How to Take Lecture Notes with AI (Step-by-Step, 2026)

By Leo Chen · Last reviewed September 2026 · Written from first-hand testing · How we test

I spent my first semester writing notes the way everyone does: frantically, badly, and with no idea what I was actually writing down. Three weeks before finals I had four notebooks of sentences I could not read and could not use.

The fix was not "use AI to write my notes for me". That produces neat summary documents that feel like revision and teach you nothing. What actually worked was handing the two mechanical parts of note-taking to AI - transcription and formatting - and keeping the part that requires my brain, which is deciding what I did not understand.

Here is the exact workflow, including the parts that went wrong first.

The 30-second answer

Three steps. Before class, generate five questions from the previous lecture so you know what to listen for. During class, write only keywords and questions by hand and let a recorder capture the rest. After class, the same day, feed the transcript and your keywords to NotebookLM and make it produce structured notes plus practice questions - then compare its version against your handwritten page and treat every difference as a gap to fix.

The third step is where all the value is, and it is the one everybody skips.

Step 1 - Before class (five minutes)

Open the previous lecture's material in NotebookLM and ask for something specific, not a summary:

Based on this material, list five questions a professor would ask on an exam, and mark which parts of the lecture I would need to have understood to answer them.

Then take that list to the lecture and listen for those five answers. This one change did more for my comprehension than any note-taking app, because it converted a passive hour into a search task.

I skipped this step for most of first year because it felt like extra work for five minutes. It is not extra work. Without it you spend the lecture deciding what matters, which is exactly when you miss what the lecturer is saying.

Step 2 - During the lecture

Two things happen at once, and one of them is not writing.

Write by hand, but only three things:

• Anything the lecturer repeats
• Anything they say will come up again, in any wording
• Anything you do not understand, written as a question with a big question mark

That is the whole list. Everything else - definitions, dates, diagrams, numbers - is the recorder's job, and this is the part people find hardest to trust at first. But the point of writing is not to have a record. It is that deciding what to write is what keeps your attention in the room.

Record on your phone, or with Otter.ai if the room is difficult. Otter gives you a live transcript with timestamps, which is worth the free tier on its own for one specific reason: when your handwritten question mark appears next to a timestamp, you can replay that ninety seconds instead of guessing.

Two honest limitations. Otter's transcription falls apart with heavy accents, mumbled delivery and technical vocabulary it has not seen - my thermodynamics lecturer produced some genuinely entertaining garbage. And a recording is not a substitute for being there, because it does not capture the lecturer pointing at a slide and saying "this bit".

Step 3 - After class (twenty minutes, same day)

This is the step that makes the other two worth doing.

1. Drop the transcript into NotebookLM alongside the lecture slides
2. Paste in your handwritten keywords and questions as a separate note
3. Ask for structured notes, then ask for exam-style questions with the answers hidden

The important part is what you do next, and it is not reading the output.

Diff the AI summary against your handwritten page. Every place where your page has a question mark and the summary has a clean answer is a gap you now know about - go and fix it while the lecture is still in your head. Every place where the summary contains something you did not write down is material you missed entirely.

That comparison takes about five minutes and it is the only part of this workflow that improves your understanding. Everything else is admin.

Then close the loop into revision. Take the structured notes and generate recall material:

• Knowt turns notes into flashcards automatically if you want speed
• Anki if you are serious about spaced repetition and want to still remember this in three months
• NotebookLM's own question generation if you just want to test yourself tonight

The tools, and what each one is actually for

If you only adopt one: NotebookLM. It is free with no usage anxiety, and it is the only tool here that works from your own materials rather than the open web - which matters, because a chatbot answering from the internet will confidently invent things your lecturer never said.

Three mistakes I made so you do not have to

Recording without asking. I recorded a seminar in second year assuming it was fine and only found out later it was not, which is not a conversation you want to have retrospectively. It takes one email before the first lecture. Rules differ by institution and some departments have banned it outright, so the answer for your school is the one in your handbook, not the one in this article.

Trusting the summary over the lecture. AI summaries average. If the lecturer spent fifteen minutes on one corner case because it appears in every past paper, a summary will compress it to a clause or drop it. The number of times a topic is mentioned is information, and summarising destroys it. This is also why Step 3 exists in the form it does: your handwritten page is the record of emphasis, and the AI output is only the record of content.

Producing beautiful notes and never testing myself. This was the expensive one. I had colour-coded, linked, well-structured summaries for every module and I could not answer a single exam question in a timed setting. Generating notes feels like studying because it is effortful and it looks like progress. It is not progress. If your notes never turn into questions you answer from memory, the workflow above is just a more efficient way to avoid revising.

What this does not replace

AI can transcribe a lecture, structure it, and turn it into practice material. It cannot tell you whether you actually understand any of it - only answering questions from memory can do that, and that part is still yours.

The honest accounting: this workflow saved me roughly two hours a week of tidying notes, and the twenty minutes after class is the part that improved my grades. The saving is real, but it is a time saving, not a learning shortcut. Anyone selling it as the second thing is selling something else.

If you want to see how we decide which of these tools are worth recommending in the first place, our review standard is written out here. And if your lectures are maths or science heavy, the free solvers I tested pair naturally with this workflow for checking your own working.