ChatGPT for Studying: Does AI Hurt Memory? How UAE Students Can Use It the Smart Way

It is 9:40 p.m. in Sharjah. A Year 10 student finishes his maths homework in twelve minutes. Every answer is correct, neatly laid out, with working shown. His mother is impressed.

The next morning, he sits a short class test on the same topic and scores far below his homework. Nothing about him changed overnight. The difference was that, the night before, ChatGPT had been open in the next tab.

Families across the UAE are asking the same question: is ChatGPT for studying a shortcut to better grades, or a shortcut around learning? The honest answer is that it can be either, and the difference comes down to one thing: who does the thinking first.

The best evidence so far comes from a study by Hamsa Bastani and colleagues, published in the Proceedings of the National Academy of Sciences in 2025. They worked with a high school in Turkey and gave nearly a thousand maths students one of three set-ups during practice sessions:

• GPT Base: a tool that behaved like standard ChatGPT.
• GPT Tutor: the same underlying model, but set up with safeguards: it gave incremental hints and common-mistake warnings rather than full solutions.
• No AI: a control group.

In a later exam, taken with no AI at all, the results split sharply.

Two lessons stand out:

• Better performance with AI is not the same as learning. Practice scores soared, but the knowledge did not travel with the students.
• Design matters. The harm was largely avoided when AI behaved like a tutor, giving hints and making the student work, rather than an answer machine. The researchers also found that students who used AI tended to overestimate how much they had learned.

A note of caution. This was one school, one subject and one version of the technology, so results may differ elsewhere. But it matches what decades of memory research would predict.

Memory is built by effort, and AI is very good at removing effort. Four mechanisms are worth knowing.

It skips retrieval. The act of pulling information out of your head is what strengthens it, a finding known as the testing effect (Roediger and Karpicke). If AI supplies the answer, there is nothing to retrieve. Our guide to The Illusion of Competence explains why recognising an answer feels like knowing it.

It skips generation. Information you produce yourself is remembered better than information you simply read, known as the generation effect (Slamecka and Graf). A neatly written AI answer is read, not generated.

It encourages cognitive offloading. People naturally lean on external tools to reduce mental effort, and research on cognitive offloading (Risko and Gilbert) suggests this can reduce what we remember ourselves. That is fine for a phone number. It is a problem for a chemistry mechanism that will be tested in a silent hall.

It creates a feeling of understanding. A clear, confident explanation feels like learning. Students who read an AI explanation often feel they have understood, but a feeling of fluency is not the same as being able to reproduce the idea. This is the same trap that makes re-reading feel productive.

There is also a practical problem: AI makes mistakes. It can state wrong answers confidently, and a student who has not tried the problem first is poorly placed to notice.

The same mechanism sits behind the pattern in The Scaffolding Trap: Why Students Freeze on Exam Questions. A tool that is always available is the most powerful scaffold a student has ever had, and the exam removes it.

AI is not going away, and the UAE is leaning in. The Ministry of Education introduced AI as a formal subject across public schools from kindergarten to Grade 12 from the 2025–26 academic year, and some private schools have introduced it too. Students are being taught to understand and use AI, which makes careful habits more important, not less.

Meanwhile, the rules for assessments are firm:

• Supervised exams are AI-free. Internet-enabled AI tools are not allowed in written IGCSE and A-Level papers.
• Coursework has strict rules. Cambridge and Pearson allow only limited, acknowledged AI use in some coursework, and treat unacknowledged or inappropriate use as malpractice. Rules differ by board and subject and change over time, and Cambridge notes that centres and teachers hold primary responsibility for spotting inappropriate use.
• Your school's policy comes first. Many schools have their own AI policy that is stricter than the board's.

In other words, in the places where marks are decided, a student's own memory is still the only tool. See our guide on independent thinking in A-Level tutoring for why this matters more at A-Level.

Not all AI use is equal. A simple traffic light helps families sort it.

1. Try first, for at least ten minutes. Before opening AI, attempt the question on paper. Struggling for a few minutes is not wasted time: it is where memory forms. If you get stuck, note exactly where.

2. Ask for hints, not answers. Use a hint ladder: first a nudge, then a hint, then a worked step, and only after another attempt, the full solution. This is the design that protected learning in the research.

3. Let AI quiz you. The best use of AI for revision is retrieval practice: it asks, you answer from memory, then it gives feedback. That is far closer to an exam than reading an AI summary.

4. Teach it back. Explain the idea in your own words, then ask AI what you missed. Saying it aloud or writing it out forces generation, and the feedback fills the gaps.

5. Check what it tells you. Treat AI as a fallible tutor, not an authority. Check facts against your textbook, notes and the official mark scheme. In maths, check answers by substituting them back (see our guide to silly mistakes in maths exams).

6. End with a no-AI test. After a study session, close the laptop and answer five questions cold. If you cannot, the learning has not stuck. This is the closest home version of the exam.

Students often ask for better “ChatGPT prompts for revision”. These are built to keep the thinking with the student. Replace the brackets with your subject, board and topic.

Always verify against the official mark scheme. AI marking is a useful second opinion, but it is not a substitute for the board's own marking guidance.

You do not need to understand the technology. You need to notice the gap between homework and understanding.

Signs AI may be doing too much of the thinking:

• Homework is polished, but the child cannot explain it aloud
• Class-test results are far below homework results
• Your child cannot do a similar question when you change the numbers
• A sudden jump in the quality of writing that does not match their usual voice
• Reluctance to show rough working or the first attempt

What to say:

A simple family agreement helps: try first, ask for hints, quiz yourself, and finish with a no-AI check. Researchers and educators suggest a short reflection after any AI-assisted work: where did I get stuck, what hint did I ask for, what did I use, and can I now solve a similar question alone? Our guide on how much parents should help with homework covers the wider balance.

Where marks count towards a qualification, honesty matters most.

• Check your school's AI policy and the exam board's current guidance before using AI for any assessed work.
• Declare any permitted use in the format the school or board requires.
• Never submit AI-generated text as your own. Passing off AI work as a student's own is treated as malpractice, with consequences that can include lost marks or disqualification.
• Keep drafts and notes. Schools often check progress at intermediate stages, and your own drafts show the work is yours.

If you are unsure whether a use is allowed, ask the teacher before you do it.

Sometimes heavy AI use is not laziness but a way of managing a feeling. A student may reach for AI immediately because:

• They feel anxious about getting it wrong. See our guide to maths anxiety in teenagers.
• They find it hard to start. See why teenagers procrastinate on revision.
• They are overwhelmed or exhausted. Sleep and workload both matter (see Sleep and Memory).
• They have a genuine gap and no one to ask.

In each case, banning the tool treats the symptom. Addressing the cause usually reduces the reliance. If a gap is the reason, a short diagnostic with a tutor or the subject teacher can rebuild the foundation, so AI becomes a coach rather than a crutch.