Prompting for Better Answers or Better Thinking?

Much of the advice about AI prompting focuses on how to get better answers. Be specific. Provide context. Assign the AI a role. Describe the desired output. Refine the prompt when the response is not what you expected.

These are useful techniques. But in education, we need to start with a different question:

What thinking should the learner be doing, and what role should AI play in supporting that thinking?

I would be hesitant to teach prompting primarily as a collection of techniques for getting better answers from AI. A technically excellent prompt can produce an excellent answer while inadvertently eliminating the very reasoning or problem solving we want a student to develop.

Consider a student learning computer programming. A prompt such as:

Write a program that solves X and explain the code.

might produce exactly what the student requested. From a prompt-engineering perspective, adding more context, specifications, examples, and output requirements could make the resulting code even better.

But is producing better code the learning objective?

If the objective is for the student to learn how to analyze a problem, develop an algorithmic approach, anticipate edge cases, debug errors, or explain design decisions, improving the AI-generated solution could actually reduce the student's opportunity to practice the intended skills.

That suggests a different way to think about prompting.

Prompting as the design of a human-AI interaction

Rather than treating prompting primarily as a formula for communicating instructions to AI, it can be approached as an iterative process for designing a productive human-AI interaction.

At a high level, that process might look something like this:

1. Define the goal

What is the learner trying to accomplish, and what knowledge, reasoning, or problem solving does the task require?

This comes before deciding what to ask AI. If the learning goal is unclear, it is difficult to determine what an appropriate role for AI should be.

2. Decide what thinking should remain with the learner

This may be the most important step.

The learner and/or teacher determines which cognitive work is important for the learner to perform and what role AI should be allowed to play.

Depending on the learning objective, AI might provide information, structure thinking, ask questions, offer feedback, challenge reasoning, or collaborate more extensively.

For example, if the objective is learning to formulate an algorithmic approach, asking AI to generate the approach defeats much of the purpose. Asking AI to question an approach the student has already developed may support the same learner very differently.

3. Construct the prompt

Only now do the familiar elements of prompt construction enter the process.

Provide the AI with the goal, relevant context, its intended role, appropriate constraints, expectations for the interaction, and the desired form of its response.

For example:

I developed the following approach to solving this problem. Review my proposed solution. Don't write the code for me. Ask me questions that will help me identify logical problems or cases I may have overlooked.

This is not necessarily a more technically sophisticated prompt than asking AI to solve the problem. But it creates a very different learning interaction.

4. Interact and iterate

The initial prompt should be a starting point, not necessarily a one-time transaction.

Learners can examine the response, clarify their thinking, challenge the AI, redirect it, ask follow-up questions, test alternatives, and refine the interaction.

This is where I see a particularly strong connection to design thinking and the iterative orientation of the DOT Framework. Prompting becomes less about discovering the perfect wording in advance and more about designing, testing, evaluating, and revising an interaction to accomplish a particular purpose.

5. Evaluate

Evaluation should address more than whether the AI produced a good answer.

• Was the information accurate?

• Was it relevant?

• Were its recommendations appropriate?

But in a learning environment, we also need another level of evaluation:

• Did the interaction support the intended learning?

• Did the learner retain responsibility for the thinking that mattered?

• Did AI help expose or extend the learner's reasoning? Or did it quietly take over cognitive work that the learner needed to practice?

6. Refine

Sometimes evaluation tells us that the prompt itself needs improvement.

In other cases, something more fundamental needs to change: the role assigned to AI.

If AI is doing too much of the reasoning, adding more detailed instructions may not solve the problem. The better design decision may be to ask AI to question, coach, critique, provide feedback, or introduce a contrasting perspective instead of producing an answer.

The role of AI can change during the learning process

There is another implication that is particularly important.

The appropriate role for AI does not have to remain constant throughout a task.

Consider the stages of solving a programming problem. Early in the process, AI might help a learner clarify requirements without proposing a solution:

Here is how I interpret the problem. Ask me questions that would help me determine whether I have overlooked any requirements or constraints.

Once the learner develops an approach, AI might shift into a critical role:

Here is the approach I am considering and why. Ask me questions that will help me test my reasoning. Don't recommend a solution yet.

During debugging, the interaction could change again:

My code produces X, but I expected Y. Here is my hypothesis about what is causing the problem. Help me test my hypothesis without fixing the code for me.

And after completing the program, AI could become a skeptical reviewer:

Ask me questions that require me to explain and defend the major design decisions I made in this program.

The learner is "using AI" in every example. But the nature of that use, and the thinking required of the learner, changes substantially.

A different measure of a good prompt

This leads to a different way of thinking about what constitutes a good prompt in education.

A good prompt should certainly help produce a useful AI response. But that cannot be our only criterion.

We also need to ask:

What thinking does this interaction require from the learner?

What role has been assigned to AI?

Does that division of cognitive work align with what we actually want the learner to learn?

Perhaps that is the more important form of prompt literacy for education.

The goal is not simply to teach learners how to get increasingly sophisticated answers from AI. It is to help them make intentional decisions about when AI should answer, when it should question, when it should challenge, when it should collaborate, and when the learner should do the thinking without AI assistance.

These questions connect to a broader area I have been exploring in my work on AI-mediated learning: understanding what actually happens in the interaction between a learner and AI, how responsibility for cognitive activity is distributed, and how we might intentionally design those interactions to support learning.