AI Tutor or Real Tutor? What Actually Helps Your Child Learn
Nelson Barrios is the Founder of Edukko, a Dubai-based EdTech platform delivering personalized tutoring, mentorship, and academic support across IB, GCSE, A-Level, American, and French curricula. A PGCE-qualified, multilingual practitioner trained by the UK Ministry of Education, he harnesses AI to transform learning worldwide.
Every week a parent asks me some version of the same question. My child is already using an Artificial Intelligence (AI) app for homework, so do we still need a tutor? It is a fair question, and the honest answer is more interesting than either side of the debate admits. The research now shows exactly where AI tutoring works, exactly where it falls short, and why the strongest results come from putting the two together. This article walks through that evidence and explains how we have built Edukko around it.
The question parents ask
I run a tutoring company, so you would expect me to tell you AI tutors are no match for a real teacher and leave it there. I am not going to do that, because it is not true, and parents can tell when someone is defending their business instead of answering the question.
Here is what I actually believe after fifteen years in classrooms and five years building Edukko. AI is the most significant thing to happen to one-to-one learning in my working life. It is also, on its own, not enough. Both of those are true at the same time, and the research backs both of them up. The parents who get the best results for their children are the ones who understand the difference between what a machine can do and what a person does, and stop treating it as a choice between the two.
What forty years taught us
Start with the number every education researcher knows. In 1984, Benjamin Bloom at the University of Chicago published a study showing that students who were tutored one to one performed two standard deviations better than students taught in a conventional classroom. In plain terms, the average tutored student outperformed 98 percent of the students taught in a normal class. Bloom called it the two sigma problem, because one-to-one tutoring clearly works, but no society could afford a teacher for every child.
That is the problem every education technology company has been trying to solve for four decades. Can a machine give every child the benefit of a tutor? For most of that time, the honest answer was no. In the last two years, the answer has started to change, and it is worth being precise about how.
Where AI genuinely wins
In 2025, a randomized controlled trial published in Nature Scientific Reports gave university physics students either a purpose-built AI tutor or a well-run active learning class. The AI group learned significantly more, in less time, and reported feeling more engaged and more motivated. I read that paper twice because the effect was larger than I expected. The researchers were careful to note that the AI tutor was designed around established teaching principles rather than simply handing out answers, and that detail matters enormously, as we will see.
So let us be fair to the technology. A good AI tutor is available at eleven at night when a student finally sits down to revise. It never gets impatient. It can explain the same quadratic equation six different ways without sighing. It can generate fifty practice questions on exactly the topic a student is weak on, in the exact format of their exam board, in seconds. For drilling, for retrieval practice, for instant feedback on a first attempt, it is better than any human, including me.
“For practice, feedback and availability, a good AI tutor beats any human. That is not a threat to teachers. It is a gift to them.”
Where AI quietly fails
Now the other half. A 2025 study comparing AI tutoring with human-led sessions found that AI followed predictable response patterns and struggled to adjust in real time when a student needed more explanation, redirection, or scaffolding. Anyone who has taught a fourteen-year-old knows exactly what that means. The moment a student goes quiet, looks at the window, or says I get it in a tone that means I do not, a teacher changes course. The machine carries on.
A literature review of AI tutoring systems across global contexts reached a conclusion that should be printed on the wall of every educational technology (EdTech) office. Effectiveness is maximized when AI is used to complement, rather than replace, human instruction. The review also flagged something parents rarely hear about: AI tools that simply deliver answers, rather than guiding a student through the reasoning, appear to produce close to zero learning, and in some cases make things worse. A homework app that gives your child the solution is not a tutor. It is a very polite way of not learning.
Then there is the part that no study measures well, because it is hard to measure. A real tutor notices that a Year 11 student has stopped trying in math because of something that happened at school, not because of math. A real tutor knows that this particular student needs to be challenged and that one needs to be reassured. A real tutor can look a parent in the eye and say that your child is not behind, your child is scared, and here is what we do about it. I have never seen a machine do that, and I do not expect to.
The study that settles it
The most useful research I have read this year did not pit AI against humans at all. A team at Carnegie Mellon ran a year-long study in real classrooms where students used an AI math tutor, and some of them also received periodic support from a human tutor alongside it. Students who had both outperformed students who only had the AI, both inside the software and on external standardized tests. The researchers' own words were that relatively small doses of human support can meaningfully improve learning outcomes over time.
Read that again, because it is the whole argument. Not AI instead of a tutor. Not a tutor instead of AI. A tutor whose impact is multiplied by AI, and an AI whose limits are covered by a tutor. That is where the best outcomes are, and it is the only model I would put my own children into.
“Not AI instead of a tutor. Not a tutor instead of AI. A tutor whose impact is multiplied by AI. That is where the best results are.”
How Edukko uses AI
This is the model we have built Edukko around, and I want to be specific about it because the phrase AI-powered tutoring is now attached to almost everything and means almost nothing.
At Edukko, every student works with a qualified teacher, matched to their exact curriculum and exam board. International Baccalaureate (IB) Diploma, International General Certificate of Secondary Education (IGCSE), General Certificate of Secondary Education (GCSE), A Level, Middle Years Programme (MYP), the specification named, not just the subject. That is the human part, and it is not negotiable. AI sits underneath that relationship and makes it better. Our sessions are monitored live by AI, which means a Learning Manager can see patterns a single tutor might miss across weeks of lessons. It supports the written note every family receives after each session, so parents are told plainly what went well, what needs work, and what happens next. It flags when a student is stalling on a topic before that becomes a bad grade. Between sessions, it gives students practice built around what their tutor actually covered, in the format of the exam they will actually sit.
What it does not do is teach the lesson. The judgement about what this child needs today, the decision to slow down or push harder, the conversation with a worried parent–those stay with a person. We use AI to give our teachers more information and more time, so that the human hour is spent on the things only a human can do.
What this means for parents
If your child is using an AI app at home, keep it, but check what kind it is. If it hands over answers, replace it with one that asks questions back. Use it for practice, for repetition, for the night before a test. Those are its strengths, and they are real.
Then be honest about what it cannot see. It cannot see that your child has stopped putting their hand up. It cannot see the gap between what they understand and what they produce under pressure. It cannot tell you where your child actually stands against the demands of their specific exam board in this specific term. Those things need a qualified person who knows your child, and a system that helps that person see clearly.
The question was never AI or a real tutor. It was always how to give your child the two sigma advantage Bloom described in 1984, at a price a family can afford. For the first time in my career, that is possible. It just takes both.
Find out where your child stands
If you want to know honestly where your child is against their curriculum, and what the right mix of human teaching and AI-supported practice looks like for them, we offer a free thirty-minute Check In with a senior Edukkator and a written plan the next day. No commitment, and the plan is yours to keep either way. Visit here to book, or follow me and Edukkoapp for more on how we are building this.
Read more from Nelson Barrios
Nelson Barrios, Founder, Edukko, EdTech & Education Innovator
Nelson Barrios is a Dubai-based education entrepreneur, PGCE-qualified teacher trained by the UK Ministry of Education, and founder of Edukko, one of the UAE’s fastest-growing EdTech platforms. Having taught across Spain, the United States, and the United Kingdom, he launched Edukko in 2021 to help families navigate academic pressure with expert support. Today, a network of 700+ vetted Edukkators has delivered over 10,000 sessions across eight countries, supporting students across IB, GCSE, A-Level, American, and French curricula. Nelson writes on EdTech innovation, AI in learning, and how families can raise confident, future-ready students in a rapidly changing world.
