A new study put AI tutoring head-to-head against human tutoring on one specific, measurable task: GRE test prep. The result — statistically equivalent score gains, at a fraction of the cost. That is a real data point. It is also a much narrower finding than the headlines suggest, and it says almost nothing about what actually gets a student into a competitive program: a real project, built with a real mentor, that the student can defend in an interview.

What the Study Found

The research, published as “StudentBench,” compared students working with AI tutoring tools against students working with human tutors on standardized test preparation. On GRE-style content, the two groups showed comparable score improvements. The AI-tutored group got there at significantly lower cost.

That is a meaningful comparison. Head-to-head, outcome-based studies like this one have mostly been missing from the “AI in education” conversation, which has leaned on anecdote in both directions. So take the finding seriously. Just take it for what it actually measures.

Why This Doesn’t Mean “Just Use AI”

Two caveats matter here.

First, this was standardized test prep: a well-defined skill, with clear right answers and a huge amount of practice material. That is exactly the kind of task AI tutoring tools are built for. It does not generalize to open-ended work — an original research project, an essay, a portfolio piece — where the value of a human mentor is less about drilling content and more about pushing back on an idea, or asking the follow-up question a student has not thought of yet.

Second, “equivalent learning gains” measures the score, not the experience. A good human mentor also builds a student’s judgment, confidence, and independence in ways a scoring metric was never built to capture.

What This Means If Your Student Is Building a Real AI Project

This is where the study’s real relevance shows up, and it is not about test prep at all.

If AI tools are already good enough to match a human tutor on a well-defined, drill-heavy task, that only sharpens the question every family should be asking about an AI program: what is the mentor actually there for? An AI tutor can drill vocabulary or a practice set because the task has a right answer. It cannot sit with a student while they decide what a project should even be, weigh the ethics of what they are building, or make the case for why it matters — that is judgment, not a drill, and it is the part STEAM in AI mentors are actually there for.

It is also why the program is not only an AI class. Every STEAM in AI student works through AI alongside Ethics, Design Thinking, and Entrepreneurship — the parts of a real project an AI tutor was never built to teach, and the reason a student can explain not just how their project works, but why it matters.

wp-block-buttons is-content-justification-center is-layout-flex wp-container-cta” style=”margin: 32px 0; justify-content: center; display: flex;”>

FAQ: AI Tutoring vs. Human Mentoring

Does this study mean AI tutors are as good as human mentors?

For standardized test prep specifically, yes — the study found equivalent score gains. For open-ended work like a research project or an application essay, the study does not make that claim, and the skills involved are different.

Should my student use AI tools to study for the SAT or GRE?

For drilling and practice, this study is a reasonable data point in favor of it. It is a lower-cost option for a well-defined task with clear right answers.

What should a student use a human mentor for instead?

Anything without a clear right answer: framing an original project, defending a research direction, revising an essay’s argument, or explaining a decision in an interview. That is judgment, not drilling, and it is what a STEAM in AI mentor is there to build with a student.

Is STEAM in AI just an AI tutoring program, then?

No. Students choose an AI Build or AI Research track, and the project is built around AI together with Ethics, Design Thinking, and Entrepreneurship, with two mentors rather than one. That combination is what a tutoring tool, AI or human, is not built to provide.

More on AI, Mentors, and Real Projects

• What Is the STEAM in AI Intensive? What Your Student Would Actually Build in 12 Weeks
• AI Project Ideas for High School Students, Grouped by Intended Major
• An AI Project With No Coding Experience: Julia Started at Zero and Went to Duke

A study can tell you what a tool is good at. It cannot tell you whether your student can defend their own project in front of someone asking hard questions. That is a mentor’s job, not a tutor’s.