If you're a parent right now, you've probably wondered whether AI is going to help your kid learn or just help them skip the work. It's a fair question. Your student can ask a chatbot almost anything, and it will happily answer.
The good news is that researchers have started testing this with real students in real classes, and the results are worth knowing. The short version: AI can help students learn a lot, but only when it's built to make them do the thinking. Here's what two recent studies found, and what it means for how your student studies.
The Harvard study: an AI tutor beat a great class
In fall 2023, researchers at Harvard ran a randomized controlled trial in an introductory physics course, the kind of study where students are assigned to groups so the comparison is fair. The results were published in Scientific Reports in June 2025.
About 200 students learned the same lessons two different ways:
In class, taught with active learning: hands-on, interactive teaching that's already known to work much better than a traditional lecture.
On their own with an AI tutor that the instructors designed to guide students through each problem one step at a time, using the same research-based teaching ideas as the class.
The comparison wasn't AI against a boring lecture. It was AI against a well-run, modern classroom. Here's what they found:
More than double the learning. The AI-tutored students' median learning gains were more than twice those of the in-class group.
In less time. The typical student spent 49 minutes with the AI tutor, compared with a 60-minute class.
More engaged. Students in the AI group said they felt more engaged and more motivated.
The researchers are careful about what this means, and so are we. They don't claim AI tutoring will always beat good teaching. They say the results depended on a tutor carefully designed by experts, with step-by-step prompts written for each problem. That detail turns out to be the whole story.
The catch: AI that gives answers can backfire
A second study shows what happens when AI isn't built that way. Researchers worked with nearly 1,000 high school math students in Turkey and published their findings in PNAS in 2025. Some students practiced with a standard ChatGPT-style assistant. Others used a version set up like a tutor, which gave hints instead of handing over answers.
While students had the AI, both groups did much better on practice problems: 48% better with the standard version and 127% better with the tutor version.
Then the AI was taken away for the exam. Students who had used the standard version scored 17% lower than students who never had AI at all. They had leaned on it as a crutch, and when it was gone, they hadn't learned the material. Students who used the hint-giving tutor didn't have that drop.
Put the two studies side by side and the lesson is clear. The same technology can help or hurt. What matters is whether it makes your student do the thinking.
What this means for parents
You don't need to ban AI or hand your student a chatbot and hope for the best. A few simple questions will tell you whether an AI study tool is the helpful kind:
Does it ask your student questions, or just answer theirs? Quizzing and step-by-step hints build memory. Finished answers mostly don't.
Could your student pass the test without it? That's the real goal. Practice should get harder to need, not easier to lean on.
Is it built around what your student is actually studying? Practice that matches this week's material is practice they'll use.
Can you see what they did? A record of what they studied and how they scored beats asking "Did you study?"
This lines up with decades of learning research, too. A major 2013 review of study techniques rated quizzing yourself and spreading study over several days as the most effective, and re-reading and highlighting as some of the least. Good AI tools simply make the effective habits easier to stick with.
How we built LearnVai
We built LearnVai (say "Learn Vay") around exactly this idea. It's an AI study tutor for students in Spanish, French and German classes, and it's designed to make your student practice, not to do the work for them.
It starts from your student's real class. They tell LearnVai what's on the quiz and when it is, then upload the study guide or snap a photo of the handout.
It builds a plan that spreads the studying out. Short nightly sessions finish the night before the test, instead of one big cram.
It quizzes them. Flash cards, vocab matching, practice quizzes and AI conversation practice for the speaking part, all from this week's material.
You can see the work. A study history shows what they finished and how they scored.
To be clear, LearnVai wasn't part of the studies above, and no tool can promise a grade. But the research points in one direction: the AI that helps is the AI that makes students think. That's the kind we set out to build.
It's $12.99 per student per month, and setup takes a few minutes. See how LearnVai works.
Sources
Kestin, Miller, Klales, Milbourne & Ponti, "AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting", Scientific Reports 15:17458 (June 2025).
Bastani, Bastani, Sungu, Ge, Kabakcı & Mariman, "Generative AI without guardrails can harm learning: Evidence from high school mathematics", PNAS 122(26) (2025).
Dunlosky, Rawson, Marsh, Nathan & Willingham, "Improving Students' Learning With Effective Learning Techniques", Psychological Science in the Public Interest (2013).
