What Is an AI Tutor? How AI Tutoring Really Works

Last Updated: September 27, 2026

Summary: An AI tutor is software that uses an AI model to explain concepts, answer questions, and adapt to a student's level, standing in for or supplementing a human tutor. AI tutoring genuinely works, but real research shows it works far better paired with a human than left fully on its own. This guide covers how AI tutors work, what the actual research found, the different types available, and how to pick one.

"AI tutor" gets pitched as a simple fix for expensive, hard-to-schedule human tutoring. The honest picture, based on real classroom research rather than marketing copy, is more specific than that.

💡 Not sure which AI tool is actually right for your business?

Get the free guide, Which AI Tool Should You Actually Use? — a straightforward breakdown of the leading AI tools to help you pick the right one for your needs.

Subscribe to AI Business Weekly for the guide, plus daily coverage of AI trends, acquisitions, and product launches.

Get the Free Guide →

An AI tutor is a piece of software, usually built on a large language model, that explains concepts, answers questions, and walks a student through problems the way a human tutor would. Instead of just handing over an answer, a well-built AI tutor is designed to guide a student toward it, adapting explanations to what that student already understands.

The category covers a wide range of products: general-purpose AI tutoring platforms, subject-specific tools (math, coding, test prep), and AI language tutors built specifically for conversation practice. Some run standalone. Others, like Khan Academy's Khanmigo, are built into an existing learning platform.

Most AI tutors follow a loop similar to any other AI agent: the student asks a question or submits an answer, the model assesses where they're stuck, and it responds with a targeted explanation, a hint, or a follow-up question rather than the raw answer.

The better AI tutoring platforms track what a student has already covered and adjust difficulty over time, similar to how a human tutor builds on previous sessions. The weaker ones are closer to a chatbot with an educational skin, giving direct answers on request with no real pedagogical structure behind it.

That distinction, guided questioning versus direct answers, turns out to be the single biggest factor separating AI tutoring that actually teaches from AI tutoring that just does the homework faster. A tool built to ask "what have you tried so far?" before offering help pushes a student to do real thinking. A tool that answers immediately on request removes that step entirely, which is convenient in the moment and counterproductive for actual learning.

This is where most AI tutor content stops short, and it's the part worth taking seriously. A two-year NBER field experiment tracking Khan Academy's Khanmigo in real middle school classrooms found that students engaged with the AI tutor on only about a third of the days they were using Khan Academy, and a large share of that engagement was off-topic or attempts to get direct answers rather than genuine tutoring.

Stanford's SCALE Initiative reviewed the broader evidence and reached a similarly specific conclusion: AI tutoring works best as support for a human tutor, not as a replacement for one. Human-led tutoring with AI support behind the scenes showed the strongest results, with learning gains equivalent to 3 to over 15 months of extra progress depending on grade level. Fully AI-led tutoring with no human involved showed the weakest evidence, with 40-47% of students never engaging with the platform at all, and the ones who did averaging only 2-5 minutes a week, far short of the roughly 90 minutes weekly researchers associate with measurable gains.

The practical takeaway: an AI tutor genuinely helps when it's used consistently and paired with some human structure, whether that's a parent, a teacher, or a tutoring service using AI as a tool rather than a full substitute. Used alone with no oversight, most students simply don't engage with it enough for it to matter.

An AI language tutor in particular tends to see strong genuine engagement, since conversation practice is something students can do in short, low-stakes bursts that don't require the same sustained focus as working through a hard math problem.

Khan Academy's Khanmigo is the most heavily studied AI tutor currently in classroom use, built into Khan Academy's existing free platform and aimed at K-12 students. Khan Academy's own writeup of its latest work on the tool describes ongoing changes aimed at pushing Khanmigo toward more guided, question-based interaction rather than direct answers, a direct response to the kind of engagement gaps the NBER study found.

General-purpose AI assistants like ChatGPT and Claude are also widely used informally as ad hoc tutors, particularly for older students and college-level work, though they weren't purpose-built for pedagogy the way a dedicated tutor ai app is. Dedicated AI language tutoring apps focus specifically on conversation practice rather than general homework help, and tend to see stronger real engagement than general subject tutoring, since a short daily conversation is an easier habit to build than a sustained homework session.

The right pick depends heavily on the use case in the table above. A general-purpose model is flexible but requires the student to ask good questions. A purpose-built AI tutor app does more of that structuring work automatically, at the cost of being narrower in scope.

Cost is one of the biggest practical reasons families look at AI tutoring in the first place. Care.com's tutoring rate data puts the average private tutor at roughly $24 an hour for K-12 subjects, climbing to around $60 an hour for college-level help and $100 an hour for specialized test prep, with rates in major cities like New York and Los Angeles regularly running $100-$200 an hour.

Most AI tutoring apps run somewhere between free and $20-40 a month for unlimited use, and general-purpose assistants like ChatGPT or Claude add tutoring on top of a subscription many families already pay for other reasons. That price gap is real. It's also exactly why the research on engagement matters so much: a $30-a-month tool that a student uses three minutes a week isn't actually cheaper than an hour of focused human tutoring, it's just a lower price for less delivered value.

Based on the actual research, this isn't really an either-or question. The strongest evidence favors human tutoring with AI support in the background, not AI tutoring as a standalone replacement for a person.

A human tutor brings a relationship, accountability, and the ability to notice when a student is disengaged in ways current AI tutors generally can't. An AI tutor brings availability at any hour, patience with repetition, and a much lower cost per session. For a family or school weighing the two, the research points toward using AI to extend what a human tutor or teacher can do, rather than treating it as a full substitute, especially for younger students or anyone who struggles to stay engaged without external accountability.

Match the tool to the actual gap. A student who needs accountability and consistent structure benefits more from a platform with built-in tracking and, ideally, some human check-in, than from an open-ended chatbot they have to remember to use on their own.

Look for tools that ask guiding questions rather than just supplying answers, since that's the behavior the research links to actual learning gains rather than just faster homework completion. And treat engagement as the real risk. Based on the Khanmigo research, giving a student access to an AI tutor doesn't mean they'll use it well, or at all, without some structure around it.

Does AI tutoring actually improve grades?

The evidence is mixed and depends heavily on how it's used. Research on human-led tutoring with AI support behind the scenes shows real learning gains, equivalent to 3 to over 15 months of extra progress in some studies. Fully AI-led tutoring with no human involvement shows much weaker results, largely because most students don't engage with it consistently enough. It doesn't guarantee improved grades on its own, and results depend heavily on consistent use. The realistic next step is picking a tool with built-in structure rather than an open-ended chatbot.

Is Khanmigo actually effective?

A two-year field study found that middle school students engaged with Khanmigo on only about a third of the days they used Khan Academy, and much of that engagement wasn't genuine tutoring interaction. Students using Khan Academy overall did show faster math gains by year two, but researchers didn't attribute that specifically to Khanmigo engagement. It's a real, heavily studied tool, but the honest read is that access alone didn't translate into consistent use. See our broader guide on what AI agents are for how tools like this are built.

Can ChatGPT or Claude be used as an AI tutor?

Yes, informally, and many students already do this. General-purpose AI assistants can explain concepts and work through problems step by step, but they weren't purpose-built for pedagogy the way a dedicated AI tutoring app is, and they'll often give a direct answer if asked rather than guiding a student toward it. Purpose-built AI tutors typically have more structure around avoiding that. It works reasonably well for a student who already knows how to ask good questions.

What's the difference between an AI tutor and a chatbot?

An AI tutor is generally designed to guide a student toward understanding, using questions and staged hints rather than giving direct answers immediately. A general chatbot will typically just answer whatever is asked. The distinction matters for actual learning: research on AI tutoring links guided, question-based interaction to real learning gains, while direct-answer interactions look more like the student outsourcing the work than learning it.

Should schools replace human tutors with AI tutors?

Based on current research, no. The strongest evidence supports AI as a support tool for human-led tutoring, not a replacement for it. Stanford's review of the evidence found AI-only tutoring showed weak results, largely due to low and inconsistent student engagement when no human structure was involved. Schools considering AI tutoring get the best documented results by using it to extend a human tutor's reach, not to eliminate the human entirely.

An AI tutor is software built to explain, guide, and adapt to a student the way a human tutor would, and the honest research shows it genuinely works, under specific conditions. It performs best as support for a human tutor or teacher, not as a fully standalone replacement, because the real barrier isn't whether the technology can teach. It's whether students actually engage with it consistently, which multiple studies found most don't do on their own. Pick a tool with real structure, pair it with some human oversight where possible, and treat "gives students access" as the easy part, not the whole solution.

By Sameer Khan

This article was AI-assisted, then reviewed by Sameer Khan before publishing.

Sameer Khan is the founder of AI Business Weekly. He has a background in research and advisory, working with HR leaders and executives across Canadian public-sector and enterprise organizations on research and AI adoption. He holds an MBA from the Ted Rogers School of Management and has spent nearly a decade in B2B sales across SaaS, research and advisory, and AI.