Why AI Tutoring Apps Are Losing Students (and What Learners Need Instead)

When AI entered education, it promised every student a personalized tutor available around the clock. Soon, EdTech startups and legacy platforms rushed to launch AI-powered chatbots, each pitched as a private tutor in every student’s pocket.

But as the hype wears off, EdTech platforms face a harder truth: students are abandoning pure AI tutors. Left to interact with software alone, students disengage quickly, avoid hard work, and fail to build long-term learning habits.

In The Early AI Wave, Chatbots Were Built to Replace Tutors

Following the AI boom, many platforms rushed out student-facing AI tools. One of the most closely studied examples is Khanmigo, Khan Academy’s AI tutor. Researchers tracked how students actually used it over two years (source), and the results were interesting:

Image created for illustration purpose. Source of data: EdWorkingPapers’s study on AI tutoring.

• Interest does not mean use: 96% of students tried the AI tutor at least once, but it accounted for only 14% of their practice sessions.
• Students avoid active engagement: Even after getting a question wrong, students messaged the AI only 17% of the time.
• Shortcuts replace learning: Nearly 40% of student replies were bare answers or single numbers, 24% were simple clicks on auto-suggested prompts like “Help me solve this,” and only 14.5% contained an actual written question or math step.
• Resistance to guidance: When the AI was programmed not to hand over direct answers, more than 12% of the students sent low-effort messages like “idk” or demanded direct answers, and another 10% shifted off-task to ask for jokes.

Without human guidance, students ignore even well-designed pedagogy. Access to software is not the same as meaningful learning.

What The Research Says Actually Helps

To understand why pure AI tutors underperform, researchers at Stanford’s SCALE Initiative evaluated AI tutoring models based on relational intensity: the depth and consistency of the human connection between student and tutor (source). Their findings suggest AI tutoring isn’t one thing, but a spectrum:

• The Spectrum of Use: AI can act as a simple answer engine, a student learning tool, or an assistant that supports human tutors.
• The Role of Human Guidance: AI tools consistently failed to improve learning when operating entirely without human guidance.
• The Key Finding: The strongest learning outcomes came from AI designed to support the human tutors, not replace them or interact with students solo.

AI is most effective when it strengthens the human relationship at the center of learning, not when it stands in for it.

EdTech is Recognizing That AI Belongs Behind The Scenes

While some platforms continue experimenting with fully autonomous student-facing apps, a growing share of the industry is moving AI into a supporting role: handling adaptivity and data while leaving direct mentorship and coaching to humans.

Global math tutoring program, Cuemath, is one example. Rather than replacing teaching with a chatbot, Cuemath uses AI in the background to enhance self-paced practice while keeping 1:1 coaching central to the experience, tracking learning history, flagging skill gaps, and building personalized practice using techniques like spacing and interleaving, so tutors can spend their time on mentorship and targeted guidance instead of manual tracking.

This model is an answer to what the recent studies exposed: when AI is the student’s primary point of contact, they take shortcuts, disengage, or resist guidance. By keeping a human tutor as the central relationship and keeping AI for the background work, this model resolves the exact failure point the research identified.

Although, it might be fair to argue that keeping a human in the loop sacrifices the scalability that made AI tutoring appealing in the first place. But scalability without engagement is a hollow win: a tool 96% of students try but only 14% actually use isn’t scaling learning, it’s scaling abandonment.

Identifying the Blind Spot

Cuemath’s model also points to a broader gap. If the industry keeps scaling what AI already does well, it risks preparing students for tasks software already automates. Students need to build cognitive skills AI can’t replicate:

• Strategic Reasoning & Logic: Knowing how to approach an unfamiliar problem, test assumptions, and evaluate alternative strategies.
• Error Analysis & Synthesis: Identifying why a solution failed and using intuition to adjust course.
• Real-World Judgment & Creative Problem-Solving: Applying abstract concepts to messy, unstructured scenarios where there is no pre-programmed algorithm.

To target these skills directly, Cuemath app introduced an Enrichment Math section: logic puzzles like KenKen alongside real-world applied problems, such as weighing trade-offs in a budget plan, that require students to reason through several steps rather than recall a memorized procedure.

Enrichment math section in the Cuemath App. Source: Cuemath

The Lesson for the EdTech Industry

AI is an extraordinary tool for scale, data analysis, and personalized practice, but it’s a poor primary motivator on its own. The Khanmigo results are a caution against assuming that access to AI is the same as engagement with it, or that engagement is the same as learning.

The more durable path, based on what the data shows, is hybrid: AI manages data, adaptivity, and administrative logistics, while human educators provide the empathy, accountability, and inspiration learners actually need.

Photo by Antoine Dautry on Unsplash