Time: ~20 minutes
Skill level: Intermediate
What you'll have: A copilot that acts as your personal AI development consultant — one that knows your role, your team's challenges, and how you like to work, so every copilot you build from here fits your context from the start.

Why build a copilot builder?

Writing a good system prompt from scratch takes thought. You need to define what the copilot does, who it's for, what it should and shouldn't do, and how it should respond. That process is much faster when you have a thinking partner who already understands your context.

Your copilot builder is that partner. Because it knows your role, industry, and communication preferences, it can give you targeted suggestions instead of generic advice — and help you refine prompts through conversation rather than trial and error.

Step 1: Define your context (5 minutes)

Before you build anything, take a few minutes to gather the details that will make your copilot builder genuinely useful. The more specific you are here, the better the output will be.

Write down:

• Your name and job title
• Your company name and what it does (one or two sentences)
• Your industry and approximate team size
• Two or three recurring work challenges where AI could help
• How you prefer to receive information: direct and concise, or detailed with context?
• Your preferred output format: structured lists, prose paragraphs, or a mix?
• A name and personality for your copilot builder (professional, conversational, encouraging — whatever fits how you work)

📘

Keep this somewhere handy. You'll paste parts of it directly into a system prompt in the next step.

Step 2: Write the system prompt (5 minutes)

Your copilot builder needs a system prompt that combines your personal context with clear instructions about how it should help you. Use this template as your starting point, filling in the brackets with your details from Step 1.

PERSONAL CONTEXT:
- Name: [Your name]
- Role: [Your job title]
- Company: [Company name and one-sentence description]
- Industry: [Your industry]
- Team size: [Approximate number of people you work with]
- Main challenges: [Two or three recurring work challenges]

COPILOT PERSONA:
- Name: [What to call this copilot]
- Personality: [Professional / conversational / encouraging / direct — your choice]
- Communication style: [How you prefer to receive information]
- Output format: [Bullet lists / paragraphs / structured sections — your preference]

---

CORE INSTRUCTIONS:
You are a specialized copilot creation assistant. Your role is to help me brainstorm, develop, and refine copilots for my specific work context.

When I bring you a challenge or idea:
1. Ask clarifying questions to understand what I'm trying to accomplish, who will use it, and what success looks like.
2. Suggest relevant copilot approaches based on my role, industry, and challenges.
3. Help me write a clear, effective system prompt — including role definition, specific instructions, output format requirements, and relevant context.
4. Help me test and improve the concept through conversation.

Always apply my personal context to your suggestions. Tailor your recommendations to my work style and communication preferences.

This template is a strong starting point, but it's meant to be yours. Edit the tone, add specific terminology your team uses, or include examples of work you do regularly.

Step 3: Create the copilot (5 minutes)

1. Open the Copilots sidebar and select My Copilot Hub, then choose Create copilot.
2. Give it the name you chose in Step 1 (something like "Alex – Copilot Builder" or "My Copilot Consultant").
3. Write a short description — one or two sentences about what it does. This appears in your copilot list and helps future-you remember its purpose.
4. Paste your completed system prompt into the System prompt field.
5. Leave the model set to the default. The default language model handles this kind of structured reasoning well.
6. Enable Search internet if you want your copilot builder to reference current AI best practices and examples.
7. Add two or three conversation starters to prompt the most common use cases, for example:
   • "Help me build a copilot for [specific task]"
   • "Review this system prompt and suggest improvements"
   • "What copilot would help my team most right now?"
8. Select Save.

For a full walkthrough of copilot creation options, see Custom copilots.

Step 4: Test with a real challenge (5 minutes)

Don't test with a hypothetical — use something you're actually working on. This gives you an immediate sense of whether the copilot's suggestions fit your context.

Try one of these to start:

Real challenge test. Describe something you do repeatedly that takes longer than it should. Ask your copilot builder what kind of copilot could help. Evaluate whether the suggestion is specific to your role or generic.

System prompt review. Paste in a system prompt from a copilot you've already built and ask for improvement suggestions. A well-configured copilot builder will identify what's missing or unclear.

Brainstorming session. Ask for copilot ideas for your team or department. The suggestions should reference your industry and specific challenges — not just general AI use cases.

As you test, look for: Does it remember your context without you repeating it? Are the suggestions specific or generic? Does the communication style feel right?

Step 5: Refine based on what you find (a few minutes, ongoing)

Most first versions need small adjustments. Here's what to tune and how:

Responses feel too formal or casual. Update the personality setting in the COPILOT PERSONA section of the system prompt.

Output format doesn't match your preferences. Be more explicit in the CORE INSTRUCTIONS — for example, "Always structure system prompts with numbered sections" or "Lead with the recommendation before explaining the reasoning."

Suggestions are too generic. Add more specificity to your context. Include the names of tools your team uses, the types of documents you typically work with, or specific workflows you want copilots to support.

It doesn't reference your challenges. Add an instruction like: "When suggesting copilot approaches, always connect your recommendation to one of my main challenges."

Update the system prompt directly in the copilot's settings whenever something isn't working. Iteration is the point.

Next steps

Now that you have a copilot builder, use it regularly:

• Before creating any new copilot, run your use case by your copilot builder first. It will help you clarify scope before you spend time in settings.
• When an existing copilot underperforms, paste the system prompt into a chat with your builder and ask what's missing.
• When onboarding teammates to copilot creation, share your copilot builder with them so they have a starting point that already fits your team's context.

To make your builder even more useful over time, upload examples of effective copilots you've created to a knowledge base and link it to the copilot. This gives it real reference material to draw from.

Related guides:

• Custom copilots — Full reference for copilot settings
• Build your first custom copilot — Step-by-step for new builders
• Scale a copilot to your team — When you're ready to share what you've built
• Knowledge bases — Connecting documents to enrich your copilot builder over time