Here’s the uncomfortable truth about AI-assisted studying: it makes you faster, less knowledgeable, and dangerously confident about both.
The numbers don’t lie
A new randomised controlled trial with 120 university students compared AI-assisted studying against traditional methods on a complex topic:
| Metric | AI Group | Traditional Group |
|---|---|---|
| Study time | ~3.2 hours | ~5.8 hours |
| 45-day retention | 57.5% | 68.5% |
The AI group finished in barely half the time. They felt productive. And 45 days later, they remembered 11 percentage points less than the students who struggled through it the old-fashioned way.
That’s not a small gap. That’s the difference between passing and failing.
The “metacognitive blind spot”
Here’s the most troubling finding: AI-assisted students consistently overestimated how much they’d learned. The speed and smoothness of AI-assisted studying creates a false sense of mastery. You feel like you understand something because the AI explained it so clearly. But understanding someone else’s explanation isn’t the same as building your own mental model.
Researchers call this “cognitive offloading” — the same mechanism behind the “Google effect,” where people remember where to find information rather than the information itself. Except with AI, you’re not even remembering where to look. You’re outsourcing the thinking entirely.
The desirable difficulty principle
Learning science has known for decades that effortful retrieval strengthens memory. This is called “desirable difficulty” — the struggle of recalling, connecting, and applying information is what makes it stick.
AI strips that struggle away. When ChatGPT synthesises your notes, writes your summaries, and generates your study guides, you’re bypassing the exact cognitive processes that create durable long-term memories.
It’s like going to the gym and watching someone else lift the weights. You’re in the building. You can see the effort. But your muscles didn’t do anything.
What actually works
The research isn’t saying “never use AI.” It’s saying how you use it matters enormously:
- AI as a quiz tool (active recall) → retention holds up
- AI as a summariser (passive consumption) → retention tanks
- Unrestricted AI access during study → worst outcomes
- AI for generating practice questions → surprisingly effective
The difference is between using AI to test yourself versus using AI to avoid thinking.
The NZ context
New Zealand universities are still figuring out their AI policies. Some ban it outright, some embrace it, most are somewhere in between. This research suggests the conversation needs to shift from “should students use AI?” to “how should students use AI?” — because the data is clear that unrestricted access during learning actively harms knowledge formation.
🔍 The Bottom Line
AI is the most powerful studying tool students have ever had. Used wrong, it’s also the most efficient way to not actually learn anything. The research is clear: if you want knowledge to stick, you need to struggle with it yourself. AI can generate the questions, but you need to fight through the answers. Every shortcut you take is a memory you don’t build.
Sources:
- RCT study (120 university students, November 2025)
- MIT/Carnegie Mellon research on AI and problem-solving persistence
- X/Twitter discussion
