How to Write a System Prompt for Any AI

How to Write a System Prompt for Any AI

A system prompt is a standing instruction you save once that shapes every reply an assistant gives you. A good one names four things in this order: the role it should take, how it should write, what it should avoid, and any context about you it should assume. Everything else is refinement.

This guide covers what to put in each of those four parts, and how to carry one instruction across ChatGPT, Claude, Gemini, Copilot and Perplexity without rewriting it for each.

What Is a System Prompt?

Prompt customization means creating system-level instructions that shape how an LLM communicates and responds. Instead of accepting default behavior, you define:

• Communication style and tone
• Response structure and format
• Personality and approach
• What to avoid or emphasize

Why Customize Prompts?

Default LLM Behavior

Most LLMs default to:

• Overly polite, corporate language
• Formulaic response structures
• Generic, one-size-fits-all tone
• Meta-commentary about being an AI

Customized Benefits

Custom prompts let you:

• Match tone to your use case
• Get more natural, conversational responses
• Reduce robotic patterns
• Improve consistency
• Save time with better defaults

Core Principles of Prompt Customization

Principle 1: Be Specific

Vague prompts produce vague results. Be concrete about what you want.

Principle 2: Balance Guidelines

Tell the LLM what TO do and what to AVOID. Both are important.

DO:
Use contractions naturally
Vary response structure
Show personality
AVOID:
Formulaic openings
Numbering everything
Generic sign-offs

Principle 3: Match Context

Different contexts need different customization:

• Professional: Clear, direct, but friendly
• Casual: Relaxed, conversational, expressive
• Creative: Vivid, metaphorical, enthusiastic
• Technical: Precise, clear, but not robotic

Principle 4: Test and Iterate

Start with a basic prompt, test it, then refine based on results. Customization is an iterative process.

Basic Prompt Structure

1. Role Definition

Define who the AI is:

You are a helpful assistant with a natural, conversational communication style.

2. Positive Guidelines

Tell the AI what TO do:

Communication guidelines:
Use contractions (I'll, you're, it's) naturally
Vary your enthusiasm
Use analogies and examples
Acknowledge complexity

3. Negative Guidelines

Tell the AI what to AVOID:

Avoid:
Formulaic openings
Numbering everything
Meta-commentary
Generic sign-offs

Complete Customization Template

Here's a comprehensive template you can customize:

You are a helpful assistant with a natural, conversational communication style.
Communication guidelines:
Use contractions (I'll, you're, it's) naturally
Vary your enthusiasm—not everything is "amazing" or "incredible"
Skip meta-commentary about being an AI
Use analogies and examples to explain complex ideas
Acknowledge when something's tricky or nuanced
Mix paragraphs with occasional lists (not everything numbered)
Show personality while staying helpful
Think out loud—show your reasoning process
Avoid:
Formulaic openings like "Certainly!" or "I'd be delighted"
Numbering everything automatically
Announcing you're an AI
Overly corporate grammar
Generic sign-offs after every response

Customization by Use Case

Professional/Work

You are a knowledgeable colleague helping with professional tasks. Be professional 
but conversational—clear and direct without being stiff.
Style: Professional but friendly, like a helpful coworker
Tone: Clear, direct, but not robotic
Structure: Varied—paragraphs and lists when helpful

Personal/Learning

You're a patient teacher explaining concepts clearly. Break things down, use examples, 
and check for understanding.
Style: Casual and friendly, like explaining to a friend
Tone: Patient, clear, encouraging
Structure: Step-by-step when needed, flowing narrative otherwise

Creative

You're a creative collaborator. Use vivid language, metaphors, and analogies to 
bring ideas to life.
Style: Expressive and vivid
Tone: Enthusiastic, creative, exploratory
Structure: Varied and dynamic

Implementation by Platform

ChatGPT

Claude

Perplexity

Gemini

Advanced Customization Techniques

1. Context-Specific Customization

Add context about your use case:

You're helping me with [specific task/domain]. When responding:
Focus on [specific aspects]
Use [specific tone/style]
Emphasize [specific elements]

2. Multi-Prompt Strategy

Create different prompts for different scenarios:

• Work prompt: Professional but conversational
• Learning prompt: Patient and clear
• Creative prompt: Expressive and vivid

3. Layered Instructions

Combine general guidelines with specific requirements:

General: Natural, conversational communication style
Specific: When explaining technical concepts, use analogies
Context: I'm learning programming, so be patient and clear

4. Dynamic Adjustments

Add instructions that the LLM can adjust based on context:

Match your tone to the context: professional for work topics, casual for personal 
questions, creative for brainstorming sessions.

Common Customization Patterns

Pattern 1: Tone Adjustment

More casual: "Use very casual language, like texting a friend"
More professional: "Professional but conversational, not overly casual"

Pattern 2: Structure Variation

"Vary your response structure. Don't default to numbered lists. Mix paragraphs, 
bullets, and narrative flow based on what works best for each response."

Pattern 3: Personality Injection

"Show personality through analogies, acknowledging complexity, and thinking out 
loud. Be helpful and accurate, but don't be robotic about it."

Pattern 4: Context Awareness

"When I ask work-related questions, be professional but conversational. For personal 
questions, be casual and friendly. For creative projects, be expressive and 
enthusiastic."

Testing Your Customization

Test Questions

Try these to evaluate your customization:

1. Complex explanation: "Explain [complex topic] simply"
2. Problem solving: "Help me solve [problem]"
3. Creative task: "Help me brainstorm [idea]"
4. Professional task: "Draft an email about [topic]"

Evaluation Criteria

• Natural, conversational tone
• Varied structure
• Appropriate for context
• Consistent across responses

• Still sounds robotic
• Formulaic structure
• Inconsistent tone
• Doesn't match context

Troubleshooting Common Issues

Issue: Still Sounds Robotic

• Add specific examples
• Increase specificity of guidelines
• Test different phrasings

Issue: Too Casual/Formal

• "More professional" or "More casual"
• Add context about your use case
• Provide examples of desired tone

Issue: Inconsistent Responses

• "Maintain this tone consistently"
• "Keep the same energy level"
• Add reinforcement of key points

Issue: Not Following Instructions

• Break down complex instructions
• Use clearer language
• Test if instructions are too long

Best Practices

1. Start Simple

Begin with basic customization, then add complexity as needed.

2. Test Regularly

Test your customization with various questions to ensure it works across contexts.

3. Iterate Based on Results

Refine your prompt based on actual responses, not assumptions.

4. Match Your Use Case

Customize for how you actually use the AI, not how you think you should.

5. Keep It Focused

Don't try to cover every scenario. Focus on your primary use case.

Key Takeaways

1. Customization = Better Results - Tailored prompts produce better responses
2. Be Specific - Concrete guidelines work better than vague ones
3. Match Context - Different use cases need different customization
4. Test and Iterate - Refinement is key to success
5. Start Simple - Begin basic, then add complexity

Ready to Customize Your LLM Prompts?

The fastest way to get started is with our interactive wizard. We'll guide you through creating a custom prompt tailored to your preferences and use case, then help you implement it on your preferred AI platform.