The difference between a mediocre and an excellent AI result almost never comes down to the model — it comes down to the prompt. Prompt engineering is the skill of giving AI models the right instructions. Master it, and you multiply your productivity.
📖 Definition: A prompt is the input you give an AI model — a question, an instruction, or a complex template. The prompt largely determines the quality, relevance, and usefulness of the AI's response.
The difference in practice is dramatic:
❌ Weak prompt:
"Write me something about marketing"
✅ Professional prompt:
"Create a 5-point checklist for planning a social media campaign for a B2B SaaS company with 50 employees. Target audience: IT decision makers in the DACH region. Tone: professional but approachable. Format: Numbered list with 2-3 sentences of explanation each."
The second example typically delivers a usable result on the first try. The first requires multiple follow-up questions.
| Element | What It Does | Example |
|---|---|---|
| 🎭 Role | Defines expertise and perspective | "You are an experienced B2B SaaS marketing consultant" |
| 📋 Context | Provides relevant background information | "Our company has 50 employees and sells HR software" |
| 🎯 Task | Precisely describes what should be done | "Create a campaign checklist with 5 points" |
| 📐 Format | Specifies the desired output structure | "As a numbered list with 1-2 sentences each" |
💡 Tip: Not every prompt needs all 4 elements. For simple questions, a clear task is enough. But the more complex the task, the more elements you should use.
Complete example with all 4 elements:
[Role] You are a Senior Data Analyst with experience in the
logistics industry.
[Context] Our company delivers 2,000 packages daily in the
DACH region. Average delivery time is currently 3.2 days.
[Task] Analyze the following delivery data from the past 6
months and identify the top 3 causes of delayed deliveries.
[Format] Present results as:
1. Executive Summary (3 sentences)
2. Causes table (Cause | Frequency | Recommendation)
3. Immediate action for each cause
| Anti-Pattern | Problem | Better Alternative |
|---|---|---|
| "Make it good" | No clear quality definition | Specify concrete criteria |
| "Be creative" | Too vague, unpredictable results | "Create 3 unconventional variants" |
| Everything in one prompt | Overwhelms the model | Break into steps (prompt chaining) |
| No length constraint | Response too long or too short | "Max 200 words" or "5 bullet points" |
| Double negation | Confusing for the model | Phrase positively |
⚠️ Caution: The most common anti-pattern is missing context. The AI knows nothing about your area of work, industry, or target audience — you need to actively provide this information.
Prompt engineering is not a one-time action. The best results come from systematic refinement:
🔑 Remember: A prompt that works well once works well every time. Invest the time in optimization — it pays off with every reuse.
🎯 Exercise: Take a task you do regularly and write a prompt using all 4 elements. Test it and refine it through 3 iterations.
Next lesson: Role Prompting and System Prompts — how to assign expertise to the AI.
Welche 4 Elemente gehören zu einem guten Prompt?