The best way to take lecture notes: by hand, typed, recorded or AI

There are four ways to take lecture notes: by hand, by keyboard, by recording, and by having an AI take them while you listen. Each one trades something, and the trade is not the one most people assume. This post starts with what notes are actually for, goes through the four methods and the research on each, covers the three classic note formats, and ends with the method we would pick and the routine that makes it work.

The two jobs notes do

Since Di Vesta and Gray's work in the early 1970s, researchers have split note-taking into two functions. The encoding function is what happens while you write: deciding what matters, compressing it, putting it in your own words. The external storage function is what the notes do afterwards: they are the thing you revise from [1].

The point of naming them is that the four methods trade one against the other. Handwriting is strong on encoding and weak on storage, because you miss things and cannot search paper. Recording is perfect storage and no encoding at all. The right method is the one that gives you both, and no single method does on its own, which is why the routine at the end matters more than the choice.

The four methods

By hand

The case for handwriting is the encoding function. You cannot write as fast as a lecturer talks, so you have to choose, and choosing is thinking. Mueller and Oppenheimer's 2014 study, the one every article cites, found that students who wrote by hand answered conceptual questions about a lecture better than students who typed, and that the typists had written more words and more of them verbatim [2]. The paper's title, "The pen is mightier than the keyboard", did the rest.

Two direct replications complicate it. Morehead, Dunlosky and Rawson in 2019 found the handwriting advantage much smaller and not reliable [3]. Urry and a large team of co-authors in 2021 ran the original study again with a larger sample and found the typists did write more verbatim, as before, but that this did not translate into worse test performance [4]. So treat the handwriting advantage as a tendency, not a law. What survives the replications is the mechanism: compressing in your own words helps, and whichever method lets you do that is the one that encodes.

The costs of handwriting are real. You miss what is said while you write the previous point. Diagrams, formulas and anything the lecturer draws come out rough or not at all. Paper is not searchable, and a term's worth of pads is hard to revise from.

Typed

Typing's advantage is speed and storage: you capture more, the notes are searchable, and they sync. The cost is that speed removes the need to choose. A fast typist can keep up with a lecturer, which turns note-taking into transcription, and transcription is the thing the encoding function is not.

The other cost is the laptop itself. Sana, Weston and Cepeda found that students who multitasked on a laptop during a lecture scored lower on a test afterwards, and so did the students sitting near them who could see the screen [5]. That is a discipline problem rather than a typing problem, but it comes with the method.

Typing works well when you force the compression back in: type in fragments rather than sentences, keep a running "one line" summary at the top of each section, and never type what is on the slide.

Recorded

Recording wins the storage function completely and loses the encoding function completely. You listen properly, which is a real benefit for a lecturer who goes fast or a subject you find hard. Afterwards you have the whole lecture.

In practice almost nobody replays a recording. Ninety minutes of audio to find a two-minute explanation is a bad trade, and the summary that recording apps generate is someone else's compression, not yours. Recording is a good safety net and a bad primary method. It also needs permission at many institutions, which the FAQ below covers.

AI notes

The newest method: the lecture is recorded, and an AI builds notes from it as it goes. In Live Notes the notes land on a board every minute or two, as headings, points and diagrams, while you listen. Afterwards the recording is saved to the board as a source, so any note can be traced to the moment it came from.

This gets you the storage function and the listening benefit of recording, with the summary already made. What it does not do, on its own, is encode. The notes are the AI's compression. If you never touch them afterwards, you have a tidier version of the recording problem: good storage, no learning. The method works when you add the twenty minutes of rework described below, which is the compression step from handwriting, done once at the end instead of under time pressure.

It also lets you do something none of the other methods do: type a question onto the board the moment you get lost, without losing the next two minutes of the lecture to it, and have it answered from what the lecturer just said.

Cornell, outline or map: the format question

Whichever method captures the lecture, the notes end up in a format, and three have stood the test.

• Outline. Headings, sub-points, indented. Matches how most lecturers structure a talk, and is what an AI note-taker produces by default. Weak at showing how points connect across sections.
• Cornell. A wide column of notes, a narrow column of cues and questions beside it, and a summary at the bottom of each page. The cue column is a built-in recall test, which is why it keeps working. Easy to add to any notes afterwards: write one question per section in the margin, and the one-line summary at the end.
• Mapping. The topic in the middle, branches out to its parts, arrows between parts that relate. Best for subjects where the relationships are the content: biology, history, law, anything with mechanisms. Hard to do live, good to do afterwards. A board is the natural place for it, because the parts can be moved as you understand them better.

Our view: capture as an outline, because that is what a lecture is, and rework into Cornell cues and a map afterwards, because that is what revising needs.

Our pick, and the routine

Let the AI take the notes, and spend the lecture thinking. In Live Notes, press record and Bord builds structured notes on the board while you listen. When the lecturer says something you do not follow, type the question onto the board rather than losing the next two minutes to it.

Then, the same day, do the part that is yours. This is the step that turns storage into encoding, and it takes about twenty minutes:

1. Read the board once, top to bottom, and cut what does not matter.
2. Fix anything the AI heard wrong. Deciding whether a note is wrong is itself a recall exercise.
3. Write the one-line version of each section in your own words, in a note beside the heading. This is the Cornell summary, and it is the compression step from handwriting.
4. Write one question per section in the margin. These are your Cornell cues, and your revision test.
5. Where the lecture was about how things connect, pull the parts into a map on an empty part of the board.
6. Ask Quizbook for a short quiz on the lecture and answer it before you close the board.

The reason for the same day is forgetting. The classic forgetting curve is steepest in the first day, and reviewing then is the cheapest point at which to stop it. Spaced practice, revisiting the material at increasing intervals, is one of the two techniques with the strongest evidence in the study-skills literature [6], and the same-day rework is the first interval.

If you missed something during the lecture, Tutor can explain it from the recording and cite the timestamp, so you hear the lecturer's own version of it rather than a generic one.

If you prefer paper

Keep writing by hand. The encoding benefit is real even if it is smaller than the 2014 study suggested, and some people think better with a pen. Then fix the storage problem: photograph the pages afterwards and drop the photos on a board as a source. Bord reads handwriting as it reads a scanned page, writes the notes up beside the originals, redraws the diagrams, and makes the whole thing searchable and quizzable. You keep the benefit of handwriting and lose the problem of never finding the page again.

Questions