How to Use Claude Projects - Complete Guide to Setting Up Per-Project AI Workspaces

Introduction: Why Claude Projects Changes How You Work with AI

If you’ve been using Claude by starting a new conversation every time you need help, you’re leaving enormous productivity on the table. Claude Projects is Anthropic’s answer to one of the most persistent frustrations with AI assistants: context loss. Every new chat means re-explaining who you are, what you’re working on, and what rules the AI should follow.

Claude Projects lets you create dedicated AI workspaces where your instructions, reference documents, and conversation history live together in one persistent environment. Think of it as giving Claude a dedicated desk for each area of your work — complete with reference manuals, style guides, and sticky notes about your preferences.

This guide is written for professionals, developers, content creators, and teams who use Claude regularly and want to move beyond one-off conversations. Whether you’re managing a software project, running a content pipeline, or building a research knowledge base, you’ll learn how to set up project workspaces that make Claude dramatically more useful.

By the end of this guide, you will have:

A clear mental model of how Claude Projects works and when to use it
At least one fully configured project workspace tailored to your real work
A system for organizing multiple projects without overlap or confusion
Strategies for writing project instructions that consistently produce high-quality outputs

Difficulty: Beginner to Intermediate | Time required: 15–30 minutes for your first project, 5 minutes for each additional one

Prerequisites

Before you start, make sure you have the following:

A Claude Pro, Team, or Enterprise subscription. Claude Projects is available on paid plans. As of early 2026, Pro costs $20/month and Team costs $30/user/month. The free tier does not include Projects.
Access to claude.ai. Projects are managed through the web interface at claude.ai. The feature is also accessible through the Claude mobile apps, though setup is easier on desktop.
Your reference materials ready. Gather any documents, style guides, code snippets, or instructions you want Claude to reference. Supported formats include PDF, TXT, CSV, and common code file types. Each project supports up to 200K tokens of project knowledge (roughly 150,000 words).
A clear use case in mind. You’ll get the most from this guide if you have a specific workflow — like “help me write documentation for my API” or “assist with weekly marketing reports” — rather than a vague goal.

Step-by-Step: Setting Up Your First Claude Project

Step 1: Access the Projects Dashboard

Log in to claude.ai and look at the left sidebar. You’ll see a section labeled Projects beneath your recent conversations. Click the ”+ New Project” button. If you’re on a Team or Enterprise plan, you’ll also see shared team projects here.

Give your project a clear, descriptive name. Avoid generic names like “Work Stuff” or “AI Help.” Instead, use names that immediately tell you the scope: “Q1 Blog Content Pipeline,” “Backend API Documentation,” or “Client Proposal Drafts – Acme Corp.”

Tip: If you manage multiple clients or products, include the client or product name in the project title. Six months from now, you’ll thank yourself for the specificity.

Step 2: Write Your Project Instructions (Custom System Prompt)

This is the single most important step. Project instructions are persistent directions that Claude follows in every conversation within the project. They replace the need to repeat yourself at the start of each chat.

Click on the project you just created, then find the “Project instructions” or “Custom instructions” field. Here’s a framework for writing effective instructions:

Role definition: Tell Claude what role it should play. Example: “You are a senior technical writer helping maintain API documentation for a Python REST framework.”
Context about your work: Briefly describe the project, its goals, and its current state. Example: “We’re building a SaaS platform for inventory management. The backend is FastAPI with PostgreSQL. We’re in beta with 50 pilot customers.”
Output rules: Specify formatting, tone, length, and style preferences. Example: “Write in active voice. Use second person (‘you’) for user-facing docs. Keep paragraphs under 4 sentences. Always include code examples in Python.”
Constraints: List things Claude should avoid. Example: “Never suggest deprecated endpoints. Don’t recommend packages not already in our requirements.txt. Don’t add type hints to test files.”
Recurring tasks: If you’ll ask for the same type of work repeatedly, define the template. Example: “When I ask you to document an endpoint, use this structure: Description, Authentication, Parameters (table), Request Example, Response Example, Error Codes.”

Example of a complete project instruction set:

You are a senior content strategist for a B2B SaaS company that sells project management tools to mid-market teams (50-500 employees).


Our brand voice is: professional but approachable, data-driven,
never condescending. We write at an 8th-grade reading level.
When writing blog posts:

Target 1,200-1,800 words
Include 1 data point or statistic per major section
End with a clear CTA linking to our product
Use H2 for main sections, H3 for subsections
Never use the word “synergy” or “leverage” as a verb

When reviewing drafts I paste:


Check for brand voice consistency
Flag any claims without supporting data

Suggest stronger headlines using the CoSchedule formula

**Tip:** Start with 10-15 lines of instructions. You can always refine them after a few conversations. Overly long instructions (2,000+ words) can dilute the most important rules.

Step 3: Upload Project Knowledge Files

Project knowledge is where Claude Projects truly differentiates itself from regular conversations. You can upload reference documents that Claude will consult across every conversation in the project — without you needing to paste them in each time.

Click “Add project knowledge” (or the equivalent button in your interface) and upload your files. Good candidates for project knowledge include:

Style guides and brand guidelines — so Claude matches your voice consistently
API documentation or schema files — so Claude knows your data structures
Product requirement documents (PRDs) — so Claude understands feature context
Code samples or architecture overviews — so Claude follows your patterns
Glossaries or terminology lists — especially useful for technical or industry-specific work
Past examples of good output — “here’s a blog post we loved, match this quality”

You can upload multiple files, and Claude will reference all of them when relevant. The total project knowledge limit is approximately 200K tokens across all files.

Tip: Don’t dump everything you have. Curate your uploads. A focused set of 3-5 highly relevant documents outperforms 20 loosely related files. Claude prioritizes relevance, but noise reduces signal.

Step 4: Start Your First Project Conversation

With instructions set and knowledge uploaded, start a new conversation inside the project. You’ll notice the difference immediately — Claude already knows your context, preferences, and reference materials.

Test your setup with a representative task. If your project is for content writing, ask Claude to draft a section. If it’s for code review, paste a code snippet and ask for feedback. Evaluate whether Claude:

Follows your specified tone and format
References your uploaded documents when relevant
Respects your constraints (avoids things you told it to avoid)
Produces output that matches your quality bar

If something feels off, don’t start over. Iterate on your project instructions based on what you observe.

Step 5: Refine Instructions Based on Real Usage

Your first version of project instructions is a draft. After 3-5 conversations, you’ll notice patterns: maybe Claude keeps using passive voice despite your instructions, or it’s not citing the uploaded docs enough.

Go back to your project settings and update the instructions. Common refinements include:

Adding explicit “Do / Don’t” lists based on repeated issues
Including a short example of ideal output format
Specifying how Claude should handle ambiguity (“If my request is unclear, ask me a clarifying question before proceeding”)
Adding workflow-specific shortcuts (“When I say ‘new post,’ use the blog post template. When I say ‘social,’ generate 3 LinkedIn post variations.”)

Tip: Keep a running note (even in the project itself) of things you find yourself correcting. After a week, batch those corrections into an instruction update.

Step 6: Organize Multiple Projects by Domain

Once you’ve set up your first project, you’ll want more. The key is organizing them so you always know which project to open. Here’s a taxonomy that works well:

Category	Example Project Names	What Goes Here
Product/Engineering	"Backend API Docs," "Frontend Component Library"	Code-related work, technical documentation, architecture decisions
Content/Marketing	"Blog Pipeline," "Email Sequences Q1"	Writing, editing, content strategy, SEO
Research/Analysis	"Market Research – APAC," "Competitor Intelligence"	Data analysis, report generation, synthesis
Operations	"SOPs and Runbooks," "Vendor Evaluation"	Process documentation, decision frameworks
Personal/Learning	"Learning Rust," "Book Notes 2026"	Study aids, concept exploration, skill building

**Rule of thumb:** Create a new project when the context, instructions, or reference materials differ meaningfully from an existing project. If two workstreams share 90% of the same context, they can share a project. If they share less than 50%, split them.

Step 7: Use Starred Conversations and Naming Conventions

Within each project, conversations accumulate quickly. Develop a naming convention from day one. Effective patterns include:

Date-first: “2026-03-18 — Homepage copy revision v2”
Task-type prefix: “[DRAFT] Q1 investor update” or “[REVIEW] Auth module refactor”
Outcome-based: “Final approved version — onboarding email sequence”

Star important conversations — the ones containing approved outputs, key decisions, or reference threads you’ll revisit. This prevents important work from getting buried under daily chats.

On Claude Team and Enterprise plans, projects can be shared across your organization. This is where Projects become a genuine team tool rather than a personal productivity hack.

To share a project, open the project settings and adjust the visibility. Options typically include:

Private: Only you can see and use the project
Team-visible: Team members can view and start conversations in the project
Team-editable: Team members can also modify instructions and upload knowledge

Shared projects are excellent for onboarding new team members, maintaining consistent brand voice across writers, and creating standardized workflows that the whole team follows.

Tip: Designate one person as the “project owner” who maintains the instructions and knowledge files, even if the whole team uses the project. This prevents instruction drift from too many editors.

Common Mistakes and How to Avoid Them

1. Writing Vague Instructions

The mistake: Instructions like “Be helpful and write well” give Claude nothing actionable to work with. You’ll get generic output that requires heavy editing.

Instead, do this: Be specific and concrete. Replace “write well” with “Write in active voice, keep sentences under 20 words, use Oxford commas, and format technical terms in code backticks.” The more specific your rules, the more consistent Claude’s output.

2. Uploading Too Many Knowledge Files

The mistake: Dumping 30 PDFs into project knowledge because “Claude might need them.” This dilutes relevance and can cause Claude to reference tangential information instead of the most pertinent documents.

Instead, do this: Start with 3-5 essential documents. Add more only when you notice Claude lacking specific context. Quality and relevance beat quantity every time. If a document is only useful for one conversation, paste it into that conversation instead of adding it to project knowledge.

3. Using One Mega-Project for Everything

The mistake: Creating a single project called “Work” and using it for code reviews, blog writing, data analysis, and email drafting. The instructions become contradictory, and knowledge files become irrelevant to most tasks.

Instead, do this: Split by domain or workflow. It takes 5 minutes to create a new project. The payoff in output quality is immediate. A focused project with 10 lines of specific instructions outperforms a catch-all project with 100 lines of general guidance.

4. Never Updating Project Instructions

The mistake: Setting instructions once and never revisiting them, even as your project evolves, your preferences change, or you discover Claude’s output patterns.

Instead, do this: Schedule a 10-minute review of your project instructions every two weeks. Look at your recent conversations — what did you correct repeatedly? Add those corrections to the instructions. Remove rules that no longer apply. Instructions are living documents.

5. Ignoring Conversation Management

The mistake: Letting dozens of unnamed conversations pile up in a project, making it impossible to find past work or reference earlier outputs.

Instead, do this: Name every conversation with a clear description immediately after starting it. Star conversations that contain important outputs. Archive or delete test conversations you won’t need again.

Frequently Asked Questions

Can I use Claude Projects with the API, or is it only available on claude.ai?

Claude Projects as a managed workspace is a claude.ai feature. However, you can replicate similar behavior through the API by using system prompts (equivalent to project instructions) and including reference content in your requests. The API approach requires you to manage the context yourself, but it offers more programmatic control. For most non-developer use cases, the claude.ai Projects interface is simpler and more practical.

What’s the difference between project knowledge and just pasting content into a conversation?

Project knowledge persists across all conversations within the project — you upload it once and it’s available forever. Content pasted into a conversation only exists in that single conversation thread. Use project knowledge for stable reference material (style guides, schemas, SOPs) and paste content for one-time requests (“review this specific document”). Project knowledge also doesn’t count against your conversation’s context window in the same way, giving you more room to work.

Is there a limit to how many projects I can create?

There’s no strict cap on the number of projects. Practical limits depend on your plan and usage patterns. Most power users maintain 5-15 active projects. If you find yourself with more than 20, consider whether some can be consolidated or archived. The key constraint is the 200K token limit per project for knowledge files, not the number of projects themselves.

Can I transfer conversations between projects?

As of early 2026, direct conversation transfer between projects is not natively supported. If you need to move context from one project to another, the practical workaround is to copy the relevant conversation content and paste it into a new conversation in the target project. For reference materials, simply download from one project and upload to the other. Anthropic has indicated that improved project management features are on their roadmap.

How do Claude Projects compare to custom GPTs in ChatGPT?

Both features let you create specialized AI configurations, but they differ in architecture. Claude Projects emphasize persistent workspaces where conversations accumulate context over time, with strong support for uploaded reference documents up to 200K tokens. Custom GPTs focus more on creating shareable, pre-configured assistants with specific capabilities. Claude Projects are generally stronger for ongoing professional work with rich context, while Custom GPTs excel at creating reusable tools for specific, repeatable tasks. The right choice depends on whether your primary need is a persistent workspace (Claude Projects) or a reusable template (Custom GPTs).

Summary and Next Steps

Here’s what you’ve learned in this guide:

Claude Projects create persistent AI workspaces where your instructions, reference materials, and conversation history live together
Project instructions are the highest-leverage element — specific, concrete instructions produce dramatically better output than vague ones
Project knowledge should be curated, not dumped — 3-5 focused documents beat 30 loosely related files
Organization matters — name projects clearly, split by domain, name conversations, and star important threads
Iterate regularly — review and update your project instructions every two weeks based on real usage patterns
Team sharing amplifies value — shared projects with consistent instructions create organizational knowledge that scales

Your next steps:

Create your first project right now. Pick your most frequent AI use case and spend 15 minutes setting up instructions and uploading 2-3 reference files.
Run 5 test conversations within the project over the next few days. Note what works and what needs adjustment.
Refine your instructions based on those 5 conversations. Add specific corrections and examples of ideal output.
Expand to 3-5 projects covering your main work domains. Use the taxonomy table above as inspiration.
If you’re on a Team plan, share one project with a colleague and compare outputs to ensure consistency.

Claude Projects isn’t a feature you configure once and forget. The teams and individuals getting the most value treat their projects like living systems — regularly updated, carefully organized, and tightly focused on specific workflows. Start small, iterate based on real results, and expand as you see the quality difference.

Explore More Tools