Claude Projects lets you create dedicated workspaces in Claude with persistent context — files, instructions, and knowledge that Claude references in every conversation within that project. Set up a project once, and every new chat starts with full context about your work, your preferences, and your standards.
It's the most underused feature in Claude Pro. Most users never create a single project. They paste the same context into every conversation, re-explain their role every time, and wonder why Claude's output feels generic. Projects fix all of this.
What Are Claude Projects?
A Project is a container with three elements:
Project files: Documents Claude can reference — your style guide, product docs, code samples, data schemas, previous reports. Upload up to 200K tokens of files per project. Claude reads these before responding to any message in the project.
Custom instructions: Persistent directions that apply to every conversation — your role, your formatting preferences, your company's voice, topics to avoid. Think of these as a system prompt that never needs repeating.
Conversations: Each project can hold multiple conversation threads. All conversations in a project share the same files and instructions. Start a new conversation for a new task; the context carries over automatically.
The practical effect: instead of explaining who you are, what you're working on, and how you want responses formatted in every single conversation, you explain once. Then every interaction in that project is contextually aware from the first message.
How to Create Your First Project
Step 1: Open Claude (web or desktop). In the left sidebar, click "Projects" then "New Project."
Step 2: Name it something specific — "Marketing Content Q3," "Backend API Refactor," or "Weekly Reports." Avoid generic names like "Work Stuff."
Step 3: Add your custom instructions. Start with these three elements:
Your role: "I'm a senior product manager at a B2B SaaS company. Our product is a project management tool for engineering teams."
Your preferences: "Write in a direct, professional tone. Use bullet points for lists. Keep explanations under 200 words unless I ask for more detail."
Your standards: "When writing customer-facing content, follow our brand voice: confident but not arrogant, technical but accessible. Never use 'leverage,' 'synergy,' or 'best-in-class.'"
Step 4: Upload relevant files. For a marketing project, upload your brand guidelines, product positioning doc, and 2-3 examples of content you liked. For a coding project, upload your API docs, data models, and coding standards.
Step 5: Start a conversation in the project. Ask Claude something related to the project. Notice how the response reflects your role, preferences, and uploaded context without you mentioning any of it.
What to Put in Your Project Files
The files you upload determine how useful the project is. Here's what works best for common use cases:
| Use Case | Files to Upload | Custom Instructions Focus |
|---|---|---|
| Writing | Style guide, brand voice, 3-5 examples, audience notes | Tone, structure, banned phrases, length limits |
| Coding | README, architecture docs, key models/schemas, conventions | Stack, patterns, what not to touch, test commands |
| Analysis | Data dictionary, prior reports, metric definitions | Default tables, executive summary format, assumptions |
| Meetings/Comms | Templates, stakeholder bios, recent notes | Voice, concision, action items, follow-up style |
Writing projects: Style guide, brand voice doc, 3-5 examples of published content you want Claude to match, target audience description, list of topics/keywords to cover.
Coding projects: README, architecture overview, key data models or schemas, coding conventions doc, example code files that show your style. This is similar to the CLAUDE.md approach in Claude Code.
Analysis projects: Data dictionary, previous reports or analyses, methodology descriptions, stakeholder preferences for report format.
Meeting/communication projects: Org chart, key stakeholder bios, communication templates, recent meeting notes for ongoing context.
The key principle: upload context that you'd normally paste at the beginning of every conversation. If you find yourself repeating the same background information, that information belongs in a project file.
📬 Getting value from this?
One actionable AI insight per week. Plus a free prompt pack when you subscribe.
Subscribe free →Custom Instructions That Actually Work
Bad custom instructions are vague: "Be helpful and professional." Good custom instructions are specific and constraining:
For a content writer: "You are helping me write blog posts for [Company]. Target audience: mid-level marketing managers at B2B companies. Posts are 1,500-2,000 words. Always include a specific example in every major section. Never use passive voice. End every post with a concrete next step, not a vague conclusion. Use 'you/your' throughout. Structure with H2 headings phrased as questions."
For a developer: "We use TypeScript, React with Next.js App Router, Tailwind CSS, and PostgreSQL with Prisma. Prefer functional components with hooks. Use named exports. Error handling: use try/catch with typed errors, never silent catches. Testing: write tests for every new function using Vitest. Don't suggest solutions that require new dependencies unless I ask."
For an analyst: "I analyze product usage data for [Company]. Our key metrics are MAU, WAU, feature adoption rate, and time-to-first-value. When I share data, start with the most surprising finding first. Use tables for comparisons, not paragraphs. Always suggest one thing we should investigate further. Assume I know SQL and statistics — don't explain basics."
Notice the pattern: role + specific constraints + format preferences + what NOT to do. The constraints and negatives are often more useful than the positives because they prevent the most common output problems. For a deeper dive on writing these instructions, see our guide to system prompts.
Projects vs Custom Instructions (ChatGPT)
ChatGPT has Custom Instructions too, but Projects are fundamentally different in three ways:
| Feature | Claude Projects | ChatGPT Custom Instructions |
|---|---|---|
| Files | Upload up to ~200K tokens of docs | No file container (text fields only) |
| Scope | Different context per project | Global across all chats |
| Workflow | Task-specific workspaces + grouped chats | Single personalization layer |
Projects hold files. ChatGPT's Custom Instructions are text-only — two small text fields. Claude Projects can hold up to 200K tokens of uploaded documents. This means Claude can reference your actual style guide, codebase docs, or product specs — not just a description of them.
Projects are task-specific. ChatGPT Custom Instructions apply globally to every conversation. Claude Projects let you create separate contexts for separate work. Your marketing project has different instructions than your coding project. Switching between them is one click.
Projects maintain conversation history. All conversations in a project are grouped together. You can reference previous threads or continue ongoing work. ChatGPT conversations are isolated — each starts fresh (unless you use memory, which is limited and imprecise).
If you're considering switching from ChatGPT, see our migration guide. Projects are one of the main reasons people make the switch.
Common Mistakes with Projects
Too many files. More context isn't always better. If you upload 50 files, Claude spends tokens reading context that may not be relevant to the current task. Keep each project focused — 5-10 files that are directly relevant.
Too-generic instructions. "Be helpful" tells Claude nothing. Specific constraints ("Under 200 words per response," "Always include code examples," "Never suggest a solution without explaining why") produce dramatically better output.
One project for everything. Create separate projects for separate work streams. A single "Work" project with marketing files and coding standards and meeting templates confuses the context. Three separate projects work much better.
Never updating. Projects are living workspaces. Update the files when your standards change, your product evolves, or your goals shift. Review project files monthly.
Claude Projects is context engineering made accessible. You don't need technical skills — just the discipline to set up your context once and maintain it over time. The payoff is instant: every conversation starts smarter.
Want to try structured prompting alongside Projects? The Prompt Optimizer applies the ICCSSE framework to any prompt — pair it with a well-configured project for the best results.
📬 Want more like this?
One actionable AI insight per week. Plus a free prompt pack when you subscribe.
Subscribe free →Frequently Asked Questions
Do I need Claude Pro for Projects?
Yes. Projects are a Claude Pro feature ($20/month). The free tier doesn't include Projects. If you're evaluating whether Pro is worth it, Projects alone justifies the cost for anyone who uses Claude regularly for work.
How many files can I upload to a project?
Up to 200K tokens total per project, which is roughly 150,000 words or about 500 pages of text. Individual files can be PDFs, text, code, or markdown. Images are supported too.
Can I share projects with my team?
Claude Team and Enterprise plans support shared projects. On Pro, projects are individual. You can export your instructions and share them manually with teammates who can set up their own projects.
What's the fastest way to improve output inside a Project?
Keep the Project focused (5-10 high-value files) and write specific instructions with constraints. If you want a shortcut for structure, run your prompt through the Prompt Optimizer first.
Disclosure: Some links in this article are affiliate links. We only recommend tools we've personally tested and use regularly. See our full disclosure policy.