The Problem With Starting From Scratch Every Time
Most people use AI like a search engine. Open the chat, type a question, get an answer, close the tab. Next day, start over. No memory. No context. No continuity.
That's like hiring someone new every morning, giving them zero onboarding, and wondering why the output is generic.
Claude Projects fix this. A Project is a persistent workspace where you upload your documents, write a set of instructions that define how Claude should behave, and then every conversation inside that project starts with full context. You set it up once. Every interaction after that is better because Claude already knows your organisation, your style, and your constraints.
I have Projects for course production, policy drafting, board prep, and client work. Each one saves me the 10-15 minutes I used to spend re-explaining context at the start of every session. Over a week, that's hours.
Here's exactly how to set one up.
What You Need
Claude Projects are available on both the free and Pro tiers. Free tier gives you a limited number of projects with lower usage caps. Pro ($20/month) gives you unlimited projects, access to the most capable models, and significantly higher usage limits.
If you're going to use this daily for real work, Pro is worth it. If you want to test the concept first, the free tier works fine.
| Feature | Free | Pro ($20/mo) | |---|---|---| | Projects | Limited (~5) | Unlimited | | File uploads | Yes | Yes (up to ~10 MB text per project) | | Custom instructions | Yes | Yes | | Models available | Sonnet | Opus, Sonnet, Haiku (your choice) | | Usage limits | Low — you'll hit limits within an hour or two of heavy use | 5x+ more capacity |
Step-by-Step: Creating Your First Project
Step 1. Open your browser and go to claude.ai. Sign in with your account. If you don't have one, click "Sign up" — you can register with an email address or Google account.
Step 2. Once you're logged in, look at the left sidebar. You'll see a few navigation items: your recent conversations, and a section labelled "Projects." Click it. If your sidebar is collapsed (common on smaller screens), look for a folder icon near the top of the sidebar.
Step 3. Click "Create a Project" or the "+" button next to the Projects heading. A new project opens with two fields to fill in: a name and an optional description.
Step 4. Give your project a clear name. Something specific to the work it supports. I name mine by function: "Course Production — AI Leadership," "Board Prep Q2 2026," "Parent Communications." Avoid vague names like "Work stuff" or "AI project." You'll end up with multiple projects and need to tell them apart at a glance.
Step 5. Now you'll see the project workspace. There are two key areas to configure before you start chatting: Project Instructions and Project Knowledge (file uploads). Let's do instructions first.
Writing Your Project Instructions
This is the most important step and the one most people skip. Project Instructions are a persistent system prompt — they shape every conversation inside this project. Think of it as a job description for the AI.
Step 6. Click the "Project Instructions" field (or the pencil/edit icon next to it). A text area opens. This is where you define how Claude should behave in every conversation within this project.
Here's the structure I use. Copy this template and modify the bracketed sections:
## Your Role
You are [specific role] helping me with [specific domain].
You write in a [tone] style. You are [personality traits].
## About My Organisation
- [Type of organisation and size]
- [Key context — audience, constraints, values]
- [Anything Claude needs to know to give relevant advice]
## How to Write
- Use [style preferences: short paragraphs, contractions, Australian English, etc.]
- Default to [output format: bullet points, structured documents, etc.]
- Always [positive instruction: cite sources, provide examples, etc.]
## Rules
- Never [hard constraint: never invent statistics, never use jargon, etc.]
- Always [hard constraint: always flag assumptions, always ask before proceeding on ambiguous requests]
- If unsure, [what to do: ask me, flag it, provide options]
Step 7. Click "Save" when you're done. Your instructions are now active for every conversation in this project.
A concrete example. Here's a simplified version of the instructions I use for my course production project:
## Your Role
You are a senior instructional designer and content writer helping me
produce an AI Leadership video course for school leaders. You write
in a confident, direct, practical style. No academic jargon. No filler.
Every sentence earns its place.
## About the Course
- 8 modules, 25 units, ~4.5 hours of content
- Audience: school leaders, principals, department heads
- Delivery: AI avatar video (HeyGen) + interactive elements
- Tone benchmark: "Monday morning useful"
## How to Write
- Short paragraphs (1-3 sentences)
- Use contractions (we're, don't, it's)
- Max 20-word sentences for scripts
- Include [PAUSE] markers after key points in scripts
## Rules
- Never use: landscape, navigate, realm, journey, dive in, unpack
- Never invent statistics — ask me for source data
- Always include learning objectives per segment
- Flag any claim that needs verification
Every conversation I start in this project inherits those rules. I never re-explain my tone, my audience, or my constraints. Claude just knows.
Uploading Your Documents
Step 8. In the project workspace, look for the "Add content" or file upload area. It's usually labelled "Project Knowledge" with an upload icon. Click it to open a file browser, or drag files directly from your file explorer into the area.
Supported file types: PDF, Word (.docx), plain text (.txt), Markdown (.md), CSV, PowerPoint (.pptx), images (PNG, JPG, GIF), JSON, and most code file types.
Not supported: Excel (.xlsx) — convert to CSV first. No audio, video, or ZIP files.
What to upload depends on your project's purpose:
| Project Type | What to Upload | |---|---| | Board prep | Previous board report, financial summary, strategic plan, meeting notes | | Policy drafting | Current policy handbook, regulatory requirements, school values statement | | Communications | Brand/tone guide, past examples of good emails, key stakeholder context | | Course production | Course architecture, style guide, research documents, example scripts |
Step 9. Upload your files. They'll appear as a list in the project. There's no folder structure — everything sits in a flat list. Name your files clearly before uploading (e.g., "Strategic-Plan-2025-27.pdf" not "Document (3).pdf").
Size limit: Approximately 10 MB of text content total per project. That's a lot — roughly five million words. You're unlikely to hit this limit unless you're uploading large datasets.
Starting Your First Conversation
Step 10. With instructions saved and documents uploaded, click "New Conversation" (or the chat icon) within your project. A fresh chat opens — but it's not actually fresh. Claude already has your instructions and access to every document you uploaded.
Step 11. Type your first message. You can immediately reference your uploaded documents by name: "Summarise the key enrolment trends from the strategic plan" or "Draft a parent email about the new homework policy, following our tone guide."
Notice the difference. You didn't need to paste the strategic plan into the chat. You didn't need to re-explain your tone. You didn't need to set up any context. It's all already there.
Step 12. When you start a new conversation tomorrow, click into the same project and start a new chat. Your instructions and documents persist. Only the conversation history resets — which is actually what you want, since long conversations eventually push important context out of the active window.
Four Projects Worth Building
If you're not sure where to start, build one of these:
1. Weekly Communications. Upload your brand voice document (or write a description of your tone), a few past emails you were proud of, and your stakeholder list. Instructions: define your organisation, preferred length, and formatting rules. Use it every Monday to batch your weekly emails.
2. Policy Drafting. Upload your current policy handbook, relevant regulations, and any governance framework documents. Instructions: define your organisation type, the audience for policies, and your preferred structure (Purpose / Scope / Statement / Procedures / Review Date). When you need a new policy, you're one prompt away from a solid first draft.
3. Board Prep. Upload the previous board report, current data (enrollment, financials, key metrics as CSV), and your strategic plan. Instructions: define report format, what the board cares about, and your voice. Use it quarterly to go from raw data to board-ready summary.
4. Research Synthesis. Upload research papers, industry reports, and data sources on a topic you're investigating. Instructions: define your analytical lens and output preferences. Use it to find patterns, contradictions, and gaps across your sources.
What This Doesn't Do
Projects don't update automatically. If your enrollment data changes, you need to re-upload the file. There's no live connection to Google Drive or your email.
Projects don't share between accounts. Your colleague can't access your project unless you're both on a Team plan ($25-30/seat/month). On free and Pro, projects are personal.
Claude can still be wrong. Uploading documents improves relevance dramatically, but it doesn't eliminate all errors. You still need to verify important outputs — especially statistics, dates, and policy references. The difference is that Claude is now drawing from your actual documents instead of its training data, so errors are less frequent and easier to catch.
The Bigger Picture
Every time you open a bare AI chat and start typing, you're leaving quality on the table. The AI doesn't know your organisation, your audience, your standards, or your constraints. It fills those gaps with generic assumptions.
Projects replace assumptions with information. That's the whole concept.
The investment is about 30 minutes to set up a project properly. The return is every interaction after that being measurably better — more relevant, more consistent, more useful. Over weeks and months, that compounds.
I teach context engineering — the skill of building AI environments like this — in depth in my AI Leadership course. Module 2 covers the RCFCC framework, Claude Projects, and NotebookLM. If you want the full system, that's where to start.
