You open Claude. You type a prompt. You get a response that’s technically correct and completely useless for your actual situation.
That’s not a Claude problem. That’s a setup problem.
The fix is eight files. These are simple text documents you set up once that give Claude permanent context about who you are, how you write, and how you work. Every conversation starts with Claude already knowing your situation, your voice, and your rules. This post breaks down exactly what each file is and what to put in it.
One infographic from Ruben Hassid’s How to AI guide laid out the full system. This is the plain-English breakdown of it.
What Are Claude Setup Files and Why Do They Work?
Claude setup files are simple text documents (mostly Markdown) that you either upload to a Claude Project or reference through Claude’s settings. They give Claude persistent context it can draw on across every conversation.
Without them, Claude is a very capable stranger. With them, it’s closer to a collaborator who already knows your situation.
The eight files below cover three layers: who you are, how you work, and what tools you use. Each one is lightweight to set up and compounds over time as you refine it.
The 4 Identity and Voice Files
These four files are the foundation. They tell Claude who it’s actually working with.
1. about-me.md: Your Identity
This is the file Claude reads before anything else. It covers your background, your job, your current priorities, and the decisions you’ve already made that Claude shouldn’t second-guess.
What goes in it: your role and industry, what you’re currently working on, your goals for the next quarter, and any standing decisions Claude should know about. The more specific you are, the better Claude can tailor its output without you having to re-explain yourself every session.
2. voice-profile.md: Your Taste DNA
This is one of the most underrated files in the system. It captures how you actually think and write: not just your style preferences, but your aesthetic sensibility, your contrarian takes, and your hard nos.
What goes in it: your writing mechanics (sentence length, rhythm, structural preferences), your beliefs about your industry that most people disagree with, what your writing looks like when it’s at its best, and what you’d never say or write. The “what I’d never say” section is especially important. It’s faster to define the edges than to describe the center.
3. anti-ai-writing-style.md: Your Boundaries
This file is specifically about what Claude should never do when writing as you or for you. It’s a list of words, structures, tones, and formats you reject.
What goes in it: banned words (synergy, leverage, utilize, delve), structural patterns you hate (long intros, numbered summaries at the end, “In conclusion”), tones that feel wrong (overly enthusiastic, corporate, hedging). This file does more to clean up AI output than almost anything else. Claude is trained on the entire internet: which means it defaults to the internet’s average. This file pulls it away from that average and toward your actual voice.
4. The Cowork Folder: Your 4-Folder System
This isn’t a single file: it’s a folder structure that Claude navigates. The four folders are:
- ABOUT ME/: your identity and writing rules (points to about-me.md and voice-profile.md)
- PROJECTS/: one subfolder per project with briefs, drafts, and references
- TEMPLATES/: finished work Claude can use as patterns for future output
- CLAUDE OUTPUTS/: the only folder where Claude delivers finished work
The discipline of keeping outputs in one place sounds minor until you realize how much time people waste hunting through conversations for things Claude produced three sessions ago. CLAUDE OUTPUTS/ is the single source of truth.
The 2 Instruction Files
5. Global Instructions: Your Persistent Rules
This is the most powerful file in the system, and the easiest to set up. Global Instructions run automatically before every single conversation: you set them once and never touch them again unless something changes.
Where to set it: Settings → Cowork → Edit Global Instructions.
What to write in it: Always read ABOUT ME/ before starting. Always read the matching PROJECTS/ subfolder. Study TEMPLATES/ before creating anything new. Only deliver work in CLAUDE OUTPUTS/. Include naming conventions, formatting defaults, and any standing operating procedures.
The compound effect here is significant. Every conversation starts with Claude already oriented to your system: no re-explaining, no reminders, no drift back to generic defaults.
6. The One Prompt: How You Start Every Chat
This is the single prompt that kicks off 80% of your conversations. It forces Claude to ask you the right questions before assuming anything and executing wrong.
The structure: “I want to [TASK] so that [SUCCESS CRITERIA]. First, read my folder. Then ask me questions using AskUserQuestion. Refine the approach with me before you execute.”
Why this matters: Claude generates a clickable form with the questions. You answer them. Claude shows a plan. You approve it. Then it executes. If something’s off, you catch it before the work is done: not after. This one prompt alone eliminates most of the frustrating back-and-forth that makes people give up on AI tools.
The 2 Tool Files
7. Connectors: Claude Inside Your Tools
Connectors plug Claude directly into the apps you already use. No copy-pasting, no screenshots. Claude operates inside your actual workflow.
What you can connect: Slack (read channels, search messages), Google Drive (pull from your Docs), Notion (reference your pages), Gmail, Calendar, Figma, and 50+ others.
How to set up: Settings → Connectors → Browse → Add.
The practical difference this makes: instead of copying content from Slack into Claude and back, Claude reads your Slack directly. Instead of describing your Google Doc, Claude reads it. Context retrieval goes from manual to automatic.
8. Plugins: Instant Skill Packs
Plugins are pre-built workflows that turn Claude into a specialist for a specific job. One-click install. Each plugin comes with its own slash commands.
What’s available out of the box: Marketing (draft-content, voice-matched posts), Data (CSV exploration, dashboard building), Sales (account research, outreach drafting).
How to access: Cowork → Customize → Browse Plugins → Install.
The difference between a plugin and a prompt is that a plugin is a complete workflow someone else has already tested and optimized. You don’t have to figure out the right sequence of instructions: you just run the command.
How to Set These Up Without Losing an Afternoon
The temptation is to build all 8 files perfectly before you start. Don’t.
Start with about-me.md and anti-ai-writing-style.md. These two files alone will improve your output quality more than anything else in the system. Write them in 20 minutes: rough is fine. You’ll refine them as you use Claude more.
Add Global Instructions next. This takes 10 minutes and pays back immediately.
Build out the Cowork folder structure over your first week of real use, not before. Set up connectors as you need them: start with the one or two tools you use most, not all 50 at once.
The system compounds. The longer you use it, the better the output gets: because your files improve and Claude’s operating context sharpens with each session.
FAQs: How to Make Claude Actually Useful
Do I need Claude’s paid plan to use these files?
Most of these features: Projects, Global Instructions, Connectors, and Plugins: are available on Claude’s Pro plan and above. The free tier has limited project and memory functionality. If you’re using Claude seriously for work, the Pro plan is where the system starts to pay off.
Can I use these files across different Claude conversations?
Yes: that’s the point. Files stored in a Project carry context across every conversation within that project. Global Instructions apply to every new chat automatically. You’re not tied to a single thread.
What if I don’t know what to write in my voice-profile.md?
Start with the negatives. List 10 things you’d never write and 10 words you’d never use. That’s more useful to Claude than a positive description of your style. You can always add the positive layer later.
Is there a naming convention for the output files?
Yes, and it’s worth using from day one: project_content_type_v1.ext: for example, rbo_blogpost_writingvoice_v1.md. It makes version tracking and file retrieval much easier, especially if you’re running multiple projects.
The Takeaway: How to Make Claude Actually Useful
Claude isn’t magic out of the box. It’s a system that gets dramatically more useful the more context it has. These 8 files are how you give it that context: permanently, without repeating yourself every session.
Set up about-me.md and anti-ai-writing-style.md today. Add Global Instructions. Start building the Cowork folder. You’ll feel the difference within a week.
For the bigger picture on using AI as a real person with a real job: not a developer, not a tech enthusiast: the full guide is here: AI for Real People: How Solopreneurs and Solo Creators Can Actually Use It (Without the Overwhelm)