How to tutor your kids with AI (and why you should)

You sit down to help your kid with homework. Ten minutes in, you are explaining something you half remember from many years ago, they are staring at the page, and both of you are getting frustrated. If that sounds familiar, this post is for you.

I have been in this situation many times with my own kids, now in their teens. My daughter will walk in from school, drop her bag, and deliver the news on her way to the fridge.

Her: What do we have for lunch? Ohh, by the way, we have a math test next Friday.

Me: Okay. What did the teacher give you for homework?

Her: To study some pages from the book. Forty-two to fifty-one, I think. The ones with the systems of equations.

Me: That’s it? Anything else, like exercises or notes?

Her: That’s it.

Me (thinking): A date, a subject, a page range. That’s all the information I have. What is actually on those pages? Do they use the same method for solving systems of equations as I did? Should I check the parents’ groups on Viber or WhatsApp to see what others are saying? Is she familiar with all the prerequisites? How do I prepare her for the test?

The above scene, in multiple variations, has happened throughout the years of elementary, middle and now high school, and many parents have probably experienced it too. Before ChatGPT marked the new age of advanced AI chatbots, parents used to search the internet for additional material, exercises with solutions and videos explaining concepts in plain language. Now they turn into AI for help.

Education has entered a new age with AI, and homework at the kitchen table is where your family will meet it first. Your kids are probably already using ChatGPT to do their homework and study, and most likely they are using it the wrong way. This post is a practical guide to reading school material together with your child and an AI assistant. Not to get faster answers but to build real understanding, and to teach your young ones how to use AI for studying properly, in ways that enhance their skills rather than hamper them.

Why teaching your own kid is hard

Every parent who has tried to tutor their own child knows the problem. It usually comes down to a few things.

You are not familiar with the material. Nobody remembers all of it. You have forgotten most of what you learned in school.

The material has changed. Curricula evolve. The way your kid’s book presents a topic is often different from the way your books did.

It gets harder every year. As the grades go up, so does the difficulty. At some point the material outruns what you can explain off the top of your head.

You are both emotionally attached. This is probably the most important obstacle. Tutoring your own kid is hard precisely because you love and care for them. You get frustrated faster when they do not get a concept, because you want it to come easy to them. And they tend to put in less effort with you than with a teacher, because they are used to you doing things for them.

As a parent, you have to teach your kid a great many things: how to ride a bicycle, how to behave and respect others, your values, and how to deal with hardship. Now that AI has made intelligence accessible to everyone, helping them with academic subjects has become a little easier. But how to use these new tools well is far from obvious.

For the first time in history, every parent can be a meaningful tutor to their kids, regardless of their own academic background. All it takes is the proper use of the AI tools now available.

A plain chatbot is not enough

Tools like ChatGPT, Claude and Gemini have general knowledge. They are not aware of your kid’s schoolbook, although they are likely familiar with all the concepts it covers. The large language models (LLMs) that power those chatbots are trained on a snapshot of the public internet, which includes, among other things, web pages, books, Wikipedia, news, code repositories, and forums like Reddit and Stack Overflow. This is supplemented with licensed and human-curated data, but not with any particular student’s specific schoolbook or a teacher’s classroom materials.

Ask a question and you get a generic answer. It may use different notation, different terminology, or a different approach than the one the teacher expects and the book teaches. Give a problem from the book to be solved, and it will solve it using a different methodology than the book dictates. That confuses a young student more than it helps.

What about AI tools like Google’s NotebookLM and Anthropic’s Claude Projects, where you can upload your whole book? These tools can turn a chapter into a video, a podcast, or a deck of flashcards. These artifacts are nice to have, but they are also passive. Your kid consumes them the way they consume everything else on a screen, and passive consumption contributes very little to understanding. It can even harm it.

There is another cost here that matters more. These artifacts are so easy to produce that the temptation is to generate them and hand them over. When you do that, you give away the most valuable thing in the whole process which is the interaction between you and your kid while learning.

Understanding cannot be handed over

Around 100 CE, the Greek biographer Plutarch, in his essay On Listening to Lectures, wrote a version of a line that perfectly describes what good tutoring should be like:

“The mind is not a vessel to be filled, but a fire to be kindled.”¹

His point was that real learning is not the pouring of facts into a passive container; it is something to be ignited in the learner, that then develops on its own.

This is the heart of the claim I made in my previous post, What does it mean to understand in the AI era?: understanding is not a thing that can be delivered to you. It is built, effortfully, inside your own head. A summary, a podcast, a video, a deck of flashcards, every one of these is a product of someone else’s understanding, and consuming it fills the vessel without lighting the fire. Your kid ends up able to recite what the chapter said, feeling as though they understood, right up until they are asked to explain it and cannot. The AI hands over the output of understanding, such as a fluent explanation, while quietly skipping the process that would have built it.

That is why the artifacts these tools produce so effortlessly are a trap, not a shortcut. The goal is to stay on the right side of that line: use the AI to kindle your kid’s thinking, never to do the thinking for them.

Why read with your kids with AI

Done properly, reading with your kid and an AI assistant gives you four things.

You connect with them. Studying together is time spent together. The learning process itself becomes a shared activity instead of a nightly battle. While this was not always possible before, now, with AI, any parent can be a tutor.

You get visibility into their academic progress. You see what they are actually studying, chapter by chapter, exercise by exercise, instead of finding out at report card time. This means you can spot gaps early in subjects like maths and language, and take action while it still matters. And you do it continuously, rather than at a few fixed points in the year like the middle and the end of the term.

You help them develop the crucial skill of properly using AI. Generative artificial intelligence is a powerful technology that is radically transforming the way we do knowledge work and scientific research. But this will not happen through AI replacing those human activities; it will happen through AI enhancing humans to do them at greater scale and depth.²³ Combining purely human attributes such as judgement, curiosity, agency and insight with the complementary strengths of AI in handling vast amounts of knowledge gives a real advantage to those who have developed the skill.⁴ It is far easier for your kids to build that skill early, during the school years when most of their knowledge training takes place, and to sharpen at the same time their judgement about where and why AI fails.

You accelerate deep understanding. Mastering a topic typically involves understanding a series of concepts that build on each other, so if your kid gets stuck at one, it becomes hard to move on. A book is a static exposition of a topic, and no matter how good and illuminating it is, it cannot suit every reader with a different level of background knowledge. It does not carry examples, exercises, illustrations, or alternative explanations for every single concept presented. With AI you can generate as many of those as you need, using the book as the ground text, and you can also fill in any prerequisite knowledge the book leaves out.

In short, there are two goals. Deepen your kid’s understanding of the material. And train them in the proper use of AI. Everything below serves those two goals.

The system, step by step

The proposed workflow has two phases, and it involves you, your kid and the AI assistant. First, a setup phase, where you establish what is going to be studied and prepare yourself. Then a study phase with your kid, which is iterative and repeats every session. The diagram below shows the steps, and who is involved at each one.

1. Choose the right tool

You need a PDF or EPUB version of your child’s actual physical book. Pick a tool that lets you upload the whole book, so every answer is grounded in the material your kid is studying. NotebookLM, Claude Projects, or Meletium all qualify. A generic chat window does not. For a detailed head-to-head of these tools on a real textbook, see AI Tools for Studying: ChatGPT, Claude, Gemini and Meletium.

Pay special attention to how precise the tool is with page references. Both you and the AI assistant will use page numbers to point to parts of the physical book, so accurate page numbers are crucial. Many AI tools fail to distinguish between the page numbers of the PDF document and the actual ones printed on the physical book. Meletium is one of the most accurate in this respect, because when a book is uploaded it first runs a preprocessing step that extracts the book’s logical structure.

2. Find out what the study material is

Start every session by asking your kid exactly what is being studied. Which chapter, which sections, which pages. Kids rarely have the full picture, so cross-check what they tell you. Handouts the teacher gave out, notes in their notebook, and the parents’ groups on Viber or WhatsApp are all good sources. Everything that follows in the session references this material, so it’s worth a couple of minutes to get an accurate account of it.

3. Go through the material yourself

Your objective here is not to fully understand the material, but to get a good idea of what it is about. You need this so you can guide the main study session in the next step.

Use the AI to get a quick summary of the chapter or pages, and read through the main ideas until they make sense to you. This should be fast, a matter of minutes. Some tools, like NotebookLM and Meletium, offer mind maps or graphs that let you browse through the main concepts easily.

Note the asymmetry. A summary is fine for you, because you only need orientation. Your kid needs mastery, and mastery does not come from summaries.

4. Read and work through the concepts

This step, together with the next three, makes up the study phase, and it should be done in this sequence, iteratively, until a satisfactory level of understanding is reached.

You, your kid, and the AI assistant will go through the main concepts of the material together. Some of the things you can ask the assistant to do include:

• explain points you find difficult
• give numerical examples
• provide illustrations, such as drawings, where appropriate
• rephrase a sentence or paragraph in simpler wording

This step is the main interaction you will have with the AI tool, and it is where the tools differ most. Some, like Meletium, break the concept down to first principles when asked to explain something, and open a conversation with the user, asking what they already know before proceeding. Others prefer to give one long answer that contains all the relevant information.

The overall design of the tool also matters a great deal in this step. Meletium and ChatPDF, for instance, offer a view where the book and the chat sit side by side, as well as a dedicated read-only mode that shows only the book page, where you can highlight any part to prompt the AI. Others, like NotebookLM and Claude Projects, keep their sources separate and offer no such view.

You have to be present in this step, especially with younger kids. Children tend to anthropomorphise these tools, since they are built to be convincing. They will treat the AI as a person and trust it as one. Your job is to be the adult in the room and to emphasise that this is just a tool. Powerful as it is, it makes mistakes and cannot judge its own answers. That part is up to us humans.

5. Quick real-time quizzes

Quick quizzes are a great way to do progress checks. Six multiple-choice or true/false questions are enough to see whether a concept has landed. The tool should be able to generate the questions in real time and grade them as well. Given their purpose, these questions are best suited to factual knowledge or small computations.

6. Create exercises with solutions

Whereas quizzes are great for progress checks, they do little for deep understanding of the material or preparation for an actual in-class exam. Customised exercises with worked solutions do that. Solving exercises has always been an important part of academic study, which is why most textbooks offer a set of problems of varying difficulty at the end of each chapter. Very few, however, provide solutions, and in the best case they give only short answers to selected problems.

Have the AI generate an exercise set from the given material and print it. Each exercise and its solution should reference the corresponding pages in the book, so your kid knows where to look. Aim for a book-quality printout, with the exercises and their solutions kept separate.

Be very specific when you ask the AI assistant to generate the exercises. State the pages the exercises should be based on, the difficulty level, the types of questions you want (computational or theoretical), and whether or not you want worked solutions. Save each exercise set together with its solutions, since these are now an integral part of the context you have built for this book, and you may want to revisit them as your kid studies.

7. Solve the exercises

Have your kid solve the exercises with pen, paper and the book, and no device anywhere near them. This matters for two reasons. It forces active work instead of passive reading, which is crucial to understanding, as we showed in the previous post. It also closely resembles the real exam, so the format holds no surprises later. Afterwards, gauge how comfortable your kid is with the material, and repeat steps 4 through 7 if needed.

The following two habits should be practised in every single session.

Always reference the book. Every claim the AI makes should point back to the book, with page numbers and, where relevant, sections. If it does not, ask it to.

Always remind them that AI makes mistakes. Teach your kid to judge answers instead of accepting them, and to challenge the AI with common sense.

The same applies beyond the book. If the AI cites external sources, check them. Even the top models will confidently cite a paper from some obscure journal simply because their search tool returned it. Teaching a kid to ask “where does this come from, and is the source reliable?” may be the most valuable lesson in the whole exercise.

The point of it all

The tools will keep getting better. The temptation to let them do the work will keep getting stronger. What will not change is the fact that understanding has to be built in your kid’s head, by your kid, with effort.

Your role is to set up the conditions for that to happen. Pick a tool grounded in the actual book. Understand the material first. Sit together. Make them work on paper. Teach them to judge the AI’s answers, and to use it to grow their understanding rather than replace it.

---

This is part of a series of essays on understanding, comprehension, and learning with AI. The first post, What does it mean to understand in the AI era?, lays out the research this workflow is built on. The third, AI Tools for Studying: ChatGPT, Claude, Gemini and Meletium, compares four AI study tools on a real textbook.

References

Footnotes

1. The line is the English rendering of a thought from Plutarch’s essay On Listening to Lectures (De auditione), written around 100 CE. Plutarch’s Greek is closer to “the mind does not require filling like a bottle, but rather, like wood, it only requires kindling”. ↩

2. Anthropic. “The Anthropic Economic Index report: New building blocks for understanding AI use,” January 15, 2026. ↩

3. “Why AI cannot do good science without humans,” editorial, Nature, May 19, 2026. https://doi.org/10.1038/d41586-026-01551-3. ↩

4. David Autor and Neil Thompson, “Expertise,” Journal of the European Economic Association 23, no. 4 (2025): 1203 to 1271. https://doi.org/10.1093/jeea/jvaf023. ↩