You will find confident answers to this in both directions, usually from people with something to sell or a position to defend. The accurate answer is that it depends, and unusually, we now have good evidence about what it depends on.
The honest answer is "it depends", and here is what on
Two randomised studies published in 2025 reached opposite conclusions. Neither is weak. Reading them together is more useful than picking one.
The variable separating them is not the model, the subject or the age group. It is whether the system was designed to make students do the thinking, or designed to be as helpful as possible.
The case for: the Harvard trial
Published in Scientific Reports.
Design. 194 undergraduate physics students at Harvard, randomly assigned to either a purpose-built AI tutor or an active-learning classroom taught by experienced instructors. Same material both ways.
Result. The AI-tutored group learned more, and took less time to do it.
Why it matters. Active learning is not a straw man. It is one of the better-evidenced approaches to teaching, and the comparison was run against experienced instructors rather than a lecture. This is a genuine result.
What it does not show. One subject, one university, students who had already chosen to study physics and turned up. It says nothing about whether an AI tutor can motivate a disengaged fourteen-year-old, because every participant was already motivated enough to enrol.
The case against: the PNAS study
Published in PNAS.
Design. Roughly a thousand high school students, given access to GPT-4 while practising maths. Two versions were tested: an unrestricted one, and a tutor-style one with guardrails on how freely it gave answers.
Result. Practice performance rose sharply with access. Then the tool was removed for the real exam. Students who had used the unrestricted version scored around 17% worse than the control group who never had it at all. The guardrailed version landed roughly level with the control.
Why it matters. This is the finding that should change household behaviour. It is not that AI failed to help. It is that unrestricted AI actively left students behind where they would have been without it, while feeling like it was working the whole time.
Reconciling them
Put side by side, the studies are not in tension:
- Designed to make students think, with answers withheld → learning improved.
- Designed to be maximally helpful, answers freely given → performance improved while present, learning got worse.
The technology is the same in both. The design is opposite. This is why "is AI good for students" has no general answer, and why the useful question is always which system, built how.
It is also, incidentally, the distinction between a study assistant and a tutor.
This is already the default
Whatever any of us conclude, students have decided. Pew Research Center reported that the share of US teens using ChatGPT for schoolwork doubled from 13% in 2023 to 26% in 2024. Later fieldwork found 54% using chatbots for schoolwork help, with about one in ten doing all or most of their schoolwork that way.
Note what that describes: unstructured access to a general-purpose assistant. That is the condition the PNAS study found harmful. For most families the live question is not whether AI enters their child's studying but whether anything shapes how.
Five things a parent can actually watch for
You do not need to audit the technology. You need to notice the symptoms.
- The closed-book check. Can your child redo yesterday's work today with everything shut? This is the only reliable signal, and it takes two minutes.
- Homework quality rising while test scores do not. The classic pattern from the PNAS result, and the earliest warning available to you.
- Speed without struggle. Homework that used to take forty minutes now taking eight is not necessarily progress.
- Cannot explain their own answer. If they cannot talk you through it, they did not produce it.
- Reluctance to work without the tool. Dependence shows up as discomfort before it shows up in marks.
What we do not know yet
Worth stating, because almost nothing written on this topic admits it:
- Whether the Harvard result holds for younger students, weaker students, or subjects that are less structured than physics.
- What long-term use does over years rather than weeks. Every study so far is short.
- Whether the guardrails that worked in a controlled trial survive contact with teenagers who are motivated to get around them.
- How any of this interacts with exam formats that are themselves changing.
Anyone offering you certainty on those points is ahead of the evidence.
So, is it good for students?
A well-designed system that makes students do the work: the evidence says yes, and it is better than we expected.
Unrestricted access with no structure: the evidence says it can be actively worse than nothing, while feeling fine.
The difference is design, and design is a choice somebody made. That is why the practical version of this question is how to choose, and why the arrangement we think holds up is AI with a human still responsible.
One risk worth knowing before it arrives: detection software flags honest work, and it flags students writing in a second language most of all. What to do about a false accusation sets out the evidence that actually helps.
eGunas provides one-to-one human tutoring today; our AI-assisted learning has not launched. We would rather publish the study that complicates our position than leave it out.
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.
