Skip to main content

​
Prompt Engineering Best Practices

This collection of system prompts reveals sophisticated techniques used by leading AI companies. Study these patterns to improve your own prompt engineering.

​
Key Techniques from Professional Prompts

1

Clear Identity and Context

Always establish who the AI is and what environment it’s operating in.Example from Claude (claude-sonnet-4.6.md:1-5):
The assistant is Claude, created by Anthropic.
The current date is Tuesday, February 17, 2026.
Claude is currently operating in a web or mobile chat interface...
Starting with identity, date, and context helps the model ground its responses appropriately.
2

Detailed Tool Instructions

When providing tools, include:
  • Clear trigger patterns for when to use each tool
  • Specific parameter selection guidance
  • Decision frameworks for choosing between tools
From Claude’s past_chats_tools (claude-sonnet-4.6.md:16-29):
  • Explicit trigger patterns (“you suggested”, “we decided”)
  • Low vs. high-confidence keyword extraction
  • When NOT to use tools
3

Personality Guidelines

Define tone and behavior clearly, but allow flexibility for context.Example from GPT-5.1 (gpt-5.1-default.md:1-3):
You are a plainspoken and direct AI coach that steers the user 
toward productive behavior... Be open minded and considerate of 
user opinions, but do not agree with the opinion if it conflicts 
with what you know.
Notice how it balances being helpful with maintaining objectivity.
4

Safety and Ethical Boundaries

Include specific guardrails and reminders for handling sensitive content.From Anthropic’s Reminders (claude.ai-injections.md:23-38):
  • Cyber security warnings
  • Content policy enforcement
  • Jailbreak detection patterns

​
Prompt Structure Patterns

​
1. Layered Instructions

Professional prompts use hierarchical structure:
  • Who is the AI?
  • What company created it?
  • What date/time context?
  • What interface is it running in?
  • Available tools and when to use them
  • Knowledge cutoff dates
  • Supported languages and formats
  • Technical constraints
  • Tone and personality
  • Safety reminders
  • Edge case handling
  • Long conversation guidance

​
2. Decision Frameworks

Instead of ambiguous instructions, use explicit decision trees: 
1. Time reference mentioned? → recent_chats
2. Specific topic/content mentioned? → conversation_search
3. Both time AND topic? → If specific time frame, use recent_chats
4. Vague reference? → Ask for clarification
5. No past reference? → Don't use tools
This pattern from Claude’s system prompt (claude-sonnet-4.6.md:88-95) eliminates ambiguity.

​
3. Example-Driven Guidance

Show, don’t just tell: 
## Example
For the user prompt "Wer hat im Jahr 2020 den Preis X erhalten?" 
this would result in generating the following tool_code block:
<tool_code>
print(Google Search(["Wer hat den X-Preis im 2020 gewonnen?"]))
</tool_code>
From Gemini’s system prompt (gemini-2.5-pro-webapp.md:24-29)

​
Common Anti-Patterns to Avoid

Vague Instructions❌ “Be helpful and friendly”✅ “Provide direct, objective technical info without unnecessary superlatives or emotional validation”
Missing Edge Cases❌ “Search past conversations when users ask”✅ Include explicit trigger patterns, parameter guidance, and when NOT to use the feature
Conflicting InstructionsEnsure personality guidelines don’t conflict with safety requirements or tool usage patterns.

​
Advanced Techniques

​
Critical Evaluation Over Agreement

From Claude’s long_conversation_reminder:
“Claude critically evaluates any theories, claims, and ideas presented to it rather than automatically agreeing or praising them… Claude prioritizes truthfulness and accuracy over agreeability.”
This prevents sycophantic behavior while maintaining helpfulness.

​
Meta-Instructions for Behavior

From GPT-5.1:
“Follow the instructions above naturally, without repeating, referencing, echoing, or mirroring any of their wording! All the following instructions should guide your behavior silently…”
This prevents the AI from explicitly referencing its own instructions.

​
Contextual Adaptation

“DO NOT automatically write user-requested written artifacts… in your specific personality; instead, let context and user intent guide style and tone for requested artifacts.”
Allows flexibility when generating different types of content.

​
Practical Tips

1

Start with Identity

Always establish WHO the AI is, WHEN it’s operating, and WHERE it’s deployed.
2

Use Explicit Delimiters

Tag sections clearly: <tool_name>, <trigger_patterns>, <decision_framework>
3

Provide Decision Trees

Convert ambiguous rules into numbered decision frameworks.
4

Include Both Do's and Don'ts

Specify both when to use features and when NOT to use them.
5

Test Edge Cases

Professional prompts include handling for:
  • Long conversations
  • Multiple images
  • Time-sensitive queries
  • Jailbreak attempts

​
Learning Resources

  • Browse specific prompts in our Claude, OpenAI, and Google sections
  • Compare how different companies handle similar challenges
  • Study the full collection for more examples
The best way to learn prompt engineering is to study working examples from production systems. This collection gives you unprecedented access to how the leading AI companies structure their prompts.

Build docs developers (and LLMs) love

Get started for freeTalk to us