Claude Projects: How to Use Them for Persistent Context and Organized Work

ClaudeHow-To GuidesUpdated May 16, 2026
Quick Answer

Claude Projects give you a persistent workspace where uploaded documents and instructions stay available across every conversation in that project. Create one from the sidebar, upload relevant files, add custom instructions, then start conversations — Claude always has your context without you re-pasting it. Projects require Claude Pro ($20/month).

What Claude Projects Are For

Projects solve a specific pain point: repeating yourself in every Claude conversation.

If you regularly work with the same codebase, the same client, the same research topic, or the same writing style, you usually paste the same background context at the start of every session. Projects eliminate this. You upload that context once, and Claude has it available in every conversation within the project — automatically, without prompting.

Common project setups:

  • Client work — upload briefs, brand guidelines, previous outputs; Claude always knows the client context
  • Coding projects — upload key files, architecture docs, coding standards; Claude reviews code in context
  • Research — upload papers, notes, source material; Claude analyzes without you re-supplying sources
  • Writing — upload style guides, tone examples, audience notes; Claude writes consistently

How to Create a Project

  1. In the Claude sidebar, click New Project (or the folder/plus icon)
  2. Give the project a name
  3. Click into the project to open it
  4. Upload files using the Add content button or by dragging files in
  5. Write project instructions in the instructions field — these appear at the start of every conversation in this project
  6. Click New Conversation to start working

What to Put in Project Instructions

Project instructions are a system prompt that runs at the start of every conversation. Think of it as a standing briefing for Claude.

Template:

Role: You are helping me with [project type].
Context: [Brief description of the project, client, or goal]
Style: [Tone, format, length preferences]
Always: [Rules that always apply]
Never: [Things to avoid]

Example — Client project:

You are helping me with content for Acme Corp, a B2B SaaS company selling HR software to mid-market US businesses.
Audience: HR managers and CHROs, not technical readers.
Tone: Professional but approachable. No jargon.
Always: Reference uploaded brand guidelines when writing.
Never: Use competitor names. Never make claims about pricing or specific features not in the uploaded docs.

Example — Coding project:

You are a coding assistant for my FastAPI + PostgreSQL backend project.
Stack: Python 3.12, FastAPI, SQLAlchemy 2.0, Alembic, Pydantic v2.
Style: Use type hints everywhere. Write unit tests with pytest. Follow the uploaded coding standards.
When I share code, point out bugs and security issues first, then suggest improvements.

Best Practices for Project Files

Upload only what Claude needs to reference. Every byte of uploaded content takes up context window space. Uploading a 200-page PDF when Claude only needs the executive summary wastes space that could go toward the conversation.

Create focused summary documents instead of raw sources. If you have a 50-page research report, write a 2-page summary of the key findings and upload that. Claude can work from a good summary more reliably than from a dense source document.

Organize by what Claude will actually reference. Ask yourself: "Will Claude need this fact in most conversations in this project?" If yes, put it in the project. If it only applies to one task, paste it into that specific conversation.

Update files when context changes. If your codebase changes significantly or a client's brief is updated, replace the old files with new versions. Outdated project files will cause Claude to give outdated answers.

Organizing Multiple Projects

Create one project per distinct workstream:

  • One project per client (not one project for all clients)
  • One project per codebase or application
  • One project per research area
  • One project for personal use vs. work use

Name them clearly. "Marketing — Acme Corp" is better than "Project 3." The sidebar displays project names and recent conversations — clean naming makes it easy to jump back in.

Common Mistakes to Avoid

  • Uploading everything — More files is not better. A project with 20 uploaded documents will have Claude struggling to prioritize which content matters. 3–5 focused files outperform a dump of everything you have.
  • Skipping the instructions — Project files without instructions leave Claude to guess how to use them. Always write at least 3–5 sentences explaining the project context and how you want Claude to behave.
  • One giant project for all work — Mixing unrelated projects into one makes the instructions and files irrelevant to any given task. Separate projects stay focused.
  • Not starting new conversations when threads get long — A very long conversation within a project fills up the context window, leaving less space for your project files. Start fresh conversations within the project when sessions exceed 50–60 back-and-forth exchanges.
  • Expecting Projects to replace your notes — Projects are for context Claude needs, not for your own reference. Keep your own documentation; treat project files as Claude's briefing, not your filing system.

Related Articles

Claude · How-To Guides

More Claude how-to guides guides

Browse all guides in this category to troubleshoot related issues faster.

Browse all guides →

Frequently Asked Questions

In a regular conversation, Claude only knows what you have written in that session. When the conversation ends, the context is gone. In a Project, you upload documents and write instructions that Claude reads at the start of every conversation in that project — it is like giving Claude a permanent briefing. This is useful when you have recurring work with the same context: a client account, a codebase, a research area, or a writing style guide.

Related Guides

Continue with nearby guides in the same topic to rule out adjacent causes faster.

Claude Artifacts: What They Are and How to Use Them

Claude Artifacts are self-contained outputs — code, documents, HTML pages, or data — displayed in a separate panel next to the chat. When Claude creates an artifact, you can view, copy, download, or directly edit it without scrolling through the conversation. Artifacts appear automatically when Claude generates code, structured documents, or interactive content. No setup required.

Can't Log In to Claude: Every Fix for Every Login Error

If you cannot log in to Claude, the most common causes are using the wrong sign-in method (Google vs. Apple vs. email), a corrupted browser session, or a temporary Anthropic authentication issue. Start by clearing cookies for claude.ai, trying incognito mode, and using the exact sign-in method you used when you registered. Verification code issues are usually email delivery delays — check your spam folder and wait 5 minutes before requesting a new code.

How to Use Claude: Complete Beginner's Guide

To use Claude, go to claude.ai, create a free account with Google, Apple, or email, and start typing in the chat box. Claude responds immediately and handles writing, analysis, coding, research, and conversation well. The free plan uses Claude Haiku; Claude Pro ($20/month) unlocks the more capable Claude Sonnet and Opus models with higher usage limits. No setup required.

Claude Payment Failed: Why It Happens and How to Fix It Fast

Claude payment failures usually come from one of four causes: your card issuer blocking the charge, an expired or incorrect card on file, a billing address mismatch, or a temporary issue on Anthropic's payment processor. Start by updating your payment method in Settings → Billing, then retry. If that fails, call your bank to authorize the charge before trying again.

Claude Prompt Tips: How to Get Better Results Every Time

Claude follows detailed instructions well, so specificity is the biggest lever for quality. Tell Claude your role, the audience, the desired format, and what to avoid. Use the full context window — Claude handles long prompts better than most models. Ask for multiple versions, give iterative feedback, and specify constraints upfront rather than correcting after. The prompt is your only interface; writing it well is the skill.

Claude Something Went Wrong: Causes and Fixes

The 'something went wrong' error in Claude is a generic message that covers session issues, browser problems, and temporary server errors. Start by refreshing the page and signing out then back in. If the error persists, clear your browser cookies for claude.ai and try a private window. This error almost never indicates an account problem — it is usually a temporary connection or session issue.