Artificial intelligence was supposed to revolutionize personalized learning by being a patient, 24/7 tutor. But a recent study shows that when AI tools refuse to give quick answers and instead force students to think, many children simply stop using them.
What Happened
According to a two-year study by researchers at the University of Toronto, middle school students quickly lost interest in AI tutoring when the technology required actual effort. The National Bureau of Economic Research draft paper tracked students across 18 Tennessee schools who were given access to Khan Academy's Socratic chatbot, Khanmigo. While almost all students tried the AI tutor initially, their usage plummeted when the system refused to provide direct answers and instead prompted them with hints and guiding questions.
The study specifically looked at low-achieving students who were at least one grade level behind in math. While students using the Khan Academy platform showed modest math improvements, the researchers found that the AI tutor itself added no additional academic benefit. As the Hechinger Report noted, the gains were no greater than what previous studies found from standard Khan Academy practice sessions without an AI assistant. Sal Khan, founder of Khan Academy, acknowledged these engagement challenges in a detailed blog post but maintained that early rollouts are essential for refining the technology safely.
The Bigger Picture
This drop-off in student engagement shows a broader challenge with AI in education. When students first gained access to advanced chatbots, many used them as an "answer machine" to bypass critical thinking. A study by the University of California, Irvine, and McGraw Hill revealed that outsourcing homework to AI eroded math skills and reduced active study time. When AI tools are programmed to block these shortcuts, students often choose to stop using them altogether.
However, when AI is tightly integrated into structured environments, the outcomes can change. For example, a study on the NUMI platform showed that while Socratic AI tutoring slowed students down, it led to greater accuracy and better error correction during mastery-based math practice. Similarly, research on the GenAI tool Socratic Mind found that university students achieved significant test score gains and reported high cognitive engagement when forced to explain their reasoning. A study on high schoolers using Edmentum Exact Path showed that AI-based personalized instruction improved immediate achievement and long-term retention compared to traditional teaching alone.
We have previously covered how schools and platforms are adapting to these technologies. As we previously reported, platforms are constantly trying to align digital courses with state standards, while school systems continue to struggle with general-purpose tools since Google opened Gemini to all students.
What This Means for Families
For parents and educators, these findings reveal that technology alone cannot replace human motivation. Students naturally seek the path of least resistance. When faced with "desirable difficulties," the cognitive friction required to actually learn, they will often abandon independent digital tools if given the choice. To make AI tutoring effective, adults must provide the structure and accountability that software cannot. AI tools work best to supplement, rather than replace, active teacher and parent involvement.
What You Can Do
You can start by modeling active learning. When your child uses an AI tutor at home, sit with them for the first few sessions. Show them how to engage with hints rather than treating the chatbot like a search engine to escape work.
Parents should also value mistake-driven progress. Teach children that working through confusion is where actual learning happens. Research shows that slowing down with AI ultimately leads to better test scores.
Finally, choose structured platforms. Look for educational tools that embed AI within a strict, mastery-based curriculum rather than open-ended chatbots where copy-pasting is easy.