Prompt Technique: It Is About Clear Instructions

Reading time: approx. 5 min

After exploring both zero-shot and few-shot prompting, it is time to gather all the knowledge and focus on the overall art of writing effective prompts. In this moment, we go through prompt technique, the fundamental principles for formulating clear, precise, and useful instructions to an AI. The goal is for you to be able to design prompts that give you (and your students) the results you want, every time.

1. Be specific and clear

The absolutely most important principle for effective prompt technique is to be as specific and clear as possible. The AI is a task solver, but it cannot read between the lines or guess your intentions.

• Context first: Start your prompt by giving the AI a clear background or role. This helps the model understand the perspective it should take.
  • Example: "As an experienced teacher in grade 8..." or "You are an expert on English history..."
• System messages: In advanced AI interfaces, you can often set system messages that define the AI's basic behavior and expertise. This affects the entire conversation and is especially powerful for setting tone and area of expertise.
• Task clearly defined: State exactly what you want the AI to do. Use verbs like "create", "list", "explain", "compare", "summarize".
  • Example: "...create a lesson plan in three steps for..." or "...list five arguments for..."
• Expected format: If you have a specific wish about how the answer should be presented (for example, bullet list, paragraph, table, dialogue), state this clearly.
  • Example: "...in bullet form", "...as a table with two columns", "...written as a dialogue between two students".

Bad example of prompt: "Tell me about photosynthesis." Why bad? Too vague. What level of detail? For whom? What format?

Better example of prompt: "As a biology teacher, create a bullet list with four key steps in the photosynthesis process for students in grade 7. Use simple and engaging language." Why better? Clear role, specific task, number of points, target audience, and tone.

2. Use roles, tone, and level of detail

To further steer the AI's answer, you can assign it a role, specify a tone, and indicate a level of detail.

• Role: Giving the AI a specific "persona" can dramatically improve the relevance of the answer.
  • Example: "You are an experienced science teacher.", "Act as a pedagogical writer.", "You are a motivating coach."
• Tone: Decide how the answer should sound. Formal, informal, fun, serious, encouraging?
  • Example: "Keep the language simple and motivating.", "Answer with a friendly and encouraging tone.", "Use formal and academic language."
• Level of detail: Tell the AI how deep it should go into the subject.
  • Example: "Give an overview explanation.", "Describe in detail with concrete examples.", "Answer with max two sentences."

3. Structure multi-step tasks

For more complex tasks that require multiple steps, it is often best to break them down into sub-prompts or to ask the AI to follow a specific step-by-step process within a single prompt.

1. Step 1: Gather information/analyze: Ask the model to first gather, extract, or analyze a given text or data.
2. Step 2: Process/Transform: Then let the AI process or transform that information according to your instructions.
3. Step 3: Present/Generate: Finally, ask it to present the result in the desired format.

Example of multi-step prompt:

1. "List five questions based on the following text about English industrialization: [paste text]."
2. "For each question, explain briefly why it is relevant in one paragraph."
3. "Conclude by summarizing the overall conclusions from the questions in three sentences."

Structured prompt templates: For complex tasks, you can use structured templates such as:

• Task: [What should be done]
• Context: [Background information]
• Constraints: [What should not be done]
• Format: [How the answer should be presented]
• Examples: [Any demonstrations]

4. Avoid common pitfalls

Being aware of common mistakes can save you a lot of time and frustration:

• Avoid overload: Trying to cram too many instructions, roles, and constraints into a single sentence can confuse the model. Break it up if it becomes too complex.
• Avoid unclear reference: Avoid vague instructions like "Make it better" or "Continue the same way" without specifying what "it" is or how "the same way" looks. Always be explicit.
• Make no assumptions: Never rely on the model understanding unspoken requirements or implicit rules. If you want something a certain way, say it outright.
• Avoid contradictory instructions: Make sure your instructions do not conflict with each other (for example, "be brief" and "give a detailed explanation").

5. Practical tips for teachers

• Rewrite and refine: Make it a habit to write a prompt, read it critically, and then try to reformulate it at least once for increased precision and clarity. Test different formulations.
• Preview with a "test prompt": If you are going to generate something big, like a whole lesson plan, start with a small test prompt to see if the AI understands your intention and tone. Then adjust before you ask it to generate the full answer.
• Build a template library: Create a collection of effective prompt structures and templates for different types of tasks (for example, for generating test questions, lesson plans, discussion materials, student feedback). This saves time and ensures quality. Share these templates with colleagues and students!

6. Reflection exercise

To consolidate your knowledge of prompt technique:

1. Take a prompt you have used recently, maybe to plan a lesson or create materials. Read through it and consider: Can you make it 30% more specific by adding details about role, tone, format, or other constraints? Rewrite the new prompt.
2. Test both your original prompt and your reformulated prompt in an AI model. Did you get a better, more precise, or useful answer with the more specific prompt? What was the difference?
3. Discuss in the work team: Which instruction elements (role, tone, format, level of detail) are most critical for getting useful answers within your specific subjects and for your student groups? Do you have any "favorite prompts" that you want to share?

Next moment: Temperature and randomness: Why AI sometimes deviates - we will explain how certain settings, like "temperature", affect the AI model's creativity and precision, and why AI sometimes gives unexpected or "random" answers. This helps you understand how you can further control the model's behavior.