The Setup: AI Tutors Go Mainstream

Two years ago, Khan Academy launched Khanmigo, an AI tutor powered by GPT-4 technology, into select classrooms. It was the kind of moment that made educators’ eyes narrow with equal parts curiosity and skepticism. An AI that could tutor students one-on-one? That could adapt to different learning styles? That could answer questions at 2 a.m. without getting tired or frustrated? The promise was intoxicating. By 2024, the tool expanded to all U.S. teachers for free, and by the end of 2025, over 2 million students in school settings were using it. That’s a lot of educational experimentation happening in real time.

But here’s what I’ve learned after twenty years in education: promises and reality are often distant cousins. So when the research started coming out, I paid attention. Because data is what separates genuine game-changers from expensive novelties. And the data on whether AI tutors actually close learning gaps? It’s messier than the headlines suggest.

The Good News: It Works, Especially for Students Who Need It Most

Let’s start with what should make every educator hopeful. WestEd, in partnership with Khan Academy, released a longitudinal study in late 2025 that followed students over a full academic year. Students who used Khanmigo for at least 30 minutes weekly showed a 0.23 standard deviation improvement in math achievement. That doesn’t sound enormous until you translate it into classroom reality: it means the average student who engaged with Khanmigo moved from the 50th percentile to roughly the 59th percentile. Real gains. The kind you actually see in student work and confidence.

But here’s where it gets genuinely exciting. English Language Learners showed a 0.31 standard deviation improvement, significantly larger than their peers. Think about what that means. These are students navigating mathematics while simultaneously processing English. They’re translating mathematical vocabulary, trying to understand a word problem when the language itself is unfamiliar. Khanmigo’s ability to explain concepts multiple ways, to pause and clarify without impatience, appears to address a real barrier. This isn’t just interesting data. This is a tool potentially closing a specific learning gap that has stubbornly resisted traditional interventions.

The Hard Truth: Engagement Collapses Without Structure

Here’s where I have to be the teacher who tells you the uncomfortable truth. A 2025 study from Stanford’s CEPA found something that made me wince in recognition. Student engagement with AI tutors dropped by 60% after the first three weeks of use when teachers didn’t actively facilitate and structure the experience. Sixty percent. That’s not a small enthusiasm dip. That’s abandonment.

This finding aligns with something I’ve observed my entire career: students don’t magically become self-directed learners just because a tool is available. The novelty wears off. The screen feels lonely. Without a teacher checking in, asking about struggles, celebrating progress, and building accountability, many students simply stop showing up. An AI tutor is phenomenally patient, but it can’t replicate the human relationship that motivates a teenager to try again after failing. This is the ingredient that research keeps confirming: teacher involvement matters more than the technology itself.

The Privacy Concern You Should Know About

Before you or your school jumps in, there’s a conversation that needs to happen. The U.S. Department of Education’s 2025 “AI in Education” guidance document flagged a significant issue. Of the EdTech AI tools reviewed, 78% did not fully comply with FERPA’s updated digital provisions. That means student data privacy is a genuine concern in this space, not a hypothetical one. When your students interact with Khanmigo, data is being collected. Some of that gets analyzed to improve the AI. Some might be stored. You deserve transparency about what happens to your students’ learning information.

Check out the U.S. Department of Education AI in Education guidance 2025 for specifics. And if you’re implementing any AI tutoring tool in your school, ask the vendor explicit questions about data handling. This isn’t paranoia. It’s responsibility. Your students’ data is valuable, and they deserve protection.

What This Means for Your Classroom or Self-Study

So should you use Khanmigo or another AI tutor? I’d say yes, but strategically. The research points to three conditions where it genuinely helps. First, structure it. Don’t just hand a student a login and assume they’ll figure it out. Build it into your lesson design. Use it during class time when you can observe. Check in with students about what they’re learning from the interactions. Second, prioritize it for students facing language barriers or specific conceptual stumbling blocks. That’s where the research shows the biggest gains. Third, use it as a supplement to your teaching, not a replacement. The students who benefited most from Khanmigo in the studies were those whose teachers remained actively involved.

If you’re exploring Khanmigo specifically, Khan Academy Khanmigo educator overview has resources for implementation. But go in with your eyes open about both the genuine promise and the real limitations.

The Bottom Line: Potential, With Caveats

Two years of classroom data tells us this: AI tutors can close learning gaps, particularly for students who face specific barriers like language challenges. They can provide patient, adaptive support that’s genuinely valuable. But they’re not magic. They don’t eliminate the need for teachers. They don’t keep students engaged without human involvement and accountability. They require careful attention to data privacy and ethical implementation.

The question isn’t whether AI tutors work. The research says they do, under the right conditions. The real question is whether we’ll implement them thoughtfully, with teacher involvement as the centerpiece rather than an afterthought. That’s where the actual learning happens. I’d love to hear what you’ve observed if you’re using these tools in your classroom or in your own learning. What’s working? What’s falling flat? Teacher voices matter in this discussion, and we’re all still figuring it out together.