A new wave of research reveals that artificial intelligence can dramatically improve student math scores, but only if the technology is designed to make learning harder, not easier. When AI tools step in to solve problems for students, learning collapses. To help children succeed, educators and parents must look for programs that encourage "productive struggle" rather than instant answers.
What Happened
In a randomized trial across eleven schools in Ghana, about 1,000 students in grades 3 to 9 used a WhatsApp-based math chatbot named Rori. Instead of giving answers directly, the chatbot offered hints and guided them toward the correct steps by correcting misconceptions. Students using the chatbot achieved an effect size of 0.36, which is equivalent to nearly an extra year of learning, at a cost of just five dollars per student. This matches data on classroom AI tutors, which shows that structured, dialogue-based tools can successfully mimic high-quality personalized instruction.
The Bigger Picture
But not all educational AI tools are built for teaching. When students use general AI tools as search engines for math answers, they develop what researchers call "metacognitive laziness." A study on AI tools and math motivation shows that effective digital tutors must build student persistence through supportive narrative messaging rather than just giving away results.
When software simply hands over the correct answer, student learning collapses. Homework scores might look perfect, but supervised test scores drop because the student never did the cognitive work. True learning requires physical changes in the brain. Research on productive failure shows that letting students struggle with a problem before explaining it can triple their long-term knowledge retention. This process builds myelin in the brain, which speeds up neural signals.
Implementing these tools in classrooms remains a challenge. A systematic review of intelligent tutoring systems shows moderate academic gains, but success is not guaranteed. In one study, eighth-grade students receiving traditional, teacher-led instruction outperformed peers using ALEKS, an AI tutoring program, because teachers were still learning how to use the software. Another trial comparing original and redesigned tutoring systems found that software changes increased active practice and time-on-task but did not raise final test scores. While some school technology focuses on administrative tracking, like digital hall passes used to monitor student movement, learning-focused AI requires a major shift in how teachers manage and integrate classroom tools.
What This Means for Families
For parents and educators, the easiest application is often the worst for learning. If a homework helper app shows a child how to get the answer in one click, it short-circuits their math development. Parents should choose tools that act as Socratic coaches, prompting the student to explain their thinking. School administrators must also ensure teachers receive thorough training before introducing AI programs to the classroom.
What You Can Do
To start, avoid answer-engine apps. Parents should screen their children's digital tools and replace apps that provide immediate, step-by-step solutions with programs that focus on hints and guided problem-solving.
Next, encourage productive struggle. Remind children that feeling stuck is a necessary part of learning. Praise their effort and persistence through difficult problems instead of focusing only on correct answers.
Finally, ask for teacher integration plans. If a school uses adaptive math software like ALEKS, parents should ask how teachers are trained to support it. The best results occur when teachers actively guide how the software is used, rather than treating it as a digital babysitter.