Claude Projects are persistent workspaces where you can:

  • Set custom instructions that apply to every conversation in the project
  • Upload documents that Claude can reference throughout the project
  • Maintain separate context for different work streams
  • Share with team members (on Claude Team plans)

If you use Claude for the same type of work repeatedly — writing in a specific style, analyzing documents in a domain, coding in a particular codebase — Projects will significantly improve your experience.

Creating a Project

  1. In the Claude sidebar, click “New project”
  2. Give it a descriptive name (e.g., “Blog Content”, “API Documentation”, “Legal Research”)
  3. You’re in the project — start configuring

Setting Up Project Instructions

Project instructions are the most powerful feature. They’re like a permanent system prompt that applies to every conversation in this project.

Click “Edit project instructions” and write your configuration.

Example 1: Content writing project

You are my writing assistant for my tech newsletter targeted at software engineering managers.

Voice: conversational but intelligent, assumes technical knowledge, avoids marketing jargon, uses specific examples over generalities

Audience: engineering managers with 5-15 years of experience, leading teams of 5-30 engineers

Format defaults:
- Keep sentences under 25 words
- Use active voice
- Avoid "leverage", "synergy", "utilize"
- Headers for pieces over 600 words

When I ask for help writing, produce a complete draft. When I ask for edits, only change what I ask about.

Example 2: Code review project

You are reviewing Python code for a financial data processing application.

Context: FastAPI backend, PostgreSQL with SQLAlchemy, Celery for async tasks, Redis cache.

Code standards:
- Type hints required on all function signatures
- Docstrings required on public functions
- Error handling must use custom exception classes from exceptions.py
- Database queries must not be in route handlers

When reviewing code:
1. Security issues first (highest severity)
2. Logic errors second  
3. Performance issues third
4. Style/standards last

Format findings as: [SEVERITY] File:line — Issue description — Suggested fix

Example 3: Research project

I'm researching the history of programming language design for a book aimed at working programmers.

Style: Accessible but technically accurate. Assume reader knows at least one programming language but not PL theory.

Key themes I'm exploring:
- How economic constraints shaped language design
- The gap between academic language research and mainstream adoption
- How communities form around languages

When I ask questions, help me understand the topic and surface interesting angles I might not have considered. Flag when historical claims need verification.

Adding Knowledge Files

Projects can hold documents Claude references throughout all conversations. Useful for:

  • Style guides, brand guidelines
  • Technical documentation for a specific codebase
  • Research papers or reports you’re working with
  • Templates or reference examples
  • Your own past writing as a voice reference

To add files:

  1. In the project, click “Add content” or the upload button
  2. Upload PDF, text files, or paste content directly

What works well as project knowledge:

  • Documents you refer to repeatedly
  • Reference material that’s too long to paste every conversation
  • Your style guide or voice examples
  • Technical specifications or API documentation

What doesn’t work well:

  • Very large files (Claude’s context window limits how much it can use)
  • Dynamic content that changes frequently (upload an updated version when it changes)
  • Content that’s easily searchable in real-time (Claude can’t browse the web from within Projects)

Organizing Multiple Projects

Effective project organization depends on your work:

By domain/topic:

  • “Marketing Copy”
  • “Technical Documentation”
  • “Research: [Topic]”

By client or context:

  • “Client: Acme Corp”
  • “Internal: Engineering”

By workflow:

  • “Writing: First Drafts”
  • “Writing: Editing”
  • “Code Review”

Don’t over-engineer it. Create a project when you find yourself typing the same context instructions in every conversation. That’s the signal that a project would help.

Effective Project Conversations

Start conversations with the task, not the context: Because the context is already in the project instructions, you don’t need to re-explain. Just say “Draft a newsletter intro about [topic]” instead of explaining your voice, audience, and format again.

Use the knowledge files with @ references: In conversation, use @[filename] to reference a specific uploaded document. Claude will focus on that document.

Don’t hesitate to correct: If Claude’s response doesn’t match your project instructions, say “Remember, we’re using [X format/voice/approach].” You can also update the instructions if you find them consistently off.

Keep project conversations focused: Long, sprawling conversations in a project context can lose quality. For a new major topic, start a new conversation within the project.

Sharing Projects with Teams

On Claude Team plans, you can share projects with your team:

  1. In the project settings, click “Share project”
  2. Invite team members by email
  3. Set permissions (view, contribute, or admin)

Useful team project types:

  • Company knowledge base: Upload your documentation, brand guidelines, and style guides. Everyone on the team gets consistent AI responses.
  • Codebase context: Upload architecture docs, coding standards, and conventions. New engineers get faster onboarding.
  • Sales materials: Upload product documentation, competitor analysis, and talking points. Sales team gets consistent, accurate responses.

Common Mistakes

Instructions too vague: “Be helpful and clear” is useless. Specify format, voice, constraints, and what to prioritize.

Too many conflicting instructions: Keep instructions focused on the most important 5-10 things. More isn’t always better.

Uploading irrelevant knowledge: Every uploaded file takes up context. Upload only what Claude actually needs for this project’s work.

Not updating instructions as you learn: After a few weeks, you’ll know which instructions matter and which you never needed. Trim and refine.

Example: The Research Project Setup

Here’s a complete example for setting up a research project:

Project name: “AI Policy Research 2026”

Instructions:

I'm researching AI regulation and policy for a series of policy briefs targeted at lawmakers and their staff.

Audience: Non-technical policy professionals. Assume intelligence but no AI/ML background.

My position: I'm neutral on most questions but lean toward outcome-based rather than prescriptive regulation.

When I share a document or ask a question:
- Summarize key points in plain language
- Note where the document's claims are contested or uncertain
- Identify implications that might not be obvious
- Flag when I should verify claims against primary sources

Format: Use headers, bullet points for summaries. Plain prose for analysis.

Knowledge files uploaded:

  • EU AI Act full text
  • US Executive Order on AI
  • Key academic papers on AI risks
  • My own previous policy brief as a voice example

This setup means every conversation in this project starts with Claude already knowing your context, audience, constraints, and writing style. No re-explaining.