The Prompt Generation: How Do We Teach Young People to Ask Deep Questions Instead of Consuming Ready-Made Answers?

Posted on Oct 24 • Originally published at scouthub.dev

A fourteen-year-old sat down to do homework about a polluted river near his home. He typed into the chat window: "Write me an essay about river pollution." A few seconds later he had five tidy paragraphs. He handed them in the next day. His teacher asked: what surprised you most while writing this? He had no answer.

This scene is no longer rare. In the United States, 26% of teens aged 13 to 17 said they had used ChatGPT for schoolwork in 2024, double the 13% of 2023. In our own Arab homes and schools we see the same trend, even if we don't yet have precise numbers.

The problem isn't the tool. The problem is the question we put to it, and what we do once the answer arrives.

At the MIT Media Lab, 54 participants wrote essays over several sessions. One group used a language model, one used a search engine, and one wrote with their own brains only. The researchers measured brain activity with EEG. The weakest neural connectivity was in the language-model group, and many of them couldn't accurately quote what they had written themselves minutes earlier. The researchers called this "cognitive debt." The sample is small and the findings are preliminary, but they look a lot like what we see in classrooms.

In a study by Microsoft and Carnegie Mellon University, 319 knowledge workers described 936 real situations from their work. The researchers found a clear pattern. The more confidence people had in the AI, the less critical thinking they did. The more confidence they had in themselves for the task, the more critical thinking they did.

This matters for education. A child who doesn't trust their own ability will hand their mind over to the machine easily. So building self-confidence isn't only an emotional matter. It's a condition for thinking.

And that study looked at adults in their jobs, not children in school. If an experienced professional relaxes in front of the tool, what about a child who has no experience yet to measure it against?

We don't want to say that AI is the enemy of learning. The evidence says something more precise.

In an experiment with nearly a thousand high school students in Turkey, researchers gave one group a standard version of GPT-4 and another group a version designed as a tutor that doesn't hand over the solution. During practice, the first group's grades improved by 48% and the second by 127%. But when the tool was taken away for the exam, the first group did 17% worse than students who had never used it. The "tutor" group was largely unharmed.

At Harvard, physics instructors designed an AI assistant that keeps its replies short, reveals one step at a time, and asks the student to try first. Students who learned with it gained more than twice as much as in the classroom session, in less time.

The difference between the two experiments isn't the power of the model. It's the design: does the tool give the answer, or hand the question back to its owner?

We named one of our projects "Talqeen Dev." The name carries a deliberate irony. In our school memories, talqeen means having the answer dictated to you so you can memorize it. But the tool does the opposite. It takes a developer's raw idea in Arabic and turns it into a plan and small milestones. To work well, the developer has to be clear about what they want, for whom, and within what limits. In other words, it teaches them how to ask, not what to memorize.

That's what we mean by "meaning engineering." A good question isn't a trick for squeezing a better answer out of a machine. It's practice in organizing an idea before asking for anything. When you write a clear request, you are really analyzing the problem: what are the givens? What do I already know? What's missing? How will I know the answer is right?

The weak question: "Write me something about river pollution."

The deep question: "There's a small river in my area that turns dark in summer. What are three possible causes? How can I check each cause myself with simple tools? Don't write the essay for me. Help me organize my thoughts."

The second question doesn't need genius. It only needs the boy to have thought before typing. That's where real learning begins.

Another example, from the world of leaders. The weak question: "Give me a game for scouts." The deep question: "I have twelve boys aged 11 to 13, in a small schoolyard, and we want a game that trains communication inside the patrol. What could go wrong, and how will I know it met its goal?" The first question gets you an activity. The second makes you understand your activity.

So why is rote learning still so strong? Because our exam systems reward the right answer, not the path to it. Students are smart. They learn quickly what gets rewarded. If the exam only asks "what?", nobody will bother with "why?" and "how?". AI didn't create this problem, but it exposed it. A tool that remembers everything has made memorizing alone almost worthless.

For more than a century, the scout method has rested on learning by doing. We don't explain to young scouts how to pitch a tent and then give them a written test. We hand them the tent and the ropes, let them make mistakes, and then ask: why did it fall? The patrol system makes a small group discuss, decide, and own the result. That's a natural environment for asking questions.

When we built the game "ScoutQuest" (درب الكشاف), we put in 1,056 questions across eight categories, with daily challenges and one-on-one duels. We should be honest: a quiz game can end up training recall only. "What color is the Cubs' scarf?" is a memory question. "Your patrol is lost in the woods with one hour of daylight left. What are your first three steps?" is a question that makes a child think and set priorities. The challenge we're still working on is adding more of the second kind.

In "Najem," our AI assistant for scout leaders, the prompt generator doesn't just ask "what do you want?" It asks the leader about the goal, the age stage, the number of participants, and the place. A leader who answers those four questions has already done half the planning before seeing any answer. We've noticed some leaders now ask themselves these questions even without the tool.

Global numbers say these skills aren't a luxury. In the PISA 2022 creative thinking assessment, in which 64 countries took part, 22% of students across OECD countries did not reach the baseline level. In the World Economic Forum's Future of Jobs Report 2025, analytical thinking remained the top core skill for employers, at 69%. And in 2024 UNESCO published an AI competency framework for students with 12 competencies across four dimensions and three levels. Its common thread is that students should understand the tool and question it, not just use it.

We don't need new curricula to start. We need small habits, repeated:

We don't want a generation that is only good at writing commands for machines. We want a generation that understands that writing a good request is, at heart, a way of thinking: defining the goal, understanding the context, doubting the first answer, and going back to the question.

What's asked of us is simple: slow down a little before the answer, and ask our children "how did you get there?" more often than "what's the answer?"

Answers have become cheap. A good question still needs a person who thinks.