How to Use Claude Projects in 2026 (Step-by-Step Guide)
TL;DR: Claude Projects are workspaces on claude.ai where you upload files, write custom instructions, and keep all your related chats in one place so Claude remembers your context every time. This guide walks you through setup, file uploads, writing effective instructions (with a full copy-paste example), and knowing when to use Projects vs Cowork Projects instead.
Search "Claude Projects" right now and you'll hit three completely different things. One is a chat workspace on claude.ai. Another is a desktop agent feature inside Claude Cowork. The third is a developer tool called Claude Code that has nothing to do with either but ranks on the same keyword because Anthropic reused the word "Projects" across all of them.
This guide is about the first one: the Projects feature on claude.ai, the one where you upload documents, write instructions, and have focused conversations with Claude that actually remember what you're working on. If you've been confused about which "Projects" people are talking about that's not your fault, and we'll clear it up in the first section below.
Here's what we'll cover: creating a project, building your knowledge base, writing instructions Claude will actually follow (with a real example you can copy), and knowing when browser Projects are the right tool versus when you want Cowork Projects instead.
Wait Which "Claude Projects" Are You Looking For?
Three different features share the name "Claude Projects" right now, and most guides online don't tell you which one they're covering. Here's the breakdown.
Claude Projects on claude.ai is a browser-based workspace. You upload documents, write custom instructions, and start chats that all share the same context. It works on any plan including free. This is the feature most people mean when they say "Claude Projects," and it's what this guide covers.
Cowork Projects is a different feature inside the Claude Desktop app. Instead of just chatting, Claude actually executes tasks on your computer reading and writing files in local folders, running code, and maintaining memory between sessions. It requires a paid plan (Pro or higher) and the desktop app. We'll compare the two near the end of this guide.
Claude Code is a developer tool for writing software from a terminal. It has its own project structure (a CLAUDE.md file, local repo access), but it's built for programmers shipping code not for the general workspace feature you're probably looking for. If you see a guide titled "Claude Projects Best Practices" that talks about BMAD methods and CLAUDE.md files, that's Claude Code, not this.
| Feature | What It Is | Where It Lives | Who It's For |
|---|---|---|---|
| Claude Projects | Chat workspace with shared files and instructions | claude.ai (browser) | Anyone free or paid |
| Cowork Projects | Desktop agent with local file access and execution | Claude Desktop app | Paid plans (Pro+) |
| Claude Code | Terminal-based coding tool | Command line / VS Code | Software developers |
Swipe sideways to see all columns
If you've Googled 'Claude Projects' and landed on a page about coding workflows and CLAUDE.md files you weren't confused. The naming is genuinely overlapping, and this table is something Anthropic hasn't published in one place yet."
Before You Start What You Need
You don't need a paid plan to use Claude Projects. Free accounts can create up to five projects with a standard knowledge base.
That said, the experience changes meaningfully on paid plans. Here's what each tier gives you:
Free ($0): Up to 5 projects. Standard context window. You can upload files and write instructions, but if your knowledge base gets large, Claude may not be able to reference everything in a single chat.
Pro ($20/month): Unlimited projects. When your uploaded content approaches the context limit, Claude automatically enables RAG mode expanding your project's effective capacity by up to 10x. You also get access to Claude Cowork, longer conversations, and the ability to choose between models.
Max ($100 or $200/month): Everything in Pro, with 5x or 20x the usage limits. Worth considering only if you're hitting Pro's rate limits daily.
Team ($25/seat/month) and Enterprise (custom pricing): The sharing features unlock here. You can share projects with teammates, set view/edit permissions, and use projects as shared knowledge bases across your organization. Current pricing is on claude.com/pricing.
We verified these details against Anthropic's official help center and pricing page in August 2026. Several guides published earlier in 2026 still describe Projects as a paid-only feature that's outdated. Free users have had access since Anthropic expanded availability.
Step 1: Create Your First Project
Creating a project takes about 30 seconds. Go to claude.ai/projects you'll see a Projects section in the left sidebar, or you can navigate there directly.
Click "+ New Project" in the upper right corner. You'll be asked for two things: a name and an optional description. Pick something you'll recognize later "Q3 Marketing Research" is more useful than "Project 1."
One detail most guides skip: Claude doesn't see your project name or description. They're for your own organization only. If you want Claude to know what the project is about, you'll need to tell it in the project instructions (Step 3 below).
If you're on a Team or Enterprise plan, you'll also see a visibility setting during creation. You can keep the project private (only you and people you invite) or make it visible to your entire organization. For personal accounts on Free or Pro, projects are private by default there's no sharing option.
After you click create, you'll land on the project's main page. The left side shows your chat history within this project. The right side is where you'll build your knowledge base which is what we're doing next.
Step 2: Build Your Knowledge Base
Your project's knowledge base is what makes the project useful. Anything you upload here becomes context that Claude can reference across every chat inside this project without you re-uploading or re-explaining each time.
Click the "+" button in the knowledge section (right side of your project page) to add content. You have two options:
Upload files. Claude accepts PDF, DOCX, CSV, TXT, HTML, ODT, RTF, and EPUB files. Each file can be up to 30MB, and you can upload as many files as you want though the total usable content is limited by the context window. On paid plans, when your knowledge base approaches that limit, Claude switches to RAG mode automatically, pulling in only the relevant sections instead of loading everything at once.
Paste text directly. You don't always need a file. You can paste a webpage, an email thread, or meeting notes straight into the knowledge base. This is faster when you're working with content that doesn't live in a document.
A few things worth knowing: the knowledge base applies to every chat in the project, but individual chats don't share their conversation history with each other. If you tell Claude something important in Chat #1, Chat #2 won't know about it unless that information is in the knowledge base itself. Think of the knowledge base as the "shared brain" and each chat as a separate conversation with that same brain.
Step 3: Write Project Instructions (With a Full Working Example):
Project instructions are the single biggest difference between a project that saves you time and one that makes you repeat yourself every chat. They tell Claude how to behave, what to focus on, and what to avoid and they apply automatically to every conversation inside the project. Click "Set project instructions" on your project page, write your instructions, and click Save. That's the mechanical part. The harder part is writing instructions that actually work. Here's what separates good instructions from vague ones: specificity. "Be helpful and concise" tells Claude nothing it doesn't already do. "Fix grammar without changing meaning, flag unclear sentences instead of rewriting them silently, and return edits in bold" tells Claude exactly what you expect. A full working example Freelance Content Editor project: You are helping me edit blog posts for my freelance clients.
Voice and tone:
- Write in active voice. Keep sentences under 20 words when possible.
- Match the client's brand voice (I'll paste their style guide in each chat).
What to do:
- Fix grammar and spelling without changing the author's meaning.
- Flag sentences that are unclear don't rewrite them silently.
- Suggest stronger headlines when the original is generic.
What NOT to do:
- Don't add content I didn't ask for.
- Don't use "delve," "crucial," "landscape," or "game-changing."
- Don't change technical terminology unless I ask.
Format: Return the edited text with changes in bold, followed by a short summary of what you changed and why.
This works because it covers three things most instructions miss. First, it defines what Claude should do not just a role, but specific actions. Second, it sets boundaries on what Claude should not do, which prevents the kind of over-helpful behavior that makes extra work. Third, it specifies the output format, so you don't spend the first message of every chat saying "actually, can you bold the changes?" Anthropic's own help center recommends keeping project instructions concise and focused on "general context, key guidelines, and Claude's role." Every token in your instructions reduces the space available for your actual files and conversation. "The difference between a project that saves you 20 minutes a day and one you abandon after a week is usually five minutes of writing specific instructions."
Step 4: Start Chatting Inside Your Project
Once your knowledge base and instructions are set, every new chat you open from the project page automatically gets both. You don't need to re-upload files or remind Claude of the rules it already has them.
To start a chat, click inside the project and begin typing. Claude will have access to everything in the knowledge base and will follow the instructions you wrote. If you want to test whether your instructions are working, try asking Claude to summarize what it knows about the project. Any gaps will show up immediately.
You can also move existing standalone chats into a project. Click the dropdown arrow next to any chat name and select "Add to project." You can search for the right project in the modal that appears. Moving chats out of a project or between projects works the same way from the dropdown menu. Bulk moves are available from your chat history page: select multiple chats and use the move icon.
One thing that catches people off guard: moving a chat into a project means Claude's memory feature (on paid plans) will associate that conversation with the project's memory instead of your general memory. Each project keeps its memory separate. That's useful for keeping work contexts from bleeding into each other your legal research project won't influence how Claude responds in your marketing project.
How to Use Skills Inside a Claude Project
Skills are pre-written instruction sets that give Claude expertise in a specific area like writing in a particular style, analyzing data a certain way, or following a review framework. They're not a separate feature you toggle on. They're just well-structured instructions you can add to your project.
The concept comes from Claude Code (the developer tool), where users create .md files that Claude loads automatically. In regular Projects on claude.ai, the equivalent is adding skill-like instructions to your project knowledge base or project instructions. You're giving Claude a reusable playbook that applies to every chat.
Here's how to do it in practice. Say you run a small content agency and you want a project where Claude edits drafts, checks SEO basics, and flags readability issues three different "skills" in one workspace.
Option 1: Add them to your project instructions. Combine the core rules from each skill into one instruction block. Keep it under a page. If your instructions start getting long, you're probably trying to make Claude do too many unrelated things in one project split it into two projects instead.
Option 2: Upload skill documents to the knowledge base. Write each skill as a separate text file (e.g., seo-checklist.txt, style-guide.txt) and upload them. Claude will reference them when relevant. This works better when your skills are detailed enough that cramming them into the instructions field would eat too much context.
The reason this matters: most people set up a project with files and generic instructions, then spend every chat re-explaining how they want Claude to think. A project with well-defined skills means Claude already knows your framework before you type a word.
Sharing Projects with Your Team
Project sharing is available on Team and Enterprise plans only. If you're on Free or Pro, your projects stay private visible only to you.
To share a project, open it and click "Share project" next to the project name. You can add individual members by name or email, or paste a list of email addresses for bulk sharing. Each person gets one of two permission levels: Can view (they can see everything and chat inside the project, but can't edit it) or Can edit (they can modify instructions, update the knowledge base, and manage other members).
You can also make a project visible to your entire organization during creation or afterward unless an admin has disabled org-wide sharing. Shared projects appear under a "Shared with me" tab on the Projects page, and members get an email notification when someone shares a project with them.
A practical use case: set up a project with your company's brand guidelines, product docs, and a set of instructions for how Claude should handle customer-facing content. Share it with your marketing team as "Can view." Everyone chats with the same context, follows the same rules, and gets consistent output without each person setting up their own version from scratch.
Claude Projects vs Cowork Projects Which Should You Use?
Claude Projects and Cowork Projects solve different problems. The chat-based Projects on claude.ai are built for conversations with context. Cowork Projects on the desktop app are built for execution Claude reads your local files, creates documents, runs code, and remembers what it did across sessions.
Anthropic launched Cowork in January 2026 as a research preview, and added Projects to Cowork in March 2026. As of August 2026, Cowork runs in the cloud in beta for Team and Enterprise plans, alongside the existing desktop version.
The confusion between the two is real. Even experienced users mix them up in Reddit threads and newsletters, because Anthropic uses the same word for both. Here's the actual difference:
Claude Projects vs Cowork Projects Key Differences
| Feature | Claude Projects (claude.ai) | Cowork Projects (Desktop) | Source |
|---|---|---|---|
| Where it lives | Browser claude.ai | Claude Desktop app (+ cloud beta) | Anthropic |
| What Claude does | Chats using your files and instructions | Executes tasks creates files, runs code, browses | Anthropic |
| File access | Upload to cloud-based knowledge base | Reads and writes local folders on your computer | Anthropic |
| Memory | Chat history stays in project; separate memory per project (paid) | Scoped memory remembers past sessions within the project | Anthropic |
| Team sharing | Yes Team and Enterprise plans | Not yet (local storage only) | Anthropic |
| Scheduled tasks | No | Yes set recurring tasks on a schedule | Anthropic |
| Plan required | Free (up to 5) or any paid plan | Paid only (Pro, Max, Team, Enterprise) | Anthropic |
Swipe sideways to see all columns
When to use Claude Projects: You want a shared knowledge base for ongoing conversations. You work in a team and need everyone chatting with the same context. You don't need Claude to create files or run code you need it to think, answer, and draft within a chat window.
When to use Cowork Projects: You want Claude to actually do the work write reports, organize files, process data. You have a recurring workflow (weekly reports, client proposals) that you want to automate with scheduled tasks. You work primarily from a desktop machine and want Claude to access your local folders directly.
You can also import a Claude Project into a Cowork Project. If you already built a knowledge base and instructions on claude.ai, you don't have to start over the desktop app lets you pull everything in and layer execution capabilities on top.