Global and folder instructions

If you've typed "use a professional tone" or "I'm a marketing manager" at the start of every Claude conversation, Claude global and folder instructions are the fix — set them once and Claude carries that context into every session automatically.

Claude global and folder instructions are persistent preference settings that eliminate the need to repeat your role, tone, and format preferences in every new conversation. Global instructions apply to all Claude sessions; folder instructions apply only to a specific project folder. Together, they create a three-layer context stack that turns Claude from a generic assistant that resets after every chat into one that already knows your world before you type the first word.

The Problem: Claude Resets Every Session Without Instructions

Here's the honest reality of how most people use Claude: they're good at asking questions, but they front-load every session with the same context. Role. Industry. Tone. Format. Then the chat ends, and the next one starts from zero again.

Say you're a marketing manager at a B2B SaaS company. Every session you want professional but conversational tone, metric-driven language, executive summary first, data formatted as tables, and client context included. Right now you add all of that manually — every time. That's not a prompt problem. That's a setup problem, and it compounds across hundreds of conversations. It's the equivalent of having a brilliant assistant with amnesia: smart, but starting fresh every single session.

Three Levels of Instructions That Stack

Think of your Claude setup as three concentric circles of context, each more specific than the last:

• Global instructions — who you are, your tone preferences, your format rules; applies to every single Claude session automatically
• Folder instructions — project-specific context tied to a specific folder Claude has access to
• Session-level requests — what you need right now, in this specific conversation

When all three stack together, Claude stops being a generic assistant and starts operating like a colleague who already knows your background, your client's name, and your preferred output format before you ask anything. That's the architecture behind the payoff.

How to Set Up Claude Global Instructions Step by Step

This is a one-time setup. Go to Settings → Co-work → Edit Global Instructions. You'll see a text box labelled "apply to all Co-work sessions." Whatever you add there applies to every Claude session going forward — no re-pasting, no re-explaining.

Three categories belong in global instructions:

• Role and background. Something like: "I'm a marketing director at a B2B SaaS company focused on enterprise security solutions. I've been in this industry for eight years." This tells Claude your expertise level and industry so it calibrates every response accordingly — whether you're analyzing content strategy or reviewing competitive positioning.
• Tone and format preferences. "Always use a professional but conversational tone — think smart colleague, not corporate robot. Format reports with an executive summary first. Use metric-driven language and avoid fluff." Claude learns to match your voice, not just your topic.
• Always-include and always-exclude rules. "Always include competitor context when relevant. Never use marketing jargon without defining it first. When suggesting campaigns, include budget estimates." These are your guardrails — they shape every response without you manually re-enforcing them each session.

Folder Instructions: Project Context Without Cluttering Global

Global instructions handle who you are. Folder instructions handle the work in front of you — and they stay separate so your global setup stays clean.

When you grant Claude access to a specific folder — say, your client projects folder — you can attach instructions that only apply in that context. Something like: "This folder contains deliverables for Acme Group. Use our standard template format. Client contact is Sara, VP of Operations. Always request approval before sharing externally."

Now Claude knows your global preferences, the project-specific rules for this client, and whatever you ask in the current session. Three layers. Each conversation starts from a far higher baseline than any single prompt could achieve on its own. And the separation matters: Acme Group's rules stay in Acme Group's folder. Your global setup never gets cluttered with client-specific context that doesn't apply everywhere else.

What a Real Three-Layer Stack Looks Like in Practice

Here's what this actually produces. Session one, no preamble: "Draft the Q3 campaign brief for Acme Group." Claude already knows you're a marketing director, the tone is professional-conversational, executive summaries go first, and Sara at Acme Group is the client contact. The prompt is six words. The output is already calibrated.

Having trained over 79,000 students across 74+ courses in AI and automation — from solopreneurs to enterprise teams — I've seen this pattern consistently: the people who extract the most value from Claude aren't the ones writing the most elaborate prompts. They're the ones who did the 10-minute setup once and let it compound across every session after that. The prompt isn't the leverage point. The setup is.

Start Simple and Let Friction Tell You What to Add Next

The most common mistake is over-engineering global instructions on day one. Start with exactly three things: your role and industry, one tone preference, one format preference. Live with that for a week.

You will notice moments mid-session where you type the same context you've typed before. That's your signal. Whatever you're repeating belongs in global instructions — not your prompts. Instructions evolve with how you work. They're not a document you perfect upfront; they're a living preference file that improves as friction surfaces.

The same applies to folder instructions: start a new project, add the client name and one template rule, then expand as the engagement grows. Most people get to a useful global setup in under 10 minutes, then refine for a few minutes each week until it stops needing attention.

The Payoff: Claude That Already Knows Your World

Once the three-layer stack is in place, your session-level prompts shrink to what they should always have been — specific, high-value requests. "Review the competitive positioning in this deck." "Draft an executive summary for the Q2 results." No front-loading. Claude's first response is already calibrated to your role, your client, and your format.

That shift — from a tool you configure every session to a colleague who already knows your context — is exactly what Claude global and folder instructions deliver. Not a shortcut. A fundamentally different baseline for every conversation you have from this point forward.

Open Settings → Co-work → Edit Global Instructions right now and add three lines: your role, your tone preference, your format rule. That's your starting point — and every session after that starts smarter than the one before.

Keep Learning

If this was useful, these are worth reading next:

• My 11-Year-Old Got Certified by Sheikh Hamdan's AI Initiative. Here's What He Built With It.
• Fix Broken AI Automations (Claude AI Troubleshooting Guide)
• Or go further with the AI Mastery Course — used by 79,000+ students across 150+ countries.

Source: Vendor documentation as of May 2026 — Anthropic, OpenAI, Google Gemini. Pricing in USD; AED equivalents at ~3.67 conversion.