a checklist to make sure nothing is missing.
Overview
Problem statement, background, and context
Goals and Success Metrics
Primary, secondary, and guardrail metrics with baselines
User Stories / Use Cases
Core scenarios with user personas
Proposed Solution
High-level approach, user flows, mockups if available
Technical Considerations
Dependencies, API changes, data model impacts
Scope
In-scope, out-of-scope, future phases
Timeline and Milestones
Phase dates, key deliverables, review checkpoints
Risks and Mitigations
Known risks with severity and mitigation plans
Open Questions
Unresolved decisions needing stakeholder input
Real Claude Code prompts you can copy-paste — not generic "use AI to brainstorm" advice.
Tested workflow from a PM with 8+ years of experience. These prompts come from writing real PRDs that shipped, not hypothetical examples.
Includes the review/iteration step that most AI writing guides skip. First drafts are starting points, the review step is where Claude Code adds the most value.
Links to 243 additional PM workflows in the AI PM OS — PRD writing is one workflow out of hundreds.
A complete PRD takes 20-30 minutes with Claude Code, compared to 4-8 hours of manual writing. The biggest time savings come from research synthesis, first-draft generation, and the review step. You still need to apply judgment and edit, but you start from a strong draft instead of a blank page.
No. Claude Code handles roughly 80% of the drafting and research work. The remaining 20% needs a human PM: product judgment, stakeholder context, strategic trade-offs, and organizational politics. Think of it as a highly capable research assistant that writes fast first drafts. It is not a replacement for product thinking.
Start with context. Tell Claude your company, product, users, and the feature you are working on. Then be specific about each PRD section. Do not ask for the whole document at once. A prompt like "Based on the research, write a problem statement that includes who is affected, what the problem is, current workarounds, and business impact" produces much better output than "write me a PRD."
Claude Code is better for PRDs because it handles multi-step workflows on its own. It can read your existing research files. It can pull context from Linear or Notion via MCP. It can draft each section in order. Then it can review its own output for gaps. You do not re-prompt between steps. ChatGPT requires you to manage the conversation and copy-paste context by hand.
No. Writing a PRD with Claude Code uses plain English prompts. You describe what you need ("draft a problem statement for our checkout redesign") and Claude Code produces the document. The only technical step is the one-time installation of Claude Code itself, which takes under five minutes.
The AI PM OS includes Claude Code skills for user research, competitive analysis, stakeholder updates, metrics definition, and more, all with the same copy-paste prompts.
Written by George Nurijanian, Product Manager with 8+ years of experience. For the complete toolkit, see AI PM OS.