Scaffolded and Unscaffolded Generative AI Use in Mathematics Teacher Education: Effects on Pre-Service Teachers’ Instructional Design Competence

Imagine a classroom where the teacher is not just a person, but a powerful computer program capable of writing lesson plans, creating math problems, and explaining complex ideas in seconds. This is the promise of generative artificial intelligence, a technology that has rapidly entered schools and universities. But there is a catch. If a student simply asks the computer for an answer and accepts it without question, they might learn to rely on the machine rather than developing their own ability to think, plan, and solve problems. This is the central tension facing teacher education today: how to use these smart tools to help future teachers learn, without letting the tools do the thinking for them. The key to solving this puzzle lies in a concept called scaffolding. In education, scaffolding is like the temporary support beams used to build a house; they hold up the structure while the builders learn their craft, and are removed once the building is strong enough to stand on its own. The question researchers asked was whether giving future math teachers this kind of structured guidance while they used artificial intelligence would make them better teachers than if they were left to figure it out on their own.

A team of researchers in Ghana set out to answer this question by working with seventy-six future mathematics teachers. These students were enrolled in a course dedicated to learning how to design math lessons, a critical skill for any educator. The researchers split the class into two groups. One group used the artificial intelligence tool with no special instructions; they were told to use the technology to help them plan their lessons, but they had to figure out how to ask the right questions and check the answers themselves. The other group used the same technology, but they were given a specific set of supports. These supports included examples of how to write good questions for the computer, checklists to verify the computer's work, and guided questions that forced them to think about whether the lesson plan made sense for their students. The goal was to see if this extra guidance changed how well the students could design a lesson, how ready they felt to use technology, how well they could manage their own learning, and how critically they could evaluate the information the computer gave them.

After fourteen weeks of working on lesson plans, the results were clear. The students who received the structured guidance performed significantly better than those who worked without it. When the researchers looked at the final lesson plans created by the students, the group with the supports had designed more effective and thoughtful lessons. They were better at organizing the material, choosing the right activities, and creating ways to test if the students understood the math. The group without the supports produced plans that were often rushed or required heavy editing because they had accepted the computer's output too quickly. The guided group also scored higher on measures of their readiness to use technology in the future. They felt more confident in their ability to talk to the computer, understand its limitations, and use it responsibly. They were less likely to be fooled by errors or biased information generated by the software.

The study also revealed that the guided students learned how to learn better. They showed a much stronger ability to regulate their own learning process. Instead of just typing a request and waiting for a result, they paused to plan their approach, monitored the computer's output as it came in, and evaluated whether the suggestions matched their goals. They treated the computer as a partner to be questioned rather than a machine to be obeyed. This critical thinking was the most distinct difference between the two groups. The students with the supports constantly asked themselves if the lesson ideas were accurate and appropriate for real children. They checked the math, compared the suggestions to their textbooks, and revised the plans when the computer made a mistake. The students without the supports often accepted the first answer they received, even when it was flawed, because they lacked the tools to verify it.

When the researchers interviewed a smaller group of students to understand their experiences, the pattern became even more vivid. The students who had the supports described a process of deep thinking. They explained that the checklists and guided questions forced them to slow down and consider the "why" behind every activity in their lesson. They realized that the quality of the lesson depended on their own decisions, not just the computer's output. In contrast, the students without the supports admitted they often felt lost or uncertain when the computer gave a strange answer. They tended to start over with a new chat rather than fixing their original question, and they frequently prioritized speed over accuracy. They felt that the computer was doing the work for them, which left them feeling less prepared to teach on their own.

The findings suggest that simply handing future teachers a powerful artificial intelligence tool is not enough to make them better educators. The tool is only as good as the guidance surrounding it. When the researchers provided a framework that encouraged planning, checking, and reflection, the students used the technology to sharpen their professional skills. They learned to design better lessons, think more critically about the information they received, and take control of their own learning. Without that framework, the technology risked becoming a crutch that prevented them from developing the independent judgment they would need in a real classroom. The study concludes that for artificial intelligence to truly help train the next generation of teachers, it must be used with intentional support that teaches students how to think, not just how to ask.