Paige Processing

One of the most useful things you can do with Claude is set up global instructions that apply to every conversation. These are like a standing briefing that tells Claude who you are, how you work, and what you expect before you even say hello.

In Cowork, these live in a file called CLAUDE.md in your project folder. In Claude Chat (the web or desktop app), you can set them under your profile settings. Either way, the idea is the same: tell Claude how to be useful to you specifically.

What to Include

Who you are and what you do. Not a full bio, just enough context that Claude doesn’t have to guess. “I’m a nonprofit treasurer and community organizer” gives Claude very different context than “I’m a software engineer.”

How you communicate. Do you prefer bullet points or prose? Short and punchy or thorough and detailed? Do you hate corporate jargon? Say so. Claude will mirror whatever voice you describe.

Your working preferences. Default file formats, whether you want clarifying questions before Claude starts a task, how long responses should be. The more specific you are, the less time you spend correcting things later.

Honesty over validation. This one matters. AI defaults to being agreeable, which means it will cheerfully tell you every idea is great even when it isn’t. You can (and should) override this. Tell Claude to push back when something could be better, skip the “love that!” filler, and trust that you can handle direct feedback. You’ll get dramatically better output when the AI isn’t just telling you what you want to hear.

What to do when things go wrong. Do you want Claude to apologize when it makes mistakes, or just fix them? (I prefer “just fix it.”) Do you want it to explain what went wrong, or move on? Set this expectation once and you won’t have to manage it in every conversation.

Example Global Instructions

My name is [Name]. I'm a [roles/context]. I live in [location].
How I communicate: [Your style. Conversational? Formal? Parenthetical
asides? No em dashes? Whatever makes your writing yours.]
How I want Claude to work with me:
- Always ask clarifying questions before starting non-trivial tasks
- Default to [file format] unless I say otherwise
- Match output length to the task
- Write in my voice when drafting communications
- Back up claims with sourced data
- If I say something's off, just fix it, no need to over-apologize
- Be honest with me, not sycophantic. Don't automatically validate
  every idea. If something could be better, say so directly. Push
  back when you see a problem and suggest a better approach.
Typical tasks: [What you usually ask Claude to help with]

Why This Matters

Without global instructions, every conversation starts from zero. Claude doesn’t know your preferences, your voice, or your tolerance for fluff. You end up spending the first few exchanges of every session re-establishing things you’ve said a hundred times before.

With global instructions, Claude shows up already knowing how to work with you. It’s the difference between training a new intern every morning and working with someone who already knows the routine.