The fastest way to make Claude useful is to stop typing the same context into every chat. The way to do that is a folder structure on your computer that Claude reads automatically — three folders, three files, one paragraph of global instructions. I borrowed mine from Reuben Hassid (LinkedIn's most-watched Claude voice), tweaked it for the way I actually work, and stopped re-explaining myself five times a day.
Last month I had four Cowork sessions go badly in one week. Same pattern every time. I'd ask Claude for a draft. It came back generic. I'd say "no, more like this." It corrected one thing and missed another. I'd say "more like THIS." It tried too hard. By the third revision I'd write the thing myself and tell my husband I was thinking of going back to plain Notes. Which is what people who quit AI tools say.
The problem wasn't Claude. The problem was that I'd given it nothing to work with except whatever I happened to type that morning. Claude can't read your mind, your past work, or what you'd never say. If you don't tell it, it makes assumptions. The assumptions are usually wrong, and the wrongness compounds.
Cowork stops feeling like a chat and starts feeling like a coworker the moment Claude has access to who you are, what you reject, and where to put things. Three folders, three files, no exceptions.
What Cowork actually is
Quick context if you're not already there. Cowork is the desktop app's power mode. Projects become departments, artifacts become deliverables, and Claude can read files on your actual computer instead of only what you paste into the chat window. You install the desktop app at claude.com/download, select Cowork from the mode dropdown, and point it at a folder.
That folder is the whole game.
The structure (Reuben Hassid's)
I've played around with a handful of Cowork setups. Reuben Hassid's is the best I've found, and I've looked. He knows his stuff. The folder structure below is his, almost unchanged. I'm walking through it because it's the one I run now. Credit where it's due.
Make a folder on your computer called Claude Cowork. Inside it, three subfolders:
Claude Cowork/
├── ABOUT ME/
├── OUTPUTS/
└── TEMPLATES/That's the whole structure. The work is in the files inside ABOUT ME/.
ABOUT ME — the three files Claude reads before every task
Three Markdown (.md) files go in here. Together they're under 5,000 words. Claude reads them at the start of every session and refers back constantly. This is the file set that turns Cowork from a chatbot into something that knows how you work.
about-me.md
Who you are, how you think, what your day actually looks like.
Don't make this a LinkedIn bio. Make it useful. Sections to include: what you actually do for a living, who your clients or audience are, how you make decisions, your quality standard ("nothing ships without X"), your pet peeves ("I will not say 'leverage'"), and the rules Claude has to follow with you ("ask before deleting anything," "always show me the plan before the work").
Target length: under 2,000 tokens. About 1,500 words.
anti-ai-writing-style.md
The list of words, phrases, and patterns you'd never say.
This is the one that changes everything. Hassid bans about 80 words from his AI writing — words like "delve," "harness," "tapestry," "synergy." He limits paragraphs to three sentences. He bans rhetorical questions and the word "elevate."
This file should be 80% what you reject and 20% what you keep. It works because Claude knows exactly which patterns to avoid before it writes a single word, instead of you correcting them after.
If you've already done VoiceMark work or anything similar, this file takes ten minutes. Paste in your voice rules. If you haven't, the fastest way to build it is to read three pieces of generic AI-written content and write down every phrase that made you wince.
my-company.md
Strategic context Claude needs to make business-correct decisions.
What you're trying to grow. What you've decided NOT to grow. The platforms that matter and the ones you've stopped caring about. Your top three goals for the quarter with specific numbers. Anything you're actively saying no to.
This is the file that stops Claude from pitching you a TikTok strategy when you've explicitly said you don't do short-form video.
Target: under 1,000 tokens. About 700 words. Keep it scannable.
OUTPUTS — where Claude writes (and only Claude writes)
Make subfolders inside OUTPUTS/ named after the projects you're working on. When Claude finishes a deliverable, it saves to the matching subfolder. You always know where to find what Claude made you, and Claude doesn't have to ask you where to put things.
The reason this works: Cowork doesn't pull OUTPUTS/ into context unless you specifically point at a file. So your folder can grow to a thousand drafts without slowing anything down or eating your token budget.
TEMPLATES — your best work, ready to be copied
When something Claude makes you turns out really good — a newsletter that hit, a sales page that converted, a proposal that closed — say "save this as a template in TEMPLATES/." Cowork drops it in the folder. Next time you start a similar project, you point Claude at the template and say "make me the next version of this."
This is the part most people skip and it's the highest-impact piece of the whole system. Your fifth newsletter doesn't have to start from scratch. It starts from your best one.
The Global Instructions (the part nobody installs)
Settings → Cowork → Edit Global Instructions. Paste this in:
I usually start my Cowork session by pointing you to my Cowork folder. Before any and every single task, you must read every file inABOUT ME/:about-me.md,anti-ai-writing-style.md,my-company.md. Never readOUTPUTS/orTEMPLATES/unless I specifically point you to a file. Save all deliverables inOUTPUTS/under a subfolder named after the project. If the brief is unclear, ask me a clarifying question before you start writing.
This is the prompt that ties the whole structure together. Without it, Claude won't read your files automatically. With it, you stop having to say "remember, I don't write like that" five times a session.
What I'd do if you're starting today
Skip the trap of trying to make all three files perfect. The trap looks like spending your whole afternoon writing about-me.md and never actually using Cowork.
What works: 30 minutes per file, in this order.
anti-ai-writing-style.mdfirst. Fastest to write, biggest impact. Open a doc and list every AI phrase you hate. 30 minutes. Done.about-me.mdsecond. Paste in the LinkedIn bio you already have, then add three specific things: your actual day, your decision-making rules, and your pet peeves.my-company.mdlast. Your goals for the next 90 days and what you're saying no to.
Then put the global instructions in. Open Cowork. Tell it "this is my Cowork folder" and ask it to do a real piece of work — not a test prompt. The first 10 minutes will tell you whether the files are tuned right. They won't be. You'll edit them. That editing IS the setup.
The thing nobody tells you
Cowork is only as good as your anti-ai-writing-style.md file. Everything else is logistics. The voice file is the work. If your AI output still sounds like a chatbot after you've installed all this, that file isn't specific enough yet.
The folder structure is the easy part. The voice file is the work. Everything Claude writes for you is downstream of how clearly you've told it what you would never say.
If you want help building the voice file specifically — the one that catches the way you actually sound and the patterns you'd never use — that's what VoiceMark is built for. It's a faster path to the version of anti-ai-writing-style.md that actually changes your output.
The structure itself? Three folders, three files, one paragraph of global instructions. You can have it running by lunch.
Frequently asked questions
Do I need Claude Cowork specifically, or can I do this in Claude Chat?
You need Cowork or another mode that can read files from your computer (Claude Code also works). Claude Chat only reads what you paste into the conversation. Cowork reads files automatically every session, which is the entire point of the folder system.
What if I don't have time to write all three files?
Write one: anti-ai-writing-style.md. Paste your voice rules and a list of phrases you hate. That single file fixes about 60% of the "Claude doesn't sound like me" problem. The other two files are real but they're optimization, not foundation.
What I'd actually do this week
Open the voice memo app on your phone. Three minutes, no editing: "When AI writes for me and it sounds wrong, here's what it's doing." Don't filter. Just talk.
Transcribe it. The list you just made is your anti-ai-writing-style.md first draft. Put it in a folder called Claude Cowork/ABOUT ME/. Open Cowork. Paste in the global instructions above. Try one task.
You'll know within 10 minutes whether it's working. If it is, write the other two files this weekend. If it isn't, the voice file needs more specifics. That's almost always the answer.