No terminal. No installation. Just paste and go.
Claude Projects give you persistent instructions, so the discovery copilot stays active across conversations.
- Go to claude.ai and create a new Project
- Name it something like "Product Discovery"
- Click "Set project instructions" and paste the entire contents of
system-prompt.md - Optionally, upload the template files from
templates/as project knowledge
Start a new conversation in the project and say:
"I want to start product discovery for [your problem area]"
The copilot will guide you through the full cycle:
- Frame the problem (starts with a braindump)
- Plan interviews (Mom Test questions, bias warnings)
- Synthesize findings (raw notes in, opportunity tree out)
- Package evidence (stakeholder one-pager with GO / PIVOT / KILL)
- Each phase works independently. You can say "I just need an interview guide" and skip straight there.
- Paste your raw interview notes directly into the chat for synthesis. Messy is fine.
- Ask it to export any artifact as a shareable markdown doc.
- Go to chatgpt.com
- Start a new conversation
- Paste the contents of
system-prompt.mdas your first message, prefixed with:
"Use the following as your instructions for this conversation. After reading them, confirm you're ready and wait for me to start."
- Go to Explore GPTs > Create
- In the Instructions field, paste the contents of
system-prompt.md - Name it "Discovery Copilot"
- Save and use it anytime
- ChatGPT doesn't have persistent project instructions in free tier, so you may need to re-paste for new conversations.
- Custom GPTs (Plus/Team plans) solve this -- the instructions persist.
The system prompt works with any AI assistant that follows instructions well.
- Start a new conversation
- Paste the contents of
system-prompt.mdas your first message - Say "I want to start product discovery"
Quality will vary by model. Claude and GPT-4 class models work best for the conversational push-back that makes this tool valuable.
Each phase produces a structured artifact you can copy-paste into your team's tools:
| Phase | What you say | What you get |
|---|---|---|
| Frame | "I want to explore [problem]" | Discovery brief with hypothesis + research questions |
| Interview | "Help me plan interviews" | Mom Test interview guide with bias warnings |
| Synthesize | "Here are my notes: [paste]" | Themes, pain ranking, opportunity tree |
| Package | "Help me package this for stakeholders" | One-pager with GO/PIVOT/KILL recommendation |
All outputs are markdown. Works in Notion, Confluence, Google Docs, Linear, Jira, email, or anywhere else.
| Chat version | CLI version (Claude Code) | |
|---|---|---|
| Setup | Paste system prompt | Install as skill |
| File output | Copy-paste from chat | Auto-saves to project |
| Persistence | Per-project (Claude.ai) or per-session | Permanent in project |
| Quality | Same discovery logic | Same discovery logic |
The discovery methodology is identical. The only difference is how artifacts are saved.
- Want the full CLI experience? See the main README for Claude Code / Cursor / Windsurf setup.
- New to AI coding tools? Check out the Claude Code Guide to get started.
- Questions? Open an issue or reach out on LinkedIn.