Search “AI tutor for kids” right now and you’ll find dozens of comparison posts, all doing the same thing: lining up Khanmigo, Photomath, Mathway, and a handful of others, then ranking them by features, price, and test-score improvements.
Here’s the problem. Test scores can go up while understanding stays flat. We’ve spent years watching kids get the right answer without knowing why it’s right — a gap that shows up the moment the problem changes shape. In our world, we call this the illusion of mastery: the appearance of learning without the cognition that’s supposed to produce it. And almost none of the AI tutor reviews out there are measuring for it.
So instead of asking “which app scores highest,” here’s a different question worth asking about any AI tutor before your kid spends real time with it: does this tool make my kid think, or does it think for them?
Two very different jobs wearing the same label
“AI tutor” gets used for two fundamentally different tools, and the difference matters more than any feature comparison.
Some tools — Photomath and the Socratic app among them — are built to give answers fast. Point the camera at a problem, get the solution, often with steps shown. That’s genuinely useful for checking work. It is not teaching. A student can run ten problems through a tool like this and walk away having verified answers without ever generating one independently.
Other tools, like Khanmigo, are built around Socratic questioning — they’re designed to withhold the answer and instead ask the question that gets the student to find it themselves. That’s a much harder product to build, and a much slower one to use. It’s also the only kind of AI tutoring that maps onto how kids actually learn to reason.
This is the same distinction we teach in the Parent Track when we talk about the parent’s role during math practice: ask before you tell, wait through the silence, respond to the child’s reasoning rather than supplying your own. A Socratic-style AI tool is trying to do that same job. An answer-giving tool is doing the opposite job while using the word “tutor.”
What to actually watch for, if you try one
Rather than take a review site’s ranking at face value, watch for three things while your own kid uses a tool:
- Can they explain the answer back, in their own words, after the tool helped? A Socratic-style tool should make this possible, because the reasoning was theirs to begin with. An answer-giving tool often won’t — the reasoning was never theirs.
- Can they try the next similar problem without help? This is the real test of transfer, not the interaction they just had.
- Does the struggle feel productive or just frustrating? A tool that sits with a kid through confusion until they get there is doing something different than a tool that removes the confusion instantly.
None of this shows up in a features table. It only shows up if you’re watching for it while your kid is actually using the tool.
The honest verdict
We’re not against AI tutoring — that would contradict the reason Auxesis exists. AI, used well, can ask better questions than a tired parent at 7pm can manage, and it can do it patiently, every single time. That’s real augmentation.
But “AI tutor” is not one category. A tool that hands over answers on demand is solving a different problem than a tool that makes a kid work for the answer — and only one of them is actually tutoring. Don’t start with a review site’s rankings. Start with the question above: after using it, could your child do the next problem alone? If the honest answer is no, the score may have gone up, but nothing else did.
Leave a Reply