Almost every comparison of AI and human tutoring is published by someone selling one of them, and concludes in their favour. We sell human tutoring, so treat this page with the scepticism that deserves. What we can do is show you the two strongest studies and be honest about which parts go against us.
Two studies, opposite conclusions, both correct
The case for AI. A randomised controlled trial published in Scientific Reports assigned 194 Harvard physics undergraduates either to a purpose-built AI tutor or to an active-learning classroom run by experienced instructors. The AI group learned more, in less time. That is a real result against a genuinely strong comparison.
The case against. A study of roughly a thousand high school students published in PNAS gave students access to GPT-4 while they practised maths. Practice performance rose substantially. Then the tool was removed for the actual exam, and the students who had used the unrestricted version scored around 17% worse than the control group who never had it. A more constrained tutor-style version landed roughly level with the control.
These findings are not in conflict. The variable is not whether AI was involved; it is whether the system was built to make students think or built to be maximally helpful. A tool that removes difficulty raises performance while it is present and lowers learning once it is gone.
Where AI genuinely wins
We would rather state these plainly than pretend otherwise.
Patience. An AI tutor will explain the same concept eleven times without a flicker. Human patience is real but finite, and children can tell when it is running out.
Availability. It is there at 11pm the night before an exam, and on a Sunday, and during the holidays.
Cost per hour. Not close. One-to-one human teaching is expensive because it is one person's hour.
Drill and repetition. Generating twenty more questions of exactly the right type is something software does better than a person, and it is a large part of what maths practice actually requires.
No social cost to being wrong. Some students will admit confusion to software that they would never admit to an adult. This matters more than it sounds.
Where humans still win
Motivation and accountability. The Harvard study measured learning among students who had already enrolled in a physics course. It did not test whether software can persuade a disengaged fourteen-year-old to start. Someone expecting you on Thursday at five is a different kind of force.
Diagnosing the unusual. Systems recognise common misconceptions well. The odd, self-invented misunderstanding is exactly where there is least data to pattern-match against, and where a person asking "show me how you got that" still wins.
Reading the room. A tutor notices a child is upset, or tired, or has stopped trying while still typing. Sometimes the correct teaching decision is to stop teaching.
Judgement about the whole child. Whether to push now or rebuild confidence first is not a question about the syllabus, and it is not one any current system answers well.
Being accountable to you. You can ask a human tutor what happened this month and receive an answer that a person stands behind.
The crutch problem
The PNAS result deserves its own section, because it is the single most practically useful finding for parents.
Students using unrestricted AI got better answers and felt like they were learning. The felt sense of fluency was real, and it was misleading. They had outsourced the difficult step, and the difficult step was the one doing the teaching.
This has a direct implication: watch what happens when the tool is not there. If your child can do yesterday's problems today, unaided, the tool is helping. If they cannot, it is not, regardless of how good the homework looked. We turn that into concrete methods in How to study with AI without quietly getting worse.
Side by side, by task
| Task | Better suited to |
|---|---|
| Unlimited practice questions | AI |
| Explaining a concept a fourth time | AI |
| Late-night help before an exam | AI |
| Marking routine work | AI |
| Finding an unusual misconception | Human |
| Getting a reluctant student to start | Human |
| Deciding what to do when a child is discouraged | Human |
| Exam technique and pacing under pressure | Roughly even |
| Building a long-term plan for a weak subject | Human |
| Keeping a student honest about what they can do alone | Human |
If a comparison you read has one column winning everything, it is marketing. Ours has AI winning four rows and we still sell the other kind.
So what should a family actually do?
For most students the honest answer is not either. It is AI for volume, repetition and availability, with a person responsible for direction, diagnosis and accountability. That is what both studies point at once you stop reading them as a contest, and how that hybrid arrangement works week to week is set out in our guide to hybrid tutoring.
If you are choosing now and can only choose one: pick based on what is actually going wrong. A student who understands the work but has not done enough of it is well served by AI. A student who has stopped trying, or who has a misconception nobody has found yet, needs a person.
Before you pay for either, twelve questions worth asking may save you a term.
If the comparison has settled it for you, the next question is quantity rather than kind, and almost nobody answers it honestly: how much tutoring a child actually needs is usually less than you would be sold.
Why we included the parts where AI wins
eGunas provides one-to-one teaching with human tutors today. AI-assisted learning is something we are building and have not launched. We have tried to write the comparison we would want to read as parents, including the parts where the other side wins.
How eGunas helps your child
eGunas teaches one child at a time. A session follows the student's own board syllabus but is built around what that particular student needs next -- their pace, the topics they are actually weak at -- rather than one lesson delivered to thirty students at once. That is what personalised learning means in practice.
eGunas AI, our AI-powered learning platform, is planned for 2026 and not available yet. It is being designed to extend that same individual attention beyond the lesson hour: practice pitched at the right difficulty, available whenever your child sits down to work, so progress can continue between sessions instead of waiting for the next one.
- Browse our courses -- CBSE, ICSE and state boards, Classes 1 to 10
- How eGunas online tutoring works -- one-to-one lessons online, wherever in India you live
- Book a free demo class -- one-to-one with a tutor, at no cost
- Ask us about your child -- tell us the subject and year, and we will say honestly whether we can help
New articles are published on the eGunas blog.
