You have three ways to end a lecture with notes: write them by hand, type them, or let an AI produce them from the slides or a recording. Each is faster than the last. The research on the first two is mature. The research on the third is new, and more interesting than the headlines suggest.
Handwriting vs typing: a small, real edge
The famous study is Mueller and Oppenheimer (2014): students taking notes on laptops did worse on conceptual questions than students writing longhand, apparently because typists transcribed while writers rephrased. It is worth knowing that a direct replication by Morehead, Dunlosky and Rawson (2019) found no consistent difference, and called a verdict premature.
The best current answer is the meta-analysis by Flanigan and colleagues (2024): across 24 studies of college students, handwritten notes led to higher achievement, g = 0.248. Small but reliable. Typists wrote far more (g = 0.919), so volume is clearly not what helps. A brain imaging study by Umejima and colleagues (2021) found richer brain activity when people recalled information written in a paper notebook, though overall accuracy was similar and two authors worked for a consulting firm, so we would not lean on it.
Why effort is the point
The handwriting edge is one case of a broader rule. Bertsch and colleagues (2007) pooled 86 studies on the generation effect: information you produce yourself is remembered better than information you only read, by about 0.40 standard deviations. And Bjork, Dunlosky and Kornell (2013) explain why students resist this: fluent, easy study feels like learning, while effortful study feels like failing, and the feelings are backwards.
What happens when AI writes the notes
The strongest study so far is a field experiment with nearly 1,000 high school maths students (Bastani and colleagues, 2025, PNAS). During practice, students with a plain ChatGPT-style assistant scored 48% higher. Then the AI was taken away for the exam, and they scored 17% worse than students who never had it. A second group used a tutor version that gave hints instead of answers: their practice scores rose 127%, and the exam penalty mostly disappeared.
Two more results point the same way. Melumad and Yun (2025), across seven experiments with 10,462 adults, found that people who learned a topic from an AI summary came away with shallower knowledge than people who worked through web links, even when the facts were the same. Stadler, Bannert and Sailer (2024) found students researching with ChatGPT reported less mental effort and produced weaker arguments than students using a search engine.
How to use AI notes without losing the learning
Take your own rough notes first
By hand if you can, in your own words. Messy is fine. This is the generation step, and no tool can do it for you.
Let AI do the restructuring
Turning ten messy pages into a clean summary, a visual or a set of flashcards is formatting work. Handing that over costs you little and saves hours.
Argue with the output
Compare it with your notes. What did it leave out? What did it get wrong? Correcting an AI summary is an effortful task, which is exactly what you want.
Close it and retrieve
Quiz yourself with the AI switched off. The Bastani study shows why: what counts is what you can do without it.
We build an AI notes tool, so here is how we read this research against our own product. If you paste a chapter into PDF to study notes, glance at the result and close the tab, the studies above say you have learned little. The parts of VisualNote AI that line up with the evidence are the ones that make you work: photographing your own handwritten notes and having them redrawn, fixing the visual through chat, and the quiz and flashcards at the end. Use it as the restructuring step, not as a replacement for the thinking.
Frequently asked questions
Is handwriting really better than typing notes?
Slightly. A 2024 meta-analysis of 24 studies found a small advantage for handwriting (g = 0.248) among college students. If typing lets you keep up and you review your notes afterwards, the gap is small enough not to worry about.
Do AI-generated notes hurt learning?
They can, if they replace your own processing. In a PNAS field experiment, students who practised with a plain ChatGPT assistant scored 17% worse on the exam than students with no AI. Students using an AI that gave hints instead of answers did not show that drop.
What is the best way to combine them?
Write rough notes yourself, use AI to restructure them into a summary, visual or flashcards, correct what it produces, then test yourself without it.
Does it matter if I write on a tablet instead of paper?
The evidence is thin. One imaging study favoured paper for a calendar task, with similar overall accuracy. Writing in your own words matters more than the surface.
