AI tutoring systems can generate mathematically correct solutions while completely missing the conceptual understanding a student is trying to build. A new study examining how students with learning differences interact with artificial intelligence in mathematics classrooms reveals that the real problem with AI-assisted learning isn't accuracy or access, but alignment between what the AI explains and how the student thinks.

What Happens When AI Answers Correctly But Teaches Poorly?

Researchers conducting a practitioner study at a secondary school in southern England observed two students working with generative AI on mathematics problems. In one case, a 15-year-old student with lower prior mathematical attainment used AI to solve a linear cost model problem. The AI provided a technically correct algebraic explanation, but the student couldn't connect it to meaning. The student wanted to interpret the expression as repeated addition, not as abstract algebra. The AI's correct procedure had become useless because it didn't match the student's reasoning process.

In another instance, a high-attaining A-level student asked a conceptually sophisticated question about when it's valid to divide by an algebraic expression that could equal zero. The AI responded with an efficient solution that bypassed the underlying mathematical concern the student was investigating. The answer was correct, but the interaction had drifted away from what the student actually needed to understand.

Why Is This an Inclusion Problem, Not Just a Teaching Problem?

AI tutoring tools appear inclusive because they're available on demand and generate explanations instantly. But availability doesn't guarantee meaningful participation. Students may use AI without making sense of its output, which matters significantly in mathematics, where learners interpret symbols, process language, and draw on prior knowledge in different ways.

The research identifies a critical gap: inclusion isn't simply whether AI provides support, but whether the interaction enables learners to build usable meaning. An explanation that's clear to one student may be too dense, too fast, or misdirected for another. This is where current AI tutoring systems struggle most.

What Teachers Do That AI Cannot Yet Replicate

The study found that teachers perform interactional work that AI systems consistently fail to match. These include:

• Slowing the pace: Teachers can pause explanations to let understanding develop, while AI often rushes toward procedural completion.
• Highlighting key ideas: Teachers draw attention to the concepts that matter most for a particular student's learning journey.
• Connecting to prior knowledge: Teachers link new explanations to what students already understand, building bridges between old and new ideas.
• Allowing flexible reasoning: Teachers permit students to solve problems in ways that make sense to them, even if unconventional.
• Reopening conceptual issues: Teachers recognize when dialogue has moved too quickly to closure and restart the conversation around the core concept.

In essence, teachers aren't just explaining mathematics; they're making explanations usable. This distinction may be the most important finding in how AI and human instruction should work together.

How Can Students Use AI as a Learning Partner Rather Than a Shortcut?

Beyond the mathematics classroom, research on university-level learning suggests that student mindset fundamentally shapes whether AI becomes a tool for growth or a crutch for avoidance. Psychologist Carol Dweck's research on motivation distinguishes between two approaches: students with a growth mindset believe intelligence and skill develop through effort, strategy, and feedback, while those with a fixed mindset see ability as unchangeable.

Students with a fixed mindset often avoid AI out of fear of failure, or over-rely on it to bypass productive struggle. In both cases, learning becomes passive. Students with a growth mindset, by contrast, engage with AI to develop skills, explore ideas, and test strategies with curiosity rather than fear.

Educators can foster this growth-oriented approach by positioning AI as a feedback partner, not an answer machine. Rather than asking an AI chatbot to fix or complete work, students can be trained to request targeted feedback. For example, instead of "fix this paragraph," a student might ask: "Read this paragraph and tell me one strength, one weakness, and one way to improve cohesion. Do not rewrite the whole paragraph for me." This shifts the interaction from passive consumption to active learning.

Steps to Help Students Develop Reflective AI Use

• Teach students to use AI for targeted practice: Rather than seeking answers, students should use AI to create practice exercises based on instructor feedback. For instance, if feedback highlights weak use of hedging language in academic writing, a student can ask AI to generate sentences using words like "may," "might," "suggests," and "could" for practice and revision.
• Build metacognitive reflection into AI sessions: At the end of each AI interaction, students should summarize how they used the tool, what they understood or improved, which ideas came from their own thinking, and where they could improve. This reflection helps students become aware of whether their AI use was responsible and effective.
• Embed reflection prompts into custom chatbots: Instructors who create course-specific AI tools can write reflection instructions directly into the chatbot's configuration. Students can then type a simple phrase like "end summary" to trigger automatic assessment of their AI use during the session, along with tips and links to additional support resources.

One study on AI-assisted learning environments found that access to AI alone does not automatically increase learner autonomy, suggesting students still need explicit support to develop reflective and growth-oriented learning habits.

What Does This Mean for How Schools Should Implement AI?

The emerging picture suggests that the central question isn't whether AI works, but for whom it works, under what conditions, and what kinds of learning it makes possible. Early findings from the mathematics study, though based on observations of just two students, point to something educators have long known: the quality of interaction matters more than the quality of the answer.

This has profound implications for how schools adopt AI tutoring. If AI systems are deployed without attention to alignment between the tool's explanations and each student's reasoning process, they may appear to work while actually undermining deeper understanding. Conversely, when AI is positioned as a reflective partner that students learn to use strategically, it can support resilience, experimentation, and long-term academic success.

The challenge ahead isn't building smarter AI tutors; it's building smarter ways for students and teachers to work with the AI tools that already exist.