Are you learning, or are you letting AI do the work for you?

My son is sixteen. He is sitting his GCSE examinations this summer, though everything that follows applies equally to anyone staring down the barrel of A levels. And, like most students I have known at that age (including the one I was myself) he is not always entirely convinced that the adults around him know anything worth listening to when it comes to exams.

So let me try a different approach.

What follows is not parental advice. It is what I know, professionally, about how learning actually works: drawn from cognitive psychology, from years in and around schools, and from watching many hundreds of students sit down to revise with the best of intentions and the wrong strategies. I am writing it here because I think it is useful. But I will not pretend I am not also writing it for him.

Most students use AI the way they used to use the highlighter: as a comfort blanket. Ask it a question, read the answer, feel reassured. Move on. The problem is that AI is designed, by default, to do the thinking for you. It summarises, explains, generates, resolves. Every interaction is optimised for your convenience. The problem is that, in revision, that is often exactly the wrong approach.

The students who get the most from AI are not the ones who avoid it, and also definetely not the ones who rely on it most heavily. They are the ones who have learned to hack it, so that it trains them, rather than the other way around: taking a tool built to remove cognitive effort and redirecting it toward creating the sort of cognitive effort that keeps them on top of their revision.

Here are five ways to do exactly that.

Here is the revision habit that almost every student gets wrong: saving it all for the end. Cramming the night before an exam feels productive because you cover a lot of ground quickly, the material feels fresh, and you feel, at least temporarily, as though you know it. But that feeling fades fast. Most of what is crammed is forgotten within days, sometimes hours.

The research on this is unambiguous. Spreading revision across multiple shorter sessions over a longer period (this is what cognitive psychologists call spacing) produces far better long-term retention than the same amount of time spent in one concentrated block. The reason is counterintuitive: when you return to material after a gap, some of it has faded, and retrieving it again takes more effort. That effort is not a sign that the strategy is failing. It is exactly what is making it work.

Closely related is interleaving: mixing different subjects or topics within a session rather than working through one thing until it feels finished. It feels disorganised. It is, in fact, more effective. Switching between topics forces you to keep deciding how to approach each one, which builds the kind of flexible understanding that exams actually require.

Left to its own defaults, AI will build you a tidy, comfortable revision schedule: one subject at a time, neat blocks, logical progression. That schedule will feel reassuring, but it will not serve you as well as one built around spacing and interleaving. So tell it exactly what you want.

Re-reading your notes feels like revising. Your eyes pass over the material, it feels familiar, you feel reassured. But recognising something when you see it is not the same as being able to recall it when you need it. Exams test recall. Re-reading trains recognition. That difference matters enormously, and it is why students can spend hours revising and still underperform.

Retrieval practice means closing your notes and trying to reconstruct what you know: writing down everything you can remember about a topic before checking what you missed, attempting past questions without the mark scheme, generating the answer on a flashcard before you flip it over. That moment when you are struggling to remember something you know you know is uncomfortable, but it is also when learning is happening. This is called productive struggle.

There is a further benefit. Retrieval practice tells you, honestly and specifically, what you do not know. Re-reading lets you skim past the gaps without noticing them. Testing yourself makes them visible. That is uncomfortable, but it is also exactly the information you need.

The temptation with AI is to use it as a faster, more responsive textbook: ask it to explain something, read the answer, feel like you have learned it. That is the wrong sequence. Flip it. Hack it.

There is an old teaching maxim that you do not really understand something until you can explain it. It turns out this is not just a maxim. It is well-supported by research. The act of putting something into your own words, as though teaching it to someone who knows nothing about it, is one of the most effective learning strategies available, and one of the least used.

The reason it works is that constructing an explanation exposes the gaps. When you read something, it is easy to nod along and feel you have understood. When you try to explain it, you find out very quickly whether you actually have. The places where your explanation stumbles or goes vague are the places where your understanding is incomplete. Those stumbles are not embarrassing. They are the most useful information you will get from a revision session.

Build this habit: after every revision session, close everything and explain what you have just studied out loud or in writing. Not a summary: an explanation. Ask yourself not just what is true, but why it is true, and what it connects to. This builds the kind of understanding where one piece of knowledge supports another, rather than a collection of isolated facts that are easily forgotten under pressure.

AI, used in its default mode, is a brilliant explainer. The hack is to reverse the roles.

One of the most consistent findings in the psychology of learning is that students are poor judges of their own understanding. Not because they are careless, but because of how learning feels from the inside. We tend to feel most confident immediately after reviewing material, which is, paradoxically, when that confidence is least reliable. The material feels familiar; we mistake familiarity for mastery; we move on.

You have probably experienced this. You revise a topic until it feels solid, go into the exam feeling prepared, and find that the questions do not behave the way you expected. The knowledge that seemed so accessible the night before is suddenly difficult to reach. Researchers call this a calibration problem, and it is one of the most common reasons capable students underperform.

The antidote is deliberate. Before attempting a set of questions, predict how many you think you will get right. Afterwards, check. Over time, this builds a more accurate picture of where you actually are. Pay particular attention to the questions you were confident about and got wrong. Those are the most important data points you have.

AI makes this harder, not easier, if you let it. When AI explains something clearly and fluently, reading that explanation can feel indistinguishable from understanding it. The confidence is borrowed, and borrowed confidence does not survive an examination. The hack is to use AI to stress-test your confidence rather than inflate it.

Everything above points to the same principle: the conditions that make revision feel most productive are very often not the conditions that make it most effective.

Ease, fluency, speed and familiarity can feel like progress. In fact, they are often signs that you are staying comfortably within what you already know. The conditions most associated with durable learning are almost the opposite: effort, uncertainty, the willingness to persevere with something that is not yet working and keep at it anyway. Psychologists call this desirable difficulty. The word ‘desirable’ is doing important work: not all difficulty is useful, but difficulty that stretches you just past what is comfortable is where lasting learning tends to happen.

In practice: revise the topics you find hardest, not the ones you feel confident about. Attempt questions you are not sure you can answer, not just the ones you know you can. Use past papers under timed conditions rather than reading through mark schemes at leisure. And when something takes a long time to work out, resist the urge to look it up immediately. That sustained effort is not wasted time. It is the point.

AI will, by default, try to make everything easier. That is what it is designed to do. The hack is to use it to do the opposite.

None of this is especially complicated. The science is solid, the strategies are straightforward, and none of them require anything more than time, consistency, and a willingness to persevere with difficulty for long enough to let it do its work.

That last part is the hard bit. It has always been the hard bit. AI has not changed what good revision looks like, it has simply made the shortcuts more tempting and the illusions more convincing.

Think of AI as a personal trainer. A good one can design your programme, correct your form, and push you to work harder than you would alone. But the trainer cannot lift the weights for you. And watching someone else exercise, however impressive their technique, will not make your muscles grow stronger.

I am writing this because I want my son to understand not just what to do, but why the effort involved is the point, not the price. Use the tools available to you. Use them well. But stay in the difficulty, check what you actually know rather than what you feel you know, and make sure that the thinking is always yours.

Because if you are not doing the thinking, you are not doing the learning. That was true before AI arrived. But understanding that feels more urgent now.

Photo by www.kaboompics.com