Master the Prompt 2.0: Instructions for a Thinking AI

Reading time: approx 9 min

In the previous lesson, you learned that ChatGPT-5 has an advanced "thinking" mode. Now it is time to learn how to become the best director for this powerful actor. A well-formulated prompt (instruction) is the difference between a generic, unusable answer and a tailored, brilliant result that saves you hours of work.

Prompting a modern AI is less about asking a question and more about giving a clear work order.

What you will learn

• Why you must move from a simple wish to a detailed instruction.
• A simple but powerful mnemonic (the CRAV method) for building perfect prompts.
• Advanced techniques for getting the AI to reason in multiple steps and explain its thinking.
• How you iterate and refine your prompts to get exactly the result you want.

The Fundamentals: From wish to instruction

A common pitfall is treating the AI like a search engine. A vague wish gives a vague result.

• Vague wish: Write a lesson about the Viking Age.
• Result: A dry, general text that you could just as easily have found on Wikipedia. It lacks adaptation for your students, your purpose, and your pedagogy.

To activate the AI's "thinking" mode, you need to give it a detailed instruction that contains all the necessary puzzle pieces.

• Detailed instruction: Act as an experienced social studies teacher for grade 5. Create an engaging introduction to a lesson about Viking Age trade. The introduction should be a short, dramatic story (max 150 words) about a young Viking traveling to Miklagård. The purpose is to awaken the students' curiosity.
• Result: A creative, audience-adapted, and directly usable lesson material.

Practical Examples: The CRAV method in practice

To systematically build powerful prompts, you can use the CRAV method. It is a simple mnemonic that ensures you include all critical information.

• Context: Give the AI all the background information it needs.
• Role: Give the AI a specific expert role to play.
• Action: Define exactly what the AI should do (the task).
• Values: Set up frames, constraints, and format requirements.

Let us see how we can build a prompt with the CRAV method, step by step:

Goal: Create an exercise about source criticism around a current topic.

Finished CRAV prompt:

Act as an expert in media and information literacy for high school. I am a high school teacher in social studies and my students in grade 11 will work with source criticism.

Your task is to create two short texts (about 100 words each) about a fictional proposal to introduce "citizen points" in Sweden.

Values:
1. Text 1 should be written from a biased, populist perspective with strong emotional arguments.
2. Text 2 should be neutral, fact-based, and cite a fictional government report.
3. End with three open questions that help students analyze the differences in the texts' credibility.

Implementation in the classroom: Advanced techniques

When you master the CRAV method, you can start using even more advanced techniques.

1. "Think step by step" prompting Instruct the AI to give a step-by-step explanation in the answer. This forces it to structure its answer logically, which is excellent for modeling problem solving. It is important to note that ChatGPT does not show its internal thought logic, but this technique gets it to present its reasoning in a clear way, often with numbered sub-steps or brief "why" justifications.

• Example: I want you to compare the Swedish and American political systems. Do this by following these steps: 1. First briefly explain the basics of a parliamentary democracy. 2. Then explain the basics of a presidential system. 3. End by listing the three most important differences in a table.

2. "Take a step back" prompting Ask the AI to first explain the overarching principles before it solves the specific task. This is fantastic for helping students understand why a method works.

• Example: I want my students to learn to calculate the area of a triangle. Before you give me the formula and a calculation example, take a step back and first explain why a triangle can be seen as half of a rectangle. Use a simple analogy.

3. Explicit invocation of "Thinking" mode As we mentioned in the previous lesson, do not be afraid to be overly clear. If the task is extra complex, add a phrase like:

• Think hard about this and use your deepest reasoning ability.
• Or select GPT-5 Thinking directly in the interface if you have a paid plan.

These methods are fully in line with OpenAI's official recommendations to ensure that the model's most capable part is used for complex tasks.

Next steps

You now have a powerful toolbox for writing instructions that get the absolute best out of ChatGPT-5. In the next lesson, 'New Tools in Practice: "Study Mode" and "Personalities"', we will see how we can combine these prompt skills with the new, built-in features that OpenAI has tailored specifically for learning.