Chapter 1: Common Prompt Failures
Why Prompts Failβ
| Failure Type | Cause | Solution |
|---|---|---|
| Hallucination | Model invents facts | Ask for sources, verify claims |
| Off-topic | Vague instruction | Be more specific |
| Wrong format | No format specified | Include format requirements |
| Too verbose | No length constraint | Add word limits |
| Wrong tone | No tone guidance | Specify tone explicitly |
| Incomplete | Missing requirements | Use checklists |
Hallucinationsβ
AI can confidently state incorrect information.
Prevention Strategiesβ
"If you're uncertain about any facts, explicitly state
your uncertainty level (high/medium/low confidence).
Do not invent statistics or citations. If you don't
have reliable information, say 'I don't have verified
information about this.'"
Off-Topic Responsesβ
When AI goes in unexpected directions:
Before:
"Write about marketing"
After:
"Write about email marketing specifically for B2B
SaaS companies with $1-10M ARR. Focus only on:
- Subject line optimization
- Send timing
- Segmentation strategies
Do NOT cover: social media, paid ads, or content marketing"
Format Issuesβ
When output structure doesn't match needs:
"Format your response EXACTLY as follows:
## Summary
[2-3 sentences here]
## Key Points
- Point 1
- Point 2
- Point 3
## Recommendation
[Single sentence]"
Diagnostic Questionsβ
When a prompt fails, ask:
- Was the instruction clear and specific?
- Did I provide enough context?
- Did I specify the format I wanted?
- Did I set appropriate constraints?
- Is this task within the model's capabilities?
Key Takeawaysβ
Summary
- Identify the failure type first
- Hallucinations: Ask for uncertainty acknowledgment
- Off-topic: Be more specific, define scope
- Format: Provide exact templates
- Iterate based on what went wrong