Prompt Optimization Guide

Reduce token count without sacrificing quality.

Before vs. After

// Before: 156 tokens
const badPrompt = "You are a helpful assistant. Please help the user with their question. Be friendly and thorough in your response. Make sure to consider all aspects of the question and provide a comprehensive answer."

// After: 23 tokens
const goodPrompt = "You are a concise technical assistant."

Result: 85% token reduction

Key Techniques

• **Remove filler phrases** - "Please", "I'd like you to", "Could you"
• **Use direct instructions** - "Summarize" not "Can you summarize"
• **Limit examples** - 1-2 few-shot examples, not 5
• **Be specific** - "List 3 reasons" not "List some reasons"

Testing Protocol

• Create A/B test with original vs. optimized prompt
• Measure quality metrics (accuracy, relevance)
• Compare token costs
• Document winning variations

Savings Calculator