Why AI Feels More Helpful Than It Is — And How to Use It Right

Here's something counterintuitive about AI and learning: the more it feels like it's helping, the less you might actually be learning.

A new study from Cambridge University Press & Assessment and Microsoft Research put this to the test — and the results are worth every parent and student understanding before exam season.

The Study

Researchers ran a randomised experiment with 405 students aged 14–15 in England, testing three different study conditions:

• LLM only — students used an AI assistant to help with their reading
• Note-taking only — students took their own notes while reading
• Note-taking + LLM — students combined both approaches

They then measured actual comprehension and memory retention — and separately asked students how much they felt the AI had helped them.

The findings were stark:

• Note-taking alone significantly outperformed LLM-only for comprehension and retention
• Note-taking combined with LLM use also significantly outperformed LLM-only
• Students who used the LLM alone felt it helped them the most — but objectively learned the least

That last point is the crucial one. Students felt more confident, more helped, more productive — while actually retaining less. This is what researchers call the perceived vs actual learning gap.

Why This Happens

When an AI answers your question fluently and completely, your brain registers that the problem is solved. The question feels resolved. The uncertainty is gone. That relief is interpreted as learning — but what actually happened is that the AI learned nothing while you did the cognitive work of reading the answer rather than producing one yourself.

Educational psychologists call the harder, slower kind of learning "desirable difficulty" — the struggle of trying to retrieve something from memory, explain a concept in your own words, or work through a problem without immediately being given the answer. That struggle is uncomfortable. It doesn't feel productive. But it's exactly what builds durable knowledge.

AI, by its nature, removes the struggle. And that's precisely when it stops being a learning tool and starts being a crutch.

The Note-Taking Insight

The Cambridge finding about note-taking is worth pausing on. Taking notes by hand or in your own words forces you to process and restate what you're reading — not just receive it. That active reformulation is a form of desirable difficulty. It's why students who take their own notes consistently outperform those who passively review pre-made notes or rely on AI summaries.

The best outcome in the study came from combining both — using AI alongside your own active note-taking. The AI adds value when it's a tool you interrogate, not a source you passively receive.

What This Means for Writing Practice

This research has specific implications for students preparing for NAPLAN, selective school exams, HSC, and VCE.

The trap to avoid

Using AI to generate or complete your writing doesn't build writing skill. If you ask an AI to write an introduction for you, or to fix your paragraph, you haven't practised the skill of writing introductions or structuring paragraphs. You've outsourced the very work that builds the ability.

This is especially dangerous in exam preparation, because exams are closed-book. When you sit down to write in 30 minutes for a selective school test, none of the AI-generated phrases you've read will be there to help you. Only the writing muscle you've actually built will show up.

How to use AI tools in a way that actually helps

The research points to a clear principle: use AI to challenge your thinking, not to replace it.

In practice for writing, that means:

• Write a draft yourself, then use AI feedback to identify what to improve
• Ask the AI why a particular sentence is weak, then rewrite it yourself
• Use AI to generate a practice prompt, then write under timed conditions
• After getting marked, ask the AI to explain one criterion in depth — then try a paragraph targeting that criterion before moving on
• Use the feedback to identify your weakest area, then deliberately practise that specific skill

• Asking AI to rewrite your essay or fix your paragraphs for you
• Reading AI-generated exemplars without first attempting the writing yourself
• Using AI feedback as a final destination rather than a starting point for more practice
• Substituting AI tutoring for the act of writing

The pattern is consistent: produce first, then use AI to improve. Not the reverse.

The Bigger Question

There's a question buried at the end of the Cambridge article that's worth sitting with:

When using AI, how do you tell whether you're actually learning rather than just getting to the answer faster?

For students and parents, a practical version of this question is: after using an AI tool, could you do the task again without it?

If you can write a better essay introduction tomorrow without AI help than you could yesterday, the AI is working as a learning tool. If you can only write a better introduction with AI assistance, it's working as a crutch. Both have their uses — but they're not the same thing, and confusing them is costly when exam day arrives.

For Parents

If your child is using AI writing tools to prepare for NAPLAN, selective school, HSC or VCE, the research suggests a simple rule of thumb:

The goal is always to build the skill, not just to produce the output. Tools that give specific, rubric-aligned feedback on essays your child has written themselves — and that challenge them to improve through their own effort — are the ones backed by the research.