Call us anytime

Last updated: November 11, 2026
AI tutoring is software that works through a problem with your child instead of for them: it asks questions, gives hints, checks each step and adapts to what your child shows they know. The research is encouraging when two things are true: the tool is built to teach, and someone makes sure your child actually uses it. In a randomized trial at Harvard, students who learned physics from a carefully designed AI tutor showed median learning gains more than twice those of students in an active-learning class, in less time. A Stanford review published in August 2026 adds the other half of the picture: tutoring led by a person remains the best-proven option, and AI does the most good when it works alongside people. Below: what AI tutoring is, what the studies show, how it compares with a human tutor, what it costs, how to set it up at home, and what a bright child needs once homework help is no longer the question.
The same AI tutor can double what one student learns and barely register for another. The difference comes down to a few habits most families never hear about. Send me the insider guide for parents →
The idea is older than chat assistants. For decades, researchers built "intelligent tutoring systems" that give feedback on each step of a problem. In a 2011 review in the journal Educational Psychologist, Arizona State University's Kurt VanLehn found that these step-by-step systems were nearly as effective as human tutors: an average effect size of 0.76, against 0.79 for human tutoring, both compared with no tutoring.
Today's AI tutors add large language models, so your child can type, speak or photograph a problem and ask questions in plain language. Stanford's 2026 brief sorts the tools into three groups:
| Type of tool | What it does | Who it serves |
|---|---|---|
| General-purpose AI chat assistants | Answer questions, write and summarize on request | Anyone; the user drives every exchange |
| Student-focused AI tutors | Built around learning goals: step-by-step guidance, hints, "productive struggle" | The student directly |
| Educator-focused AI tools | Suggest teaching moves to a live tutor, flag students who need help | The tutor or teacher |
A general assistant can be a superb study partner, but it is built to respond. A student-focused tutor is built to hold back the answer and ask your child for the next step.
The evidence is young but growing. The studies worth knowing:
| Study | What was tested | What it found |
|---|---|---|
| Harvard, Scientific Reports, 2025 | College physics students learned from a custom AI tutor designed by instructors, compared with an in-class active-learning lesson | Median learning gains more than double; median time on task 49 minutes, against about 60 in class |
| PNAS, 2025 | Nearly 1,000 high school math students practiced with a general chat assistant or with a tutor version built with teacher-designed guardrails | Practice grades rose 48% with the general assistant and 127% with the tutor version. When the tools were removed, the general-assistant group scored 17% lower on the exam; the guardrails largely prevented that drop |
| Stanford, Tutor CoPilot, 2024 | More than 700 human tutors, half given an AI tool that suggests expert teaching moves in real time, working with about 1,000 students | Students were 4 percentage points more likely to master the topic (66% versus 62%); 9 points more for students of lower-rated tutors |
| Stanford SCALE brief, 2026 | Review of research on AI and tutoring | Human-led tutoring has the strongest evidence; AI-led tutoring "does not yet meet the established evidence base"; results depend on how well the tool is built into a routine |
Three lessons stand out for parents.
Design matters. The Harvard tutor worked because instructors wrote detailed, question-specific prompts. In the PNAS study, students who used a general assistant to get answers leaned on it and did worse once it was gone. Set up to give hints instead, the same technology protected the learning.
Use matters as much as design. The Stanford brief cites a study of 181,000 students on a supplemental math platform: only 5% reached the recommended 30 minutes a week, and 41% never logged on. A tool your child does not open teaches nothing.
People still matter. The strongest result for school-age students here, Tutor CoPilot, put the AI behind a human tutor rather than in place of one.
In one study Stanford cites, only 1 student in 20 used the math tool enough to benefit. Families who get results set it up differently from day one. Show me how to find it →
Most families do not need to choose. Each does something the other cannot do as well.
| AI tutor | Human tutor | |
|---|---|---|
| Availability | Any hour, instantly | Scheduled sessions |
| Cost | Low or free for many tools | Higher, set by the tutor's rate and hours |
| Patience | Unlimited re-explanations | Re-explanations shaped by knowing your child |
| Accountability | Your child must decide to open it | A person your child shows up for |
| Spotting the real problem | Improving, strongest on step-by-step work | Reads tone, hesitation and the quiet "I'm lost" |
| Evidence base | Emerging | Strong |
Stanford's brief describes what it calls high-impact tutoring, the best-studied model: a consistent tutor, three or more sessions a week totaling about 90 minutes, for at least 10 weeks, with no more than four students per tutor. Students who receive it show learning gains of three to more than 15 months. Practitioners name the relationship as a key reason it works, and consistent tutoring with the same tutor improves attendance, motivation and learning.
That relationship is the thing many parents sense they are paying for. A good tutor notices that your child nods along but never asks a question, and changes course. AI fills the gaps between sessions: the 9 p.m. homework question, the extra practice set, the second explanation of a step. For a child who needs to catch up in one subject, our guide to finding a math tutor covers what to look for, and our SAT tutoring guide covers test prep.
The US Department of Education's July 2025 guidance on AI in schools puts the goal well: students should learn "with – rather than exclusively from – AI." At home, that comes down to a few habits:
A two-minute question after each AI session shows what your child truly understood. Get quiet insights like it from PhD mentors at top-20 universities. Send me the insider guide for parents →
Costs vary by tool and plan, and many families start with free options. One example from a nonprofit: Khan Academy's AI tutor, Khanmigo, costs $4 a month or $44 a year for families. A parent subscription can cover up to 10 children and lets the parent view each child's chat history and receive moderation alerts. Khan Academy says Khanmigo is designed to guide children to the answer rather than give it, and that its content library and learning platform "will always remain 100% free." To buy a parent or learner plan, the account holder must be 18 or older and live in the United States.
A human tutor costs more per hour, because you are paying for a person's time, preparation and judgment. Our math tutor guide shows what tutors earn according to the US Bureau of Labor Statistics.
The best AI tutor is the one built for learning that your child will use three times a week. Judge any tool, including free ones, on five points:
Khanmigo, from the nonprofit Khan Academy, is one option built around these goals and tied to Khan Academy's free lessons. Teachers can also tell you what the school already provides.
AI tutoring and human tutoring both do the same valuable job: they help your child keep up, catch up or move ahead in the courses school sets. For many children, that is exactly what is needed.
Some children reach a different point. They finish the homework in ten minutes, ask questions the class never reaches, and say school feels like waiting. More of the same material, delivered faster, does not answer that. What they need is a real question with no answer in the back of the book, and an expert who takes their thinking seriously. Our guide on how to challenge a gifted child covers the options at school and at home, and how to get into research in middle school shows how a curious 12-year-old can start.
This is where learnSTEM fits. A student works one-on-one with a PhD mentor from a top-20 university on a question in the field they love, from first idea to finished research paper. A mentor does what the best tutors do, and then more: as one learnSTEM student put it, "If you don't understand her first explanation she will change it to better suit you." The difference is the destination. Instead of the next unit, your child builds something original, and finds out what kind of work they want to do. For older students, see whether research helps college admissions.
Picture your child explaining their own research question to a PhD mentor who is genuinely curious about the answer. See how families make that leap. Show me the experiments →
It depends on the tool. Some are free; Khanmigo, from the nonprofit Khan Academy, costs $4 a month or $44 a year for a family plan covering up to 10 children. Human tutoring costs more per hour.
Not yet, on the evidence. Stanford's 2026 brief finds that human-led tutoring has the strongest research base, and AI-led tutoring does not yet meet it. Well-designed AI tutors have produced strong results in specific studies, and AI that supports a human tutor shows promise.
A general chat assistant usually will if asked. Student-focused tutors are designed to give hints and ask questions instead. Either way, have your child explain the method back to you.
Choose tools with parent visibility, clear privacy terms and an age requirement your child meets. The US Department of Education asks that AI tools in education protect student data and that parents understand how they work.
The phrase also names a job: people with subject expertise who write and grade answers to train AI models. That is a different thing from tutoring a child. For students who want to work in AI, a research project in machine learning is a strong first step; see our list of research topics for high school students.
All bright students reach a point where school stops challenging them. learnSTEM provides that challenge: the space to explore a field of their choice, guided by a world-class PhD mentor from a top-20 university, completely one-on-one and on your own schedule. Students finish with a passion for a new subject, a research project they are proud of, and a clear sense of what they want to pursue next.
See if your child qualifies for the STEM Explorers Program → · See if your child qualifies for the STEM Research Program → · Questions about eligibility? Contact us →
"It has only strengthened my love for learning about the world and given me a paper that I am proud to say I've made." — learnSTEM student
Early Admission – November 15, 2026
Regular Admission – December 13, 2026
Late Admission – January 31, 2027
The Program runs four times per year: Spring, Summer, Fall, and Winter.
