Custom AI Tools
Create specialist AI assistants that remember your context, learn your style, and work exactly how you want.
What Are Custom AI Tools?
A custom AI tool is a personalized version of an AI that "knows" your role, your preferences, your style, and your domain knowledge. Instead of explaining yourself every time, you set it up once, and it remembers forever.
Generalist AI
ChatGPT out of the box: Knows everything about everything, but nothing about YOUR business, YOUR students, or YOUR style. You have to re-explain context every conversation.
Problem: Repetitive context-setting. Inconsistent output format. No memory of your preferences.
Specialist AI (Custom Tool)
Claude Project + Your Instructions: Knows YOUR role, YOUR company, YOUR students, YOUR style. You set it up once. Every conversation assumes this context.
Advantage: Faster. Consistent. Aligned with your needs. Remembers your preferences.
The Analogy
Think of it like hiring a consultant. A generalist consultant needs to ask you basic questions every time: "What's your industry? Who are your customers? What's your budget?" A specialist consultant already knows this. You just say "here's the problem" and they dive in.
Claude Projects Deep Dive
Claude Projects is the easiest, most powerful way to create a custom AI. Here's how.
Create a New Project
- Go to claude.ai
- Look for 'Projects' in the sidebar (or create a new Project button).
- Click 'New Project'.
- Give it a name (e.g., 'My Teaching Assistant', 'Data Analysis Bot').
Write System Instructions
- In the Project, find the 'Project Instructions' field.
- Write instructions that tell Claude: who you are, what you do, your style, your constraints.
- Be specific. The more detail, the better the AI performs.
- See examples below.
What to include in Project Instructions:
- Your role and expertise (e.g., "You are a business analyst")
- Your domain knowledge (e.g., "You work in SaaS")
- Your preferences (e.g., "Always use bullet points")
- Your tone (e.g., "Professional but conversational")
- Key constraints (e.g., "Never share real company names")
Add Knowledge Documents
- In the Project, upload documents (PDFs, text files, images).
- Upload things the AI should reference: style guides, company policies, previous work, templates.
- The AI will use these as background knowledge when you ask questions.
Start Using It
- Save your Project.
- Go to the Project.
- Type a message. The AI will respond using your instructions and documents as context.
Pro tip: Each Project has its own conversation history. You can have multiple long conversations within one Project, and the AI remembers all of them.
Full Example: Teacher's AI Assistant
Now, every time this teacher opens the Project and asks for help, the AI knows exactly who they are, what they need, and how to respond. No re-explaining.
ChatGPT Custom GPTs: An Alternative
If you use ChatGPT (OpenAI), you can create Custom GPTs. The concept is the same as Claude Projects, but the interface is different.
How to Create a Custom GPT
- In ChatGPT (paid tier), click "Create" â "Create a GPT"
- Give it a name and description
- In "Instructions," write your system prompt (same as Claude Projects)
- Upload files (knowledge documents)
- Configure capabilities (web search, code execution, image generation)
- Save and start using
Google Gemini Gems
Google Gemini is adding a feature called "Gems" (similar to Projects and Custom GPTs). Here's what you need to know.
Example Gem: Student Research Assistant
What Makes a Great Custom AI? (5 Elements)
Whether it's a Claude Project, Custom GPT, or Gem, here are the 5 elements that separate "okay" from "excellent":
1. Role & Purpose (Crystal Clear)
The AI knows exactly what job it's doing and for whom. Not "help with writing" (vague), but "help a 3rd-grade teacher write differentiated math lessons in 10 minutes" (specific).
How to write it: Start with "You are a [role]. Your job is to [specific outcome]."
2. Tone (How You Sound)
Define your tone. Professional? Friendly? Formal? Casual? Encouraging? The AI adopts your voice.
Example: "Tone: Professional but warm. No jargon. Encouraging even when pointing out problems."
3. Knowledge Context (What It Should Know)
What domain knowledge should the AI have? Specific company info? Industry standards? Upload documents or explain in the instructions.
Example: Upload your company handbook, style guide, org chart, or previous work samples.
4. Output Format (How Responses Should Look)
Should responses be bullet points? Tables? Essays? Long form? Tell the AI explicitly.
Example: "Always format responses as: Executive Summary (1 para) â Key Points (3-5 bullets) â Recommendations (numbered)."
5. Guardrails (What NOT to Do)
What should the AI never do? Avoid certain topics? Warn before something risky? State your boundaries.
Example: "Never include real student names in examples. Always flag if you're uncertain about accuracy."
Fill-In Template
6 Custom AI Ideas by Profession
Here are starter prompts for 6 professions. Copy these and customize them for your specific needs.
For Teachers
For Managers/Leaders
For Developers
For Analysts/Data People
For Healthcare Providers
For Business/Product People
Hands-On: Create Your First Claude Project
Set Up Your Project
- Go to claude.ai.
- Click 'Create Project' (or find Projects in the sidebar).
- Name it something descriptive, like 'My [Job Title] Assistant' or '[Task Name] Helper'.
- Click 'Create'.
Write Your Project Instructions
- In the Project, find 'Project Instructions.'
- Copy the template from Section 5 above.
- Fill in all 5 sections: Role & Purpose, Tone, Knowledge Context, Output Format, Guardrails.
- Be specific. The more detail, the better the AI will perform.
- Save.
Add Knowledge Documents (Optional)
- Think of 2-3 documents that would help the AI understand your work.
- Examples: past projects, templates, style guides, checklists, examples of good output.
- Upload them to your Project.
- Tell the AI in your instructions that these documents are available.
Not required, but it makes the AI much more aligned with your needs.
Test It With 3 Real Tasks
- Open your Project.
- Give it a real task you'd actually do at work/school.
- Look at the response. Does it match what you need? Is the tone right? Is the format useful?
- Do this 3 times with different tasks.
- If something is off, refine your Project Instructions and test again.
Start Using It for Real
- Use your Project for actual work.
- Save good outputs to a folder for reference.
- After 1 month of use, revisit your instructions. What's working? What needs tweaking?
- You're building a personal AI that gets better over time.
Claude Projects (easiest, best for most people) | ChatGPT Custom GPTs (if you use ChatGPT) | Google Gemini Gems (coming soon)
1. Role & Purpose (clear job) | 2. Tone (how you sound) | 3. Knowledge Context (what it should know) | 4. Output Format (how responses look) | 5. Guardrails (what not to do)
1. Create Project/GPT | 2. Write instructions (use the template) | 3. Upload documents | 4. Test with 3 real tasks | 5. Refine based on results
Too vague instructions ("help with writing") | Not specific enough about your role/context | No examples of good output | Not testing before real use
Use specific, vivid language. Include constraints (what NOT to do). Upload examples of good output. Refine after 5 conversations. Keep it in one place for easy access.
Create multiple Projects for different roles (Teacher Project, Manager Project, etc.) | Revisit instructions monthly | Share your best Projects with colleagues