On this page
Many families already pay for home tuition on top of school and coaching. Now AI tutors are available on a phone at any hour. Choosing between them, or deciding how to use both, is easier once you know what the research actually measured and where each one falls short.
What does research say about human tutors?
The famous number is "two sigma". In his 1984 paper in Educational Researcher, Benjamin Bloom reported studies in which "the average tutored student was above 98% of the students in the control class". Those were small studies: students in grades four, five and eight, learning probability or cartography, in blocks of 11 periods over three weeks, with a good tutor for one to three students at a time.
Later reviews found real but smaller effects. Kurt VanLehn's 2011 review put human tutoring at an effect size of 0.79, not the widely believed 2.0. A 2023 meta-analysis of school tutoring experiments by Nickow, Oreopoulos and Quan found a pooled effect of 0.288 standard deviations. Effects tended to be largest when tutors were teachers or paraprofessionals, in earlier grades, when sessions ran at least three days a week, and when they happened during school.
What does research say about AI tutors?
Before today's chatbots, VanLehn found that intelligent tutoring systems, most of which work step by step, reached an effect size of 0.76, close to human tutors. The newer studies are encouraging but conditional:
In a Harvard physics trial with 194 students, Kestin and colleagues found students learned more from a carefully designed AI tutor than from an in-class active-learning lesson, with an estimated effect of 0.73 to 1.3 standard deviations. The median student spent 49 minutes with the tutor, against a 60-minute class. The authors note its accuracy "relied on pre-written answers".
In a World Bank pilot in Nigeria, six weeks of after-school AI tutoring focused on English, with teacher support, produced gains of "about 0.3 standard deviations".
In a PNAS field experiment, students given unrestricted GPT-4 for practice later did 17% worse than students who never had it. A version with learning safeguards largely avoided this.
The pattern: AI tutoring helps when it is built to teach and used with some structure. It can hurt when it simply hands over answers. Our post on AI tutors in real classrooms goes deeper.
Where is each one stronger?
A good human tutor is stronger at:
Noticing that a student is tired, anxious or bluffing, without being told.
Accountability: a fixed weekly time and someone who expects the homework done.
Reading handwritten work, messy diagrams and half-formed ideas.
Judgement about exam strategy, priorities and when to stop a topic.
An AI tutor is stronger at:
Being there at 11 pm when a doubt comes up, so it does not wait a week.
Patience with the tenth "why?" and with many practice variations.
Access: it is available wherever there is a device and an internet connection, not only where good tutors live nearby.
Their weak spots mirror each other. Human tutors vary a lot in quality, usually cost more per hour, and can only see you for a few hours a week. AI tutors can be confidently wrong, only know what you tell them, and make copying easy if you let them. Our 8-test checklist for evaluating an AI tutor helps with the second set.
How can you combine them?
A simple weekly split, which is our suggestion rather than a research finding:
Daily, with AI (20 to 40 minutes): first-line doubts, hints on practice problems, quizzing yourself on the day's chapter. Write down anything that stays unclear after two attempts.
Weekly, with the human tutor (60 to 90 minutes): start with that "stuck list", then review the week's test or mock, then plan next week. Do not spend this time on problems the AI could have handled.
Weekly, with a parent (10 minutes): look at the stuck list together. Is it getting shorter? Are the same topics repeating?
Before paying for a tutor, ask: How will you know what my child already understands? How do you handle a wrong answer: tell, or ask? Will you set work between sessions? What will you need from us?
If the plan is JEE Main or Advanced preparation, Fermi Testprep offers one combined route: an AI tutor, live classes, adaptive mocks, mentorship and a study plan that updates based on mastery. For families still weighing options, our guide on coaching, online classes or self-study lists the questions to ask first.
