Most explanations of AI tutoring are written by companies selling one, which means the definition tends to describe whatever that company built. Here is an attempt at a more useful version: what the term should mean, how such a system actually operates, and what it is still bad at.
A definition that is not a sales pitch
An AI tutor is a system that decides what a student should work on next, based on evidence about what that student currently knows.
The important word is decides. A tool that answers whatever you ask is doing something valuable, but it is not tutoring, because you are still the one directing the session. If a student does not know what they are weak at, and most do not, then a system that only responds will faithfully help them avoid the thing they most need to practise.
The four things a tutor does that a chatbot does not
1. It assesses before it teaches
A tutor starts by finding out where you actually are, which usually means asking you to attempt something rather than asking what you want help with. Students are unreliable narrators of their own understanding, generally in the optimistic direction.
2. It keeps a model of you
Something in the system has to remember that you have now missed three questions involving negative signs. That model is what makes the next decision different from the last one. A chat session that starts fresh each time cannot do this, however capable the underlying model is.
3. It chooses the next task
This is the actual tutoring. Given what it knows about you, the system selects what you should attempt now: a step back to a prerequisite, another rep on something shaky, or a harder variant because you have got it.
4. It withholds
The hardest one, and the one most systems fail. A good tutor frequently knows the answer and does not give it to you. It asks a narrower question instead, because the struggle is where the learning happens. A system optimised to be helpful in the ordinary sense will hand over the answer and feel excellent to use while teaching you nothing.
How it decides what to ask next
Underneath, the loop is unglamorous and roughly the same across serious implementations:
- Present a task.
- Observe the response, including how long it took and where it broke down.
- Update the internal estimate of what the student knows.
- Select the next task from that estimate.
- Repeat.
Generative AI changed step 1 and step 2 dramatically: a system can now produce an unlimited supply of tasks and can read a free-text answer rather than a multiple-choice selection. It did not remove the need for steps 3 and 4, and a product that skips them is a very good answer engine rather than a tutor.
What the Harvard trial showed
The strongest evidence that a well-built AI tutor works comes from a randomised controlled trial published in Scientific Reports in 2025. The researchers took 194 undergraduate physics students at Harvard and randomly assigned them either to a purpose-built AI tutor or to an active-learning classroom with experienced instructors. Active learning is not a weak comparison; it is one of the better-evidenced ways to teach.
The AI-tutored group learned more, in less time.
Two things are worth holding onto. The first is that the tutor was deliberately designed around teaching principles, not simply wired up to a chat model. The second is that this was one subject, one institution, and university students who had already chosen to study physics. It is a strong result, not a universal one.
What AI tutors are still bad at
Motivation. The trial measured learning within a structured course that students had signed up for. It did not test whether an AI tutor can get a reluctant fourteen-year-old to open the app on a Tuesday.
Noticing what is not said. A human tutor sees a student go quiet, or answer quickly and wrongly in a way that suggests guessing rather than misunderstanding. Some of that is inferable from data; a lot of it is not.
Diagnosing unusual misconceptions. Systems are good at recognising the common wrong answers. A genuinely idiosyncratic misunderstanding, the kind a child invents themselves, is exactly the case a model has the least to go on.
Knowing when to stop. Sometimes the correct instructional decision is to abandon the plan because the child is upset or exhausted. No current system makes that call well.
These are the gaps the comparison with human tutors turns on, and they are why the arrangement we think holds up is AI alongside a human teacher.
AI tutor, AI teacher, AI study assistant: not synonyms
These get used interchangeably and should not be.
An AI study assistant responds to requests. You bring the question; it brings the answer. ChatGPT used for homework is a study assistant.
An AI tutor directs the session. It decides what you attempt, tracks what you know, and chooses when to make things harder.
An AI teacher would additionally own the curriculum, the assessment and the pastoral judgement of a class. Nothing available today is credibly this, whatever the branding suggests, and whether anything should be is a separate question.
The practical difference between the first two is what you are paying for, and we break it down in AI study assistant vs AI tutor.
Are they accurate?
Not reliably, and this is worth saying plainly. Systems built on large language models can state incorrect things fluently and confidently. In well-built tutoring products this is constrained: questions are drawn from a vetted bank, or answers are checked against a known solution rather than generated freshly each time.
Ask any product you are considering where its questions come from and how it knows the answer is right. A vague reply is informative.
One question sits before all of this, and it is worth asking first: whether your child needs a tutor of any kind. We have written the signs that genuinely warrant paying for help, and three that usually do not, because the answer for a good many families is not yet.
Why a tutoring company wrote this definition
eGunas currently provides one-to-one teaching with human tutors. We are building AI-assisted learning, and it has not launched, so nothing here describes a product we are selling today. We have written the definition we would want as parents rather than the one that would be most convenient to us as a company.
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.
