Re-briefing Claude from scratch every conversation is like re-onboarding the same employee every Monday morning. Projects, CLAUDE.md, and Skills solve this.
The Problem
Without persistent context, every Claude conversation starts at zero. You end up:
- Pasting the same background information over and over
- Getting generic outputs that don't match your voice or standards
- Spending the first 5 minutes of every session catching Claude up
- Settling for "good enough" instead of "exactly right"
This is the single biggest time waster for Claude users. And it's completely fixable.
Claude Chat: Projects
Projects are the most important feature in Claude Chat that most people don't use.
What a Project is: A container that holds instructions and files that Claude references in every conversation within that Project.
Think of it this way: if Claude is a brilliant new hire, a Project is their onboarding packet. Everything they need to know to do great work, loaded automatically before every conversation.
What to Put in Project Instructions
You are the AI assistant for [Name], Marketing Director at [Company].\n\nCompany context:\n- B2B SaaS, $8M ARR, 200 customers\n- ICP: Mid-market e-commerce (50-500 employees)\n- Main competitors: [Competitor A], [Competitor B]\n\nWriting style:\n- Professional but conversational\n- Data-driven — always cite numbers when available\n- Short paragraphs, clear headers\n- No jargon: 'use' not 'leverage', 'help' not 'empower'\n\nKey context:\n- Currently running a product launch campaign (Q2)\n- Budget: $50K/quarter for paid acquisition\n- Team: 3 people (content, demand gen, product marketing)\n\nDefault format preferences:\n- Briefs: Start with objective and key message\n- Emails: Under 150 words, clear CTA\n- Reports: Executive summary first, then details
What to Upload as Project Files
Project files are reference documents Claude can access in every conversation. Upload things that don't change often:
- Brand/style guide — How you write, what words you use and avoid
- Product documentation — What you sell, features, pricing
- ICP or persona docs — Who your customers are
- Competitive intel — How you position against competitors
- Templates — Proposal structures, email formats, report layouts
Pro Tip
You can also upload files to individual conversations (not just Projects). Use Project files for things Claude should always know, and conversation uploads for one-time reference materials.
Setting Up Your First Project
Here's the fastest path to a useful Project:
5-minute Project setup
Create a new Project in Claude Chat
Click 'Projects' in the sidebar, then 'New Project.' Name it something clear — 'Marketing Work' or 'Daily Ops.'
Write your core instructions
Start with 4 things: your role, your company (2-3 sentences), your writing style (3-5 bullets), and one format preference.
Upload 1-2 key files
Your brand guide and your product one-pager are the highest-impact uploads. Don't overdo it on day one.
Test with a real task
Open a conversation in the Project and try a task you've done before. Compare the output to what you got without Project context.
Iterate
Add instructions based on what's missing. If Claude keeps using jargon, add 'never use jargon.' If it writes too long, add 'default to concise.' Build the instructions over time.
Multiple Projects for Different Roles
Most operators benefit from 2-4 Projects:
| Project | What It Contains | When You Use It |
|---|---|---|
| Marketing & Content | Brand voice, content strategy, audience docs | Writing content, planning campaigns |
| Sales & Deals | Pricing, objection library, proposal template | Drafting proposals, prepping for calls |
| Operations | Team structure, processes, tool stack | Planning, process design, reporting |
| Strategic Thinking | Business context, competitive landscape, goals | Decision-making, brainstorming, analysis |
Scenario
You use Claude for both marketing content and sales proposals. Your marketing voice is casual and educational, but your sales proposals are more formal and ROI-focused.
Claude Code: CLAUDE.md
CLAUDE.md is Claude Code's version of Project instructions — but more powerful because it lives in your file system and can be version-controlled.
It's a simple markdown file at the root of your project folder. Every time you start a Claude Code session in that folder, it reads CLAUDE.md first.
Real example
“The gap between default Claude Code and configured Claude Code is wild. Context setup and rules files were doing 90% of the heavy lifting.”
— @bpizzacalla
Built a full agentic sales platform with Claude Code
Think of CLAUDE.md as an onboarding document for a brilliant new hire. You write it completely differently when you think of it that way.
What to Include in CLAUDE.md
- Business context — What your company does, who your customers are, your stage
- Your role — How you use Claude Code, what decisions you make
- Repository structure — What's where, so Claude can find things
- Voice and style guide — How outputs should sound
- Key tools and integrations — APIs you connect to, tools you use
- Standard processes — Workflows you repeat
- Important files — Where key documents live
- Rules — Things to always or never do
We go deep on CLAUDE.md in the CLAUDE.md: Your Business Operating System playbook.
Claude CoWork: Shared Context
CoWork lets teams share context across conversations and members. Think of it as Projects, but for your whole team.
When to use CoWork over Chat Projects:
- Multiple team members need the same context
- You want Claude to pull from shared team knowledge
- You need integrations with team tools (Slack, Google Drive, Notion, etc.)
- You want consistency across how different team members get AI output
CoWork Context Best Practices
The same principles apply — give Claude persistent context — but with a team lens:
- Shared knowledge base — Company info, processes, standards that everyone needs
- Role-specific instructions — Different team members can have different default behaviors
- Connected tools — Link to the data sources your team actually uses
- Living documents — Context that updates as your business evolves
Skills: Reusable Instruction Packages
Skills take persistent context one step further. Instead of general project instructions, Skills are specific instruction packages for specific tasks.
The difference between Project instructions and Skills:
| Project Instructions | Skills | |
|---|---|---|
| Scope | Everything in this Project | One specific task type |
| Loading | Always loaded | Loaded when relevant |
| Example | "We're a B2B SaaS company..." | "When writing proposals, follow this template..." |
Practical Skill Examples
Proposal Writer Skill: Knows your pricing tiers, your standard proposal structure, your terms, and your closing style. Loaded when you ask Claude to write a proposal.
Meeting Prep Skill: Knows your meeting prep format, pulls from your CRM context, generates agenda + talking points + follow-up template. Loaded when you say "prep me for a meeting."
Weekly Report Skill: Knows your report format, your KPI definitions, where to find the data, and who the audience is. Loaded when you ask for your weekly report.
How Skills Work Across Products
- Chat: Upload Skills as files or define them in Project instructions with clear triggers
- CoWork: Available to the team, can be triggered by events
- Code: Drop a markdown file in your
.claude/commands/directory — it becomes a slash command
Note
Skills are covered in depth in the Skills Deep-Dive section. For now, just know they exist and that they're how you scale from "Claude does one-off tasks" to "Claude runs repeatable workflows."
The Context Hierarchy
Here's how all of this fits together, from broadest to most specific:
| Layer | Scope | Persistence | Best For |
|---|---|---|---|
| CLAUDE.md (Code) | All Code sessions in folder | Permanent (file-based) | Full business context for automation |
| Project instructions (Chat) | All chats in that Project | Permanent | Company/role context for conversations |
| Skills | Loaded when relevant | Permanent | Task-specific workflows and templates |
| Uploaded files | Project or conversation | Depends on where uploaded | Reference documents and data |
| Conversation messages | Single chat thread | Session only | Specific task instructions and refinements |
The key insight: Push as much context as possible into the higher layers. The more Claude knows before you type your first message, the less work you do in every conversation.
Memory: What Claude Remembers Between Conversations
Claude also has a memory feature that saves facts across conversations — things like "I prefer bullet points over paragraphs" or "Our fiscal year starts in April."
Memory is useful but limited. Think of it as Post-it notes on Claude's desk, not a full onboarding document. For serious context, use Projects or CLAUDE.md.
What memory is good for:
- Personal preferences ("always use Oxford commas")
- Quick facts ("our company was founded in 2018")
- Corrections ("when I say 'the platform,' I mean Shopify")
What memory is not good for:
- Complex context (use Project instructions)
- Documents and data (use file uploads)
- Task-specific instructions (use Skills)
Your Action Step
Right now, set up persistent context for your product:
Create a Project with your core business context. Start with your role, company, writing style, and one format preference. Upload your brand guide if you have one.
- Click "Projects" in the sidebar, then "New Project"
- Name it something clear — "Marketing Work" or "Daily Ops"
- Write 4 things in the instructions: your role, your company (2-3 sentences), your writing style (3-5 bullets), and one format preference
- Upload 1-2 key files — your brand guide and product one-pager are highest impact
- Test with a real task you've done before and compare the output
You can always add more later. The point is to stop starting from zero.
Quick check
What's the most impactful thing you can do to improve Claude's output quality?