Study With AI Without Letting It Do the Learning

The summary of chapter six fills the screen in tidy bullets, cleaner than anything you'd write. Somewhere under your elbow are your own notes, two pages of biro.

Someone asks what the chapter says, and your eyes go to the screen.

Reading a good summary feels like understanding. But spotting a thing when it's in front of you is one skill, and producing it from nothing is a different skill.

Exams test the second, and no summary waits in the exam hall.

AI is dangerously good at the first skill, because in thirty seconds it can make any topic feel clear. The clarity is real, but it belongs to the AI until you've done something with it.

Skip the lists of a hundred study prompts, because you need one rule and a routine.

Here's the rule. The AI never produces the thing you're supposed to be able to produce. One routine covers a GCSE, a degree, a professional exam, or something you're picking up for work.

Before the first session

Check what your school, university or exam board allows. Testing yourself with AI is one thing, but having AI write your coursework is another, and the rules usually say so.

Straight after the lecture

Maybe you have messy notes and a photo of the whiteboard, or a 40-page PDF. Turning that into something usable is admin, and admin is what AI is for.

Here are my notes on [topic]. Tidy them into clear headings and bullet points. Don't add anything that isn't in my notes, and don't explain anything. If something looks incomplete, mark it with [?] so I can fill it in.

The last line matters most, because you want the gaps marked and left empty. Filling them is your job, and that's where the learning starts.

What comes back is tidier, but it's still yours.

That evening

Close the notes and type out the topic in your own words, as if to a friend who's never heard of it.

Rough is fine and wrong is fine, because a blank page is the only failure. Then hand your explanation over.

Here's my explanation of [topic] from memory. Don't rewrite it and don't give me the correct version. Tell me what I got right, what I got wrong, and what I left out. Then ask me one question that would show whether I actually understand the hardest part.

Answer the AI's question. Then explain the topic back again, and watch the list of wrong and missing things get shorter.

The parts you can't explain turn out to be the parts you'd have skimmed past in a summary.

Every session after that

Once you can explain the topic, let the AI quiz you, and stop it being nice.

Using my notes on [topic], ask me exam-style questions one at a time. Wait for my answer before showing the next one. Mark strictly, the way an examiner would, and say what a full-marks answer needed that mine didn't. Don't tell me I'm close when I'm not.

Write down what a full-marks answer needs, in your own words from class, and tell the AI to mark against that. Check your exam board's rules first. Some, like AQA, don't allow their papers in AI tools.

Keep a list of every question you got wrong. That list is your revision plan, and nothing off the list needs your time.

Left alone, an AI slides back into teaching. It asks, you hesitate, and it answers for you. Then the AI is just an examiner reading out the answers.

Give the AI the rules once, somewhere permanent. Claude and ChatGPT both have a Project, a space that remembers your files and instructions. Put your notes and the quiz rules there, so every session starts with the rules already in place.

You are my examiner, not my tutor. Ask one question at a time and wait. Never show the answer until I've given mine or typed "pass". Only ask about what's in my notes. Mark strictly. At the end of each session, list the questions I got wrong.

Mid-quiz, type "pass", never "tell me", because asking for the answer trains the AI back into a tutor. "Pass" gets you a wrong-answer mark and the next question, and the misses wait for you at the end.

In the first session, give one wrong answer with total confidence. Does the AI say "close", or quietly correct you without marking you wrong? If so, the AI isn't being strict. Tell the AI so and try again.

That takes a minute and saves a week of false comfort.

An AI will never want to quiz you as much as you need quizzing, because it's built to be helpful. To a chatbot, helpful feels like handing over the answer, so keeping it in the examiner's chair is your job, every session.

Even when told to mark strictly, the AI goes softer over a long session. When every answer starts coming back as "great", open a new chat and set the rules again.

Once a week

The questions come from your notes, so a mistake in the notes becomes a mistake in the quiz. Pick one question the AI marked you right on, and check the answer against the textbook or the mark scheme.

The AI can also mark you wrong when you're right, which happens most on unusual topics and recent material. If you're sure of yourself, the textbook settles it.

When it finally clicks

Diagrams, mind maps, flashcards and comics all help, and most AI tools will make them. The order you do things in decides whether they teach you anything.

Made before you understand the topic, a diagram is something pretty to look at. Made after, the diagram is a check.

Draw yours first, badly if need be. Then ask the AI for its version, spot the differences, and where they disagree, work out which one's right.

Flashcards work the same way. Write the questions yourself from your wrong-answer list, and let the AI check only that the answers on the back are correct.