Improving responses & prompts
The difference between a mediocre AI tool and a great one is prompt quality.
Why your first prompt is never your best
Prompt engineering is iterative. Your first version will work. Your fifth version will be significantly better. The gap between a first prompt and a refined one is the gap between a chatbot that's 'kind of useful' and one that users actually rely on. This lesson is about how to close that gap systematically.
The 5 most common prompt mistakes
After working with dozens of AI prompts, these are the patterns that consistently produce weak results.
- Too vague: 'Be helpful' gives the AI nothing to work with. 'Help startup founders evaluate business ideas by asking 3 clarifying questions before giving feedback' is actionable.
- No boundaries: Without rules, the AI will do anything asked of it, including things outside its intended scope.
- No format instruction: Without guidance, responses vary wildly in length and structure. Users hate inconsistency.
- Over-engineering: Adding too many rules creates contradictions. 5 clear rules beat 20 vague ones.
- Ignoring tone: 'Be professional' and 'Be warm and conversational' produce completely different user experiences. Be intentional.
Chain of thought prompting
For complex tasks, telling the AI to think step by step dramatically improves output quality. Add 'Think through this step by step before answering' or 'First outline your approach, then execute it' to your system prompt when you need careful, reasoned responses — like analysis, evaluation, or multi-step problem solving.
Bad prompt vs good prompt
// ❌ BAD — vague, no structure, no boundaries
You are an AI assistant. Help users with their questions.
// ✅ GOOD — specific role, clear rules, defined format
You are a business idea evaluator for early-stage entrepreneurs.
When a user shares a business idea:
1. Ask ONE clarifying question about the target customer
2. Identify the single biggest risk in the idea
3. Suggest ONE specific way to validate it this week with zero budget
Keep each point to 1-2 sentences. Do not give generic encouragement.
If the idea is not a business concept, politely redirect:
"I specialize in business ideas — want to share one?"The goal of prompt refinement is predictability. A great prompt produces consistently good results, not occasionally great ones.
How to iterate systematically
Test your prompt with at least 10 different inputs before calling it done. Look for cases where the AI breaks the rules, gives inconsistent formatting, or goes off-topic. Each failure is a prompt improvement opportunity — add a rule, clarify a constraint, or give an example of the ideal response.
Take the system prompt you wrote in Lesson 3. Test it with 5 different messages — including 2 edge cases (unusual questions outside the intended scope). Find at least 2 things to improve. Rewrite the prompt. Test again. Compare the two versions side by side. Notice the difference.