5 Common Prompt Engineering Mistakes to Avoid

Key takeaways: Five common prompt engineering mistakes significantly reduce LLM output quality and can be corrected immediately. First, vague prompts produce vague results: always specify word count, tone, target audience, and subject scope. Second, failing to provide examples wastes the power of few-shot prompting, which drastically improves output reliability especially for specific formats like JSON. Third, neglecting chain-of-thought reasoning by not asking the model to think step by step leads to hallucinations on complex logic problems. Fourth, skipping persona assignment means missing the vocabulary and tone calibration that comes from giving the AI a specific expert role. Fifth, iterating endlessly in the same chat pollutes the context with previous errors; instead, edit the original prompt or start a new conversation. Prompt engineering is an iterative skill requiring refinement, testing, and documentation of best-performing prompts. Ikasia offers hands-on AI training covering these techniques for managers and technical teams across French enterprises.

1. Being Too Vague

• Bad: "Write an article about AI."
• Good: "Write a 500-word blog article about the impact of generative AI on the marketing sector in 2025. Adopt a professional but accessible tone. Target an audience of CMOs." Rule: The more context you give, the better the result.

2. Not Giving Examples (Zero-Shot vs Few-Shot)

LLMs learn by example. If you want a specific format (e.g., a JSON list), give an example of the expected output.

• Technique: This is called "Few-Shot Prompting". It drastically improves reliability.

3. Not Asking to "Think" (Chain of Thought)

For complex logic problems, if you ask for the answer directly, the model may hallucinate.

• Tip: Add "Let's think step by step" or "Explain your reasoning before giving the answer". This forces the model to decompose the problem.

4. Neglecting the "Persona"

Assigning a role to the AI helps it select the right vocabulary and tone.

• Example: "You are an expert senior developer in Python..." vs "You are a primary school teacher..."

5. Iterating on the Same Chat

If the conversation goes wrong, don't persist in the same thread. The context is "polluted" by previous errors.

• Tip: Edit your previous prompt or start a new conversation (New Chat) to reset the context.

Conclusion

Prompt Engineering is an iterative skill. Don't be satisfied with the first draft. Refine, test, and document your best prompts.

Enjoyed this article? Check out our AI for Managers Training — 2 days to drive AI strategy across your organisation.