Claude has no idea who you are.

Every time you open a new chat, it starts fresh. No memory of your business. No memory of what you're building. No memory of the conversation you had yesterday that took 45 minutes to get right.

Projects fix this.

Not in a complicated way. You create a workspace, upload a few documents about your business, write some instructions on how you want Claude to behave,1 and from that point on, every conversation inside that project starts with all of that context already loaded.

That's really it.

The rule of thumb for when to create a project.

Simple: if you need Claude to show up differently for a specific type of work, create a project.

Asking for restaurant recommendations? No project needed. Strategizing about your business, creating content, analyzing data, managing client work, anything where you want Claude to actually know your context and produce specific, tailored results, that needs a project.

Think of each project as a dedicated room. One for business strategy. One for content creation. One for client work. One for data analysis. When you walk into each room, everything relevant is already there.

The structure of a good project, the pyramid.

Before building anything, understand the hierarchy.

Files, the foundation. Everything Claude needs to know. SOPs, business documents, analytics, transcripts, legal documents, anything with information Claude should always have access to. Without files, Claude has nothing to work from. Everything else in the project sits on top of this.

Instructions, the behavior layer. How Claude should act, respond, and think inside this project. What format it should use. What it should always reference. What it should challenge versus accept. This goes on top of the files.

Memory + automations, the active layer. What Claude has learned about you over time within the project. Scheduled tasks. Prompts that run automatically. The intelligence that builds as you use it.

Build from the bottom up. Files first. Then instructions. Then automation.

Step 1: Create the project.

In Claude, go to the Projects section. Click New Project.

Give it a name that describes the type of work, "Business Strategy," "Content Repurposing," "Client Research," whatever it is. Add a short description of what you're trying to achieve. This is just for your own organization, it doesn't affect how Claude behaves.

Create the project. You now have a dedicated workspace.

Quick tip: Star the projects you use most. They pin to the top of your sidebar. The ones you're in every day should be immediately accessible.

Step 2: The business overview file (put this in every project).

The single most valuable document you can give Claude is a business overview.

Not because it's complicated, because it's foundational. Every output Claude gives you is only as specific as the context it has about you. Generic context produces generic advice. Specific context produces advice that's actually useful.

Here's what a business overview should include:

Company identity: Who do you help? What do you do? One-sentence description. How long have you been in business?

Revenue model: What are your offers? What do you charge? What's your pricing structure?

Offer stack: Each offer explained in depth. What's included, what it costs, who it's for.

Team and operations: Who's on your team? What does each person do? How many hours? What are the key processes?

Market position: How are you different from competitors? Why do clients choose you?

Growth profile: How are you acquiring customers right now? What channels are working?

90-day priorities: This is the most important section. What are you trying to accomplish right now? Claude is best when it can see a specific target. Give it your actual goals, not vague ambitions.

Upload this to every project. Even if the project is about content creation or data analysis, having Claude understand your business context means it can flag when advice doesn't align with your actual goals.

How to build it without writing it from scratch:

Open a new Claude chat and paste this:

“I want to create a business overview document that I can upload
to my Claude projects to give Claude persistent context about my business.

Interview me with questions across these seven sections:
company identity, revenue model, offer stack, team and operations,
market position, growth profile, and 90-day priorities.

Ask me about each section. Then produce a clean markdown file
I can download and upload to any project.”

Claude walks you through a Q&A. You answer. It produces the document. Download it. Upload it to every project you create.

Step 3: Add your other files.

The business overview is the baseline. Everything else is specific to the project.

For a business strategy project: P&L statements, pricing history, customer survey data, competitor analysis, your last 90 days of revenue breakdown.

For a content project: Video transcripts, past high-performing posts, your brand voice guide, examples of writing you want to emulate.

For a client project: The client's brief, past deliverables, communication history, their brand guidelines.

For an operations project: SOPs for every key process. Onboarding checklists. Hiring criteria. Meeting templates.

One practical note on capacity: 15 full YouTube video transcripts, each 30 minutes of dense content, uses roughly 10% of a project's total capacity. You have far more room than you think. Upload generously.

And files stay. Unlike a regular chat where uploaded files disappear when you start a new conversation, project files are always there. You upload once. You never upload again.

Step 4: Write your project instructions.

This is where the project becomes genuinely powerful.

Project instructions are backend prompts that shape how Claude behaves inside this specific project. They're separate from your account-level settings, they only apply here, only in this context.

Here's what makes instructions actually work:

Put Claude in a role. Not because it dramatically changes the model's capability, but because framing Claude as a business strategist makes it act like one. It approaches problems from that lens, uses that vocabulary, thinks at that level.

Tell it what to always reference. If you have a business overview in your files, tell Claude to read it before answering anything about your business. If you have a style guide, tell it to check every piece of content against that guide. Make the file reference automatic, not dependent on you remembering to ask.

Tell it how to format outputs. You prefer tables? Say so. You want bullet points? Bullet points. You want it to always produce an artifact? Say that. Formatting preferences inside instructions apply to every response, forever, without you asking.

Tell it what to challenge. This is the underrated one. If you tell Claude to poke holes in every idea before helping you build it, it stops being a yes-machine. It starts being something that actually helps you think.

Tell it what to flag. If something doesn't connect to your 90-day priorities, tell Claude to flag it. You'll be surprised how often advice sounds good but is actually pulling you off course.

Here's a template for a business strategy project:

“You are a business strategist. Help me make decisions about my
business direction, pricing, offers, and operations.

Before answering anything business-related, reference my business
overview document to ground your answers in my real numbers and goals.

Rules:

  • Use my real offers, real numbers, and real priorities. No generic advice.

  • Lead with your recommendation. Then explain why.

  • If I pitch an idea, poke holes in it before helping me build it.

  • Name the trade-off in every recommendation.

  • If something doesn't connect to my 90-day priorities, flag it.

  • Use tables, calculations, and data breakdowns whenever possible.

  • Never validate an idea just because I seem excited about it.”

Save the instructions. Now every conversation in this project runs through that filter, without you having to ask for any of it.

What this actually looks like in practice.

With a business overview uploaded and instructions set, try a prompt like this:

“What are the blind spots in my current pricing strategy?”

With no context, Claude gives you a generic answer about pricing psychology.

With a business overview and proper instructions, it reads your actual offer stack, finds the specific gap, say, a 5x price jump between your $297 course and your $1,497 group program, and tells you exactly what that gap is doing to your conversion rate and what to do about it.

Try something bigger:

“I'm thinking about eliminating my one-on-one consulting.”

Without context: "That's a significant decision. Here are some pros and cons."

With a business overview that shows one-on-one consulting is 55% of your revenue: "That's a $23,000-a-month decision. Here's the case against doing it today. Here's the case for sunsetting it over 90 days. Here are the exact steps."

Same Claude. Completely different quality of answer. The difference is entirely in the context you gave it.

Claude Cowork projects: the next level.

Everything above lives in Claude's web interface. It's powerful. But Claude Cowork projects on the desktop app take it further, from having a conversation to getting work done.

The fundamental difference: Cowork projects are task-focused, not conversation-focused. You don't chat with Claude about what to do. You tell it to do something and it executes, writing files to your computer, calling APIs, running research, producing actual deliverables.

Setting up a Cowork project:

Open the Claude desktop app. Go to the Cowork tab. Click ProjectsNew Project.

You get three setup options: start from scratch, import a project from Claude chat, or connect a folder from your computer. The folder option is powerful, it gives Claude access to files on your machine, not just files you've uploaded.

Give it instructions. Upload files, your business overview, any anti-AI-slop writing guides, brand voice documents, whatever is relevant. Set a project folder location on your computer. Create it.

The interface looks slightly different from Claude web. You still have your prompt bar, your instructions, your files. But you also have two new sections: Scheduled and Connectors.

Scheduled tasks, automation on autopilot:

Scheduled lets you set up recurring tasks that run automatically inside this project, while you're sleeping, while you're in another room, while you're doing something else.

You could set up:

  • Daily LinkedIn post generation from your latest content at 7am

  • Weekly email triage every Monday morning

  • Monthly analytics summary the first of each month

  • Hourly check of a specific inbox or source

If your computer is on, the task runs. It uses all the context in your project, your files, your instructions, your memory, every time.

Connectors, Claude inside your tools:

Cowork connects directly to your apps. Gmail, Slack, Notion, Google Calendar, and more. Once connected, Claude can read from and act on those tools as part of any project task.

This is what turns a project from "a place I chat with Claude" into "a system that runs parts of my business."

Multi-step tasks in action:

Here's the difference you'll feel immediately.

In Claude web, you ask Claude to do something and it produces an answer in the chat. You copy it, go somewhere else, paste it, move on.

In Cowork, you describe a multi-step outcome and Claude handles the whole thing:

“I want you to create seven LinkedIn posts from this YouTube transcript. Research what's working for top creators in the AI space first. Then use my anti-slop document to make sure none of them sound like AI.

Save each post as a separate file in my projects folder when you're done.”

What happens: Claude researches LinkedIn best practices. It reads the transcript. It extracts the key themes. It writes seven posts. It reviews each one against your style guide. It saves seven separate files to your computer.

While you do something else.

Seven posts, fully produced, in about 90 seconds. Not perfect on the first run, the anti-slop document helps but you'll still refine, but the starting point is dramatically better than generic AI output, and the finished files are waiting for you when you come back.

Build your first one today.

Pick one type of work you do regularly. Business strategy. Content creation. Client research. Data analysis. Any recurring workflow.

Create the project. Five minutes.

Build your business overview. Use the interview prompt from Step 2. Thirty minutes.

Write your instructions. Use the template from Step 4 as a starting point. Ten minutes.

Upload two or three relevant files. Whatever is most specific to this type of work.

Have one real conversation. Ask something you'd normally spend an hour figuring out on your own.

Do that once. Then come back tomorrow and notice that Claude already knows what you discussed yesterday, without you explaining anything.

That's the moment it clicks.

If this changed how you use Claude, send it to one person still starting from scratch every session. That's all I ask.

See you next week.

Keep Reading