On this page
Learn how an AI Project combines instructions and files into your own version of Claude or ChatGPT that knows your business.
Key takeaways
- An AI Project is your own version of Claude or ChatGPT that knows your business.
- AI Project = Instructions + Knowledge Base.
- Instructions are the rules: how it behaves, its tone, what it always or never does.
- The knowledge base is the files: what it knows, like your website, past emails, SOPs and examples.
Do this now: Think of one job you'd hand to a project. Write down what its instructions would say and which files it would need.
Class Notes
🪜 Quick Context: The AI Efficiency Ladder
Before we dive into projects, let's remember where they fit.
Think of your AI skills like climbing a ladder:
| Level | What It Is | Best For |
|---|---|---|
| 1. On-the-Fly Prompting | Quick, one-time prompts you make up on the spot | Random tasks, quick answers |
| 2. Repeatable Prompts | Prompts you save and reuse | Tasks you do over and over |
| 3. AI Projects | Trained AI with instructions + knowledge | Complex work that needs your voice, your data, your rules |
| 4. Automation | AI that runs without you | Repeatable AND predictable tasks |
Here's the pattern:
- Good on-the-fly prompts become repeatable prompts
- Repeatable prompts you use constantly become projects
- Projects that run the same way every time become automations
Today we're focused on Level 3: AI Projects.
This is where AI stops being generic and starts thinking like you
👀 What Is an AI Project?
An AI Project is a custom AI system you train to think and work like you (or an expert you design).
It's not regular ChatGPT or Claude. It's YOUR version of ChatGPT or Claude.
The simple definition:
An AI Project = Instructions + Knowledge Base
When you combine clear instructions with the right knowledge files, you get an AI that…
- Knows your business
- Uses your voice
- Follows your rules
- Gives outputs that actually sound like you
🧱 The Two Building Blocks of Every AI Project
1. Instructions (The Rules)
Instructions tell the AI how to behave.
Think of it like a job description for a new employee:
- What is your job?
- How should you talk?
- What rules do you follow?
- What should you always/never do?
Examples of what goes in instructions:
- "Always write in a casual, friendly tone"
- "Never use jargon or complex words"
- "When someone asks about pricing, respond with these 3 questions first"
- "Format all responses with bullet points"
Key point: Instructions set the personality, tone, and rules.
2. Knowledge Base (The Files)
The knowledge base is what the AI knows.
These are files you upload so the AI has information to pull from:
- PDFs
- Google Docs (exported as text)
- Transcripts
- Templates
- Examples of great work
- Data about your business
Examples of what goes in a knowledge base:
- Your website content
- Your best social media posts
- Your ideal client profile
- Past emails or scripts that worked
- Case studies or testimonials
- SOPs or processes
Key point: Knowledge is the data. Instructions are the rules for using that data.
Together they create: An AI that knows your stuff and follows your rules.
🙋♀️ Where Do You Build AI Projects?
Different platforms call AI Projects different names. Here's the breakdown:
| Platform | What They Call It | Share Externally? | Best For |
|---|---|---|---|
| ChatGPT | CustomGPT | ✅ Yes (anyone with link) | Client tools, public-facing projects |
| ChatGPT | Project | ✅ Yes (invite people by email) | Your own work or small-team collaboration |
| Claude | Project | ❌ No (you + team) | Internal tools, team collaboration |
| Microsoft Copilot | Copilot Agent | ✅ Within org | Enterprise teams on Microsoft 365 |
| Google Gemini | Gem | ✅ Yes (anyone with link) | Google Workspace users |
Platform Comparison: Which Should You Use?
Need to share with clients or the public? → Custom GPT (ChatGPT)
Only for you or your internal team? → Claude Project or ChatGPT Project
Your company runs on Microsoft 365? → Copilot Agent
You live in Google Workspace? → Gemini Gem
Good news: The instructions work the same way across all platforms. Build once, adapt anywhere.
👷♀️ Platform Specific Instructions
How to Build in ChatGPT: Custom GPT
Use this when you want to share with others.
Step 1: Go to "Explore GPTs"
Click the Explore button in the left sidebar, then click "Create."
Step 2: Switch to "Configure" Tab
This is where you build. Skip the "Create" tab.
Step 3: Add Your Instructions
This is the job description. Tell it:
- What its role is
- How to respond
- What voice/tone to use
- Any rules to follow
Step 4: Upload Your Knowledge Base
Click "Upload files" and add your documents (up to 20 files).
Step 5: Save and Test
Click "Update" in the top right. Start chatting and tune it.
How to Build in ChatGPT: Project
Use this for personal organization—keeps your chats grouped.
Step 1: Click "Projects" in the Left Sidebar
Then click the + to create a new project.
Step 2: Name Your Project
Example: "Brand Guardian - Marketing"
Step 3: Add Instructions
Click the settings icon and add your custom instructions.
Step 4: Upload Files
Add up to 10 files to your project knowledge base.
Step 5: Start Chatting
All chats inside this project will use your instructions and files.
Note: ChatGPT Projects remember everything inside the project, and you can share a project with specific people by email.
How to Build in Claude: Project
Use this for internal team tools and deep work.
Step 1: Click "Projects" in the Left Sidebar
Then click "New Project."
Step 2: Name It and Add a Description
Example: "Brand Guardian - [Your Company] - Marketing Copywriter"
Step 3: Add Your Instructions
Click the instructions area and paste in your rules.
Step 4: Upload Your Knowledge Base Files
Upload PDFs, text files, or paste in plain text.
Step 5: Start a New Chat Inside the Project
Each new conversation uses your instructions and knowledge automatically.
Note: Claude Projects cannot be shared outside your organization. For client-facing tools, use a Custom GPT instead.
How to Build in Microsoft Copilot: Agent
Use this if your company runs on Microsoft 365.
Step 1: Open Copilot Studio
Go to copilotstudio.microsoft.com or access through your Microsoft 365 apps.
Step 2: Create New Agent
Click "Create" and select "New Agent."
Step 3: Name and Describe Your Agent
Example: "Brand Guardian TUB"
Step 4: Add Instructions
Paste your custom instructions in the instructions field. (Same instructions you'd use in ChatGPT or Claude work here.)
Step 5: Add Knowledge Sources
Connect to SharePoint, OneDrive, or upload files directly.
Step 6: Publish and Share
Share within your organization through Teams or other Microsoft apps.
Note: Copilot has an 8,000 character limit for instructions—same as Custom GPTs.
How to Build in Google Gemini: Gem
Use this if you live in Google Workspace.
Step 1: Open Gemini
Go to gemini.google.com
Step 2: Click "Gem Manager"
Find it in the left sidebar, then click "New Gem."
Step 3: Name Your Gem
Example: "Brand Guardian - Marketing"
Step 4: Add Instructions
Write out your custom instructions in the instructions field.
Step 5: Save and Use
Your Gem is now available in your Gem library.
Note: Gems don't have the same robust knowledge base upload as other platforms (yet). They work best for instruction-heavy projects where you'll paste context into each chat.
⬅️ Back to your pre-watch list: Helping Hands (all videos) · General · Advanced
