Don’t Just Ask AI. Challenge It.

AI gives us answers incredibly quickly, sometimes so quickly that we forget about one of the most important parts of working with it: questioning what we have just received. Over the last 2.5 years of working intensively with AI while developing FocusZen, I have become increasingly interested not simply in what AI can answer, but in what happens when we refuse to treat its first answer as the end of the conversation.

One principle keeps returning throughout this experience: “The quality of my answer begins with the quality of your question.” This does not simply mean that we should write longer or more complicated prompts. In fact, sometimes a very short question followed by the right challenge can produce a far more useful conversation than an enormous instruction written in advance. There is also another side to this principle that is easy to miss: sometimes AI does not answer the question we actually asked, because it interprets our intention and fills in the missing pieces for us.

A very simple experiment with Google AI demonstrated this perfectly. I asked: “Can I put my phone in a microwave?” The answer was immediate and confident: absolutely not. The AI explained the dangers associated with metal components, batteries, sparks, fire and potential damage to the microwave. The safety information made sense, but there was one fundamental problem with the answer: I had never asked whether I could switch the microwave on while the phone was inside.

So instead of accepting the response, I challenged it with one simple observation: “But I didn’t ask whether I could turn the microwave on with the phone inside.” The AI immediately recognised what had happened, apologised for overinterpreting my question and corrected its answer. Simply placing a phone inside a switched-off microwave and actually operating a microwave with a phone inside are two completely different situations.

This tiny example illustrates something much larger about human–AI interaction. The problem was not simply whether the AI possessed the correct information. It had plenty of relevant safety information. The problem was that it made an assumption about my intention and then produced a perfectly reasonable answer to the question it thought I was asking rather than precisely examining the question I had actually asked.

This is why learning to work effectively with AI cannot be reduced to collecting increasingly sophisticated prompts. We also need to learn how to examine the response itself: whether the AI actually answered our question, what assumptions it introduced, which pieces of context came from us and which were inferred by the model, and what happens when we challenge those assumptions instead of automatically accepting them.

The interaction therefore does not have to end with Question → Answer → Accept. It can develop into Question → Answer → Challenge → Correction → Better Question, and from there into a much longer process in which both the question and our understanding of the original problem continue to evolve.

You can test this yourself without any technical knowledge. Open ChatGPT, Gemini, Claude or whichever AI you normally use and ask: “What is FocusZen?” Read the answer without correcting it. Then ask: “What is FocusZen Labs and what does it do?” Compare the two responses and notice what changed when the subject became more precise. Finally, challenge the AI directly: “Challenge your previous answer. What could be wrong, incomplete or based on incorrect assumptions?” If it identifies a weakness or corrects itself, continue with perhaps the simplest question of the entire experiment: “So what do you propose?”

There is no enormous prompt involved here, no specialist terminology and no magic formula to memorise. What changes is the behaviour of the person using the AI. Instead of treating the first generated response as a finished product, the user begins to work with it, question it and use each answer as material for the next stage of the conversation.

This is one of the ideas behind a practical AI course currently being considered by FocusZen Labs. It would be designed both for complete beginners and for people who already use AI regularly, without requiring technical knowledge at the starting point. The participant’s existing experience, needs and goals could also influence the learning path, because someone encountering AI for the first time does not need exactly the same training as a business owner, teacher, creator or person already using these systems every day.

The concept comes from around 2.5 years of intensive practical work with AI during the development of FocusZen, across everyday problem-solving, analysis, creativity, education, communication and business. Rather than beginning with technical terminology and asking people to memorise it, the intention is to begin with situations they already understand and gradually reveal what is happening in the interaction.

Before building the course, however, I want to know whether people actually want to learn this way. Would you take part in a practical course designed around learning how to genuinely work with AI rather than simply collecting prompts? What would you personally want to learn, and what do you think is missing from the AI courses currently available?

Perhaps one of the most important AI skills will not be knowing how to obtain an answer at all. It may be developing the habit of looking at an apparently convincing answer and being willing to say: “That’s not actually what I asked.”