I stumbled across a LinkedIn post that made me stop scrolling. The original poster laid out a brutally simple claim: most teams are using Claude wrong. Not because the tool is bad, but because nobody bothers to set it up properly. And honestly? After reading through the full breakdown, I think this AI professional nailed it.
The problem is familiar. You open Claude, type a prompt, get a decent answer, and move on. But then tomorrow you do the same thing again. And the day after that. Your teammates do it too, each one explaining the same context, the same tone requirements, the same formatting rules. It’s like training a new intern every single morning, except the intern has amnesia.
The expert behind this post shares a five-step system for setting up Claude Projects so your entire team works from the same playbook. No more starting from scratch. No more re-explaining everything. Here’s exactly how to do it, step by step.
Step 1: Let Claude Identify Your Projects
Before you create anything, you need to figure out what your team actually produces on repeat. The creator suggests opening a fresh Claude chat and pasting this prompt:
I work at [company + industry]. My team specifically helps [clients/goal]. You are helping me set up Claude for my team. We need to identify the 3-5 recurring deliverables my team produces. Interview me. Ask me ONE question at a time about: 1. What my team does day-to-day. 2. What we deliver to clients, leadership, or each other. 3. What tasks feel repetitive every week or month. 4. What each person always ends up redoing.
After about four back-and-forth exchanges, Claude gives you a clean list of your recurring deliverables. This is your Project list. The reason this step matters so much is that most people skip the discovery phase entirely. They jump into building without knowing what they’re building for. Four quick Q&As save you from creating Projects nobody actually needs.
Step 2: Create Each Project
Now head to Claude’s Projects page. Click “Teams,” then “New Project.” Create one Project per deliverable you identified in Step 1. Set the visibility to “entire workspace” so your whole team can access it. Repeat for all three to five deliverables.
This takes about two minutes per Project. The key here is one Project per deliverable, not one giant Project for everything. Keeping them separate means each Project stays focused, and Claude’s responses stay relevant. Think of it like having separate folders instead of one messy drawer.
Step 3: Upload the Right Context
This is where the magic happens. Open each Project and upload exactly three things:
- One gold-standard example: the best version of this deliverable your team has ever produced
- Relevant background docs: brand guides, client info, industry context
- The brief your team currently follows: whatever template or checklist people use today
Why three items specifically? Because it gives Claude enough context to understand quality, audience, and structure without drowning it in noise. Too little context and you get generic output. Too much and the AI loses focus. Three well-chosen uploads hit the sweet spot.
Step 4: Generate Tailored Instructions
Here’s a clever move from the post’s author. Instead of writing Project instructions yourself, you let Claude write them. Open a separate Claude chat (the contributor recommends using Opus with Extended Thinking) and paste this prompt:
I’m setting up this Claude Project for: [name of deliverable]. I’ve uploaded example outputs and background docs. Generate a Project Instruction set that includes: WHAT THIS IS: One sentence. WHO IT’S FOR: The audience. TONE & FORMAT: How it should read. QUALITY BAR: What separates good from bad. GUARDRAILS: What to never do. Format as a ready-to-paste instruction block.
Copy the output, go back to your Project, paste it into the Instructions field, and save. The rationale is smart: Claude is better at writing structured instructions for itself than most humans are. You get a consistent, well-formatted instruction block that covers tone, audience, quality standards, and guardrails, all in one paste.
Step 5: Test and Validate
Don’t skip this one. Open a new chat inside the Project and paste this prompt:
Based on the instructions and examples in this Project, produce a sample [deliverable name] for [a recent or fictional scenario]. Then critique your own output: what matches our standards, what doesn’t, and what should I add to make it better?
Claude will generate a sample, then self-critique. This is incredibly useful because it tells you exactly what’s missing from your setup. If it says something like “I don’t have enough info about client naming conventions,” you know to upload that document. The self-critique loop turns your first test into a diagnostic tool.
I think this step is what separates a Project that kinda works from one that genuinely saves hours. Your team’s first real interaction with the Project becomes the quality check.
🔄 What Changes After Setup
- Every chat in the Project already knows the rules, no re-explaining needed
- Your team stops prompting from scratch every time
- You stop re-explaining tone, format, and context in every conversation
- New hires get your team’s standards on day one, built right into the tool
I was genuinely impressed by how practical this whole system is. It’s not theoretical, it’s not “10 ways AI might help someday.” It’s a concrete setup you can finish in an afternoon that pays off every single day after. The innovation here isn’t any one prompt. It’s the sequence: discover, create, contextualize, instruct, validate. Each step builds on the last, and by the end, you’ve got a Claude workspace that actually reflects how your team works.
If you’ve been using Claude as a glorified search bar, this five-step setup from the original poster is worth your time. Check out the full LinkedIn post for all the details and prompts in one place.