Most people treat AI like a blank-slate conversation every time. You ask a question, Claude answers, session closes, memory gone. It's like briefing a new consultant from scratch on every call — you're paying for the same onboarding over and over.

There's a different way. If you're using Claude Code or Claude's Cowork mode, there's a file called CLAUDE.md that acts as persistent memory across all your sessions. You write it once, Claude reads it every single time, and suddenly you're not starting from zero anymore. You're working with someone who already knows the playbook.

What a CLAUDE.md File Actually Does

Think of it as the difference between having a new intern walk through your door every morning versus having a chief of staff who's been with you for two years. The chief of staff doesn't need you to explain your decision-making process, your writing style, or why you killed that idea last quarter. They know the context. They know what matters.

Your CLAUDE.md file holds exactly that kind of institutional knowledge. It's a markdown document where you tell Claude: who you are, what you're building, how you like to work, what your projects look like, what you've already decided. When I work with Claude now, I don't waste time in preamble. I just work.

Here's what that looks like in practice. Instead of:

That saved context isn't just about ego. It's operational. It means Claude doesn't waste your attention on generic advice. It means you can be more specific about what you need. It means collaboration is actually faster.

Where the File Lives and How to Create It

There are two places where CLAUDE.md can live, depending on your setup.

Global setup (persistent across all your projects): If you're using Claude in multiple projects, put CLAUDE.md in a hidden .claude folder at the root of your user directory. On Mac, that's ~/.claude/CLAUDE.md. On Windows, same structure — it persists across everything you do.

Project-specific setup: If you want different instructions for different projects (maybe one project needs a data analyst mindset, another needs a product builder), put CLAUDE.md in the project root. Claude checks the project first, then falls back to the global file.

Most people I know do a hybrid: global file with the fundamentals (who you are, how you write, what you care about), then project-specific files that layer on context (what you're building right now, what's already been decided, what you're testing).

What to Actually Put in It

Don't dump your entire life story in there. This isn't a diary. It's operational instructions. Here's the structure I use:

## Who I Am CMO-level marketing executive, 20+ years luxury brands (Treasury Wine Estates, Diageo). Currently building Forge (AI-powered social platform for men 18-25) and exploring consulting work. Building a LinkedIn audience around working with Claude and AI systems. ## Writing Style Direct executive tone with humor. No rhetorical-question stacking. Avoid choppy short sentences — vary rhythm between short and long. Clean formatting for sharing with teams. No AI-generated waffle. ## Project Status: Forge Five-layer architecture: oblique emotional assessment (Layer 0), CBT reframing (Layer 1), activity-based social skills training (Layer 2), experience marketplace (Layer 3), warm matching (Layer 4), post-game review (Layer 5). Current phase: Phase 1 interviews to validate oblique assessment approach and test education gap concern. What's been decided: Men-first positioning. Behavior-first vs. feeling-first assessment. Anti-dependency as ethical imperative, not user abandonment. ## Working Style Default mode is constructive critic, not cheerleader. Identify gaps and weak logic first. Only praise work that genuinely earns it. I want to know what's wrong before I hear what's right.

See the pattern? Each section is specific enough to actually be useful, but short enough to stay relevant. You're not writing a novel. You're leaving sticky notes for your AI collaborator.

What NOT to Put in It

Don't include sensitive financial information, API keys, passwords, or personal data you wouldn't want in a document that persists across sessions. CLAUDE.md is useful but it's not encrypted. Keep it operational, not confidential.

Don't write filler. "I like good ideas" is not useful. "I want design recommendations that prioritize accessibility and test against WCAG 2.1 AA standards" is useful. The specificity is what makes the file work.

Don't assume Claude will remember everything if it's not in the file. The file is your backup. If it's critical context, put it in the file.

How to Maintain It as Your Work Evolves

Your CLAUDE.md isn't static. It's a living document. When you finish a phase of work, update it. When you make a big decision, add it. When your writing preferences shift or a project ends, edit it.

I review mine quarterly and add a "Last updated" note. It becomes a record of what you've decided and why, which is weirdly valuable even outside of AI collaboration. It forces clarity. I've caught my own inconsistencies just by writing them down.

Update the file when:

Why This Matters for Your Work

The real power of CLAUDE.md isn't that it saves you three minutes of explanation. It's that it turns Claude from a generic assistant into something closer to a collaborator who knows your work. Someone you can be blunt with because they understand what you're actually trying to do. Someone who doesn't waste your time with irrelevant options because they know what you've already ruled out.

In marketing, we talk about the efficiency of operating with aligned teams — everyone knows the strategy, everyone knows what success looks like, so execution is faster and tighter. CLAUDE.md is that alignment, but between you and the AI you're working with.

If you're serious about using AI as a real tool (not just a novelty), stop treating every session like a fresh conversation. Set up the file. Make it specific. Update it. Then work.