# How to Take Notes From Video Courses You'll Actually Use

> A before, during, and after workflow for video course notes: timestamped lines, pause-and-predict, a five-minute recall, and dated review sessions.

- Canonical page: https://learnhubplay.nl/blog/notes-from-video-courses
- Published: 2026-11-04
- By: LearnHubPlay, an independent software studio in Amsterdam
- Topic: Notes & Research
- Tags: Note-taking, Learning, Retrieval practice, Spaced review, Templates

Say you finish a six-hour course and your notes are a thinner copy of the transcript. Three weeks later a colleague asks how the thing works, and you can recite the chapter titles but not the mechanism. Video makes this easy: the speaker is fluent, the slides advance without effort, and a smooth explanation feels like understanding.

The workflow below fixes that. Before a module you write the questions it should answer. During it you capture timestamped lines and make short predictions. After it you recall from memory for five minutes and put review dates in your calendar. Two debates get a straight answer on the way: playback speed, and typing versus handwriting.

## Why a smooth video feels like learning, and what to do about it

Roediger and Karpicke (2006) had 120 undergraduates study a short passage (about 260 words), then either read it again or take a free-recall test with no feedback. Five minutes later the rereaders recalled more, 81% against 75%. After two days the tested group led 68% to 54%, and after a week 56% to 42%. In a second experiment, students who studied a passage in four five-minute periods expected to remember it better than students who studied once and took three recall tests (4.8 against 4.0 on a seven-point scale), yet recalled 40% a week later against 61%. Rewatching is the video version of restudying; by analogy, expect the same extra confidence without the same retention. Dunlosky and colleagues' 2013 review of ten techniques likewise rated practice testing and distributed practice high utility, and rereading and highlighting low.

Video adds an attention problem. In Szpunar, Khan, and Schacter's (2013) second experiment, 48 undergraduates watched a 21-minute statistics lecture in four segments of about 5.5 minutes:

| Between segments | Mind wandering | Slides with added notes | Last-segment accuracy |
|---|---|---|---|
| Quizzed after each | 19% | 24% | 89% |
| Restudied after each | 39% | 9% | 65% |
| Tested only at the end | 41% | 7% | 70% |

With a small sample and one lecture, treat it as a direction: put a retrieval attempt inside the video, not only after it.

## Before you press play: one question per module and a note page that is ready

1. **Write one to three questions** the module should let you answer, after a glance at the lesson titles. Carpenter and Toftness (2017) had people answer prequestions before each of three segments of a short video; later they beat a control group on both prequestioned and other material. A 2018 follow-up with lectures over 20 minutes found a clear benefit for the prequestioned items but at most a marginal one for the rest, so aim your questions at what you most need to keep. Both gave participants ready-made questions, so self-written ones are an adaptation.
2. **Name the use.** One line on the task this should help with; it filters what deserves a note.
3. **Open a module page** from this template:

```text
Module:   <course> / <module title>              Started: <date>
Use:      <the task or decision this should help with>
Link:     <lesson URL>
Questions
  1. <what this module should let me answer>
  2.
Notes     (format: [mm:ss] tag, then your own words)
```

## During the video: timestamped lines and the pause-and-predict routine

### Write each line as a time, a tag, and your own words

Five tags cover most captures: `=` a claim in your words, `?` something you could not follow, `G` a gotcha, `>>` something to try, and `P` a prediction. A hypothetical entry from a Git course:

```text
[04:20] =   Rebase replays my commits onto a new base, so each gets a new hash.
[06:05] P   After the rebase my old commits are gone.  -> wrong: see reflog
[09:40] G   Don't rebase commits teammates have already pulled.
[11:10] ?   Why does --force-with-lease exist if --force works?
[12:30] >>  Redo the 06:05 demo in a scratch repo; find the old hash.
```

The time makes a note cheap to check: when a line stops making sense in a fortnight, you rewatch 90 seconds, not the lesson. Use your own words; the original laptop study blamed verbatim copying, though that penalty did not hold up in its replication. Cap each line at about ten seconds. If you screenshot a slide, add a one-line caption in your own words and the time; a bare screenshot is a copy (our judgment).

On YouTube, click the player first; then K pauses, J and L jump 10 seconds, and < and > change speed. To link a line, click Share, check Start at, and enter the time. If your course platform's notes don't show the time, type the `[mm:ss]` yourself.

### Pause-and-predict: stop before the reveal

Pause at three kinds of moment and write one `P` line before pressing play:

1. The instructor poses a problem or asks what will happen.
2. Just before code runs or a result appears.
3. The end of each section, or about every five minutes if there are no breaks.

One sentence is enough: `P  I expect X because Y`. Press play, mark it right or wrong, and add a `G` line for what you missed. Review wrong predictions first; they show where your model differs from the instructor's. Time your first three pauses; if each takes 30 seconds, six pauses add three minutes to a module.

> **Note:** Szpunar's students were quizzed after each segment, and Carpenter and Toftness supplied the questions. Neither tested a self-made one-line prediction, so this routine adapts their findings; it is not a tested method.

### Adjust the note to the kind of video

This table is our synthesis, not a tested classification; the last column feeds the question bank below.

| Kind of video | What goes in the note | Where the detail lives | Bank question type |
|---|---|---|---|
| Coding tutorial (follow along) | Decision, gotcha, timestamp; one line each | The code, with `[mm:ss]` in a comment or commit message | Rebuild the function from the spec |
| Conceptual lecture | Claims in your words, `?` lines, one hand sketch | The slides or handout | Why does B follow from A; when does it fail |
| Software walkthrough | Goal, menu path, and version as numbered steps | Your own recording or a clean test project | Do the task on a fresh project |

Say you are watching a statistics lecture (hypothetical):

```text
[07:15] P   Quadrupling the sample size will halve the width of the confidence interval.  -> right: width scales with 1/sqrt(n)
```

## Playback speed and typing versus handwriting: what the studies found

### Speed: small costs up to 2x, a clearer drop at 2.5x, in one lab study

Murphy and colleagues (2022) had 231 UCLA undergraduates watch two lectures, on real estate appraisals (about 13 minutes) and the Roman Empire (about 14.5), at one of four speeds, then tested them immediately and a week later. They were told not to pause or take notes. Averages over both lectures:

| Speed | Immediate test | One-week test |
|---|---|---|
| 1x | 65% | 59% |
| 1.5x | 60% | 54% |
| 2x | 62% | 53% |
| 2.5x | 55% | 50% |

Speech ran from about 160 words a minute at 1x to about 400 at 2.5x. The experiments were preregistered; 123 students who never saw the videos (a separate sample, not preregistered) scored 41%. Only 1x against 2.5x differed reliably, hence the authors' "minimal costs" up to 2x, but the raw averages at 1.5x and 2x sit five or six points below 1x at one week. "Minimal" is not "none".

In two follow-ups, watching twice at 2x back to back matched watching once at 1x (63% against 64%). Watching at 2x, then again at 2x right before a test a week later, beat watching once at 1x (63% against 55%). The authors flag a confound: the 1x group had no refresher, so the study cannot separate watching twice from watching recently.

The limits: short lectures, undergraduates, multiple-choice and true/false questions, no pausing, no notes. The authors caution that harder material, other speech rates, or less overlap between audio and visuals may differ. Our reading: for straightforward lectures like these, a faster speed is a fair trade if you spend the minutes saved on a second look shortly before you need it. The study tested rewatching, not recall, so adding recall is our extension.

### Typing versus handwriting: contested, small, and about verbatim copying

| Study | Design | Finding |
|---|---|---|
| Mueller and Oppenheimer, 2014 | Three studies, laptop versus longhand | Laptop note takers did worse on conceptual questions and tended to transcribe verbatim |
| Morehead, Dunlosky, and Rawson, 2019 | Two experiments (eWriter, no-notes groups) plus a mini meta-analysis | No consistent difference; small, nonsignificant combined effect favoring longhand; verdict called premature |
| Urry et al., 2021 | Direct replication of Study 1; 74 laptop, 68 longhand; no study time before the quiz | More words and verbatim overlap on laptops, but no quiz difference; eight similar studies agreed |
| Flanigan et al., 2024 | Meta-analysis, 24 studies, college students | Taking and reviewing handwritten notes: higher achievement, g = 0.248; typing produced more notes, g = 0.919 |

So the original mechanism, better encoding through longhand, was not supported in Urry and colleagues' replication, while a later meta-analysis of taking and reviewing notes finds a small handwriting edge in achievement. In the original and the replication, notes with more words went with better quiz scores and more verbatim overlap with worse ones, though the overlap link did not hold up in the replication; neither was manipulated, so these are correlations.

All four studies concern college students at lectures; none that we read tested a multi-hour video course or a coding tutorial, where you type code regardless. Our rule: use the medium you can keep up with, make every line a rephrase, and if you catch yourself transcribing, pause and restate.

## After each module: a five-minute recall, then a question bank

Do this before the next module; set a timer.

1. **Close the video and your notes.**
2. **Minutes 0 to 2, recall on a blank page.** Answer your pre-module questions first, then write everything else you can.
3. **Minutes 2 to 4, compare.** Open your notes, mark every gap with its timestamp, and rewatch the 90 seconds behind the biggest one.
4. **Minute 4 to 5, bank it.** Turn each gap, `G` line, and `?` line into a question with an answer and a timestamp:

```text
Q: Why does a rebase change commit hashes?
A: Each commit is replayed onto a new parent, so it is new.   [M3 04:20]
Q: Predict: after a rebase, are the old commits gone?
A: Not at once; git reflog still lists them.                  [M3 06:05]
```

Write questions that demand an answer, not cues such as "rebase". Four forms cover most modules: when does X fail, what does this return, why does Y come before Z, and do this from the spec without looking. The Cornell method's recite step is the same idea on paper; our [comparison of note methods](https://learnhubplay.nl/blog/note-taking-methods-compared) covers it.

## Spaced review: dates that fit when you need the skill

Cepeda and colleagues' 2006 meta-analysis pooled 839 assessments of distributed practice; the best gap between sessions grew with how long the material had to be remembered. Their 2008 study taught more than 1,350 people a set of facts, reviewed them after gaps of up to 3.5 months, and tested up to a year later: the best gap was about 20 to 40% of a one-week test delay but only 5 to 10% of a one-year delay.

Those were facts with one review, so no study gives you a schedule for a video course. Ours is a reading of the pattern: ask when you will need the material, review first after two days, make each gap about two and a half to three times the last, then shorten the final gap so the last review lands two or three days before the need date.

Say you finish module 3 on a Monday and will use it in a project that starts on a Monday six weeks (42 days) later:

| Session | Day | Weekday | Gap | Five-minute task |
|---|---|---|---|---|
| Module done | 0 | Monday | none | Recall and question bank |
| Review 1 | 2 | Wednesday | 2 days | Answer the bank cold; re-queue misses |
| Review 2 | 7 | Monday | 5 days | Same |
| Review 3 | 21 | Monday | 14 days | Same |
| Final | 39 | Friday | 18 days | Answer cold, rewatch missed segments at 2x, redo the `>>` items by hand |

The last gap (18 days) is shorter than the rule gives because the final review must land three days before day 42. Rewatching missed segments at 2x is our adaptation of Murphy's result, not a tested method.

Keep the bank in one sheet: Question, Answer, Source (`M3 04:20`), Due, Streak. Due comes from the module's finish date (plus 2, 7, and 21 days, then the need date minus 3). A right answer adds one to Streak and moves Due on; a wrong one resets Streak to 0 and sets Due to your next session. With twelve modules, block one daily slot and answer every row due that day or earlier, rather than 48 calendar events. Our rule, not a tested one: after two right answers in a row, skip a question in the next review, but never in the final one.

## Where to start this week

Run the full routine only on modules you will apply within weeks (about 25 minutes of follow-up each at five minutes a session); for background courses, do the pre-module question and the recall (about 5). That split is our judgment, not a tested rule.

- **Today:** pick an unwatched module, write two questions, open the template.
- **While watching:** make three pauses and write a `P` line at each.
- **Right after:** do the five-minute recall and calendar the review dates.

The goal is not tidy notes. It is being able to answer your own questions on the day you need the skill.

*[TabNotes](https://tabnotes.learnhubplay.nl/), our Chrome extension, can attach a Markdown note to a lesson's exact URL so it reappears when you return to that page, and it has reminders. A text file and a calendar do everything above just as well.*

## Frequently asked questions

### Can a transcript or an AI summary replace my notes?

Use them as an answer key, not a substitute: a summary does the restating for you, and in Roediger and Karpicke's study restudying raised confidence while testing produced better week-later retention. [Learning with AI without fooling yourself](https://learnhubplay.nl/blog/learning-with-ai-without-fooling-yourself) covers prompts that quiz you instead.

### What if the course platform has no notes feature?

Keep the player in one window and a text file in the other, and type the `[mm:ss]` yourself. The [tab note-taking workflow](https://learnhubplay.nl/blog/tab-note-taking-workflow) covers attaching notes to a URL so they greet you when you return.

### Should I write while the video plays or only at pauses?

We did not find a study comparing the two for self-paced video courses, though work on instructor-imposed pauses is related; Murphy's participants were barred from both. Speech keeps coming while you write, so a full sentence risks missing the next one (our reasoning). Jot the time and two keywords while it plays, then finish the line at the next pause.

### Is rewatching at 2x a substitute for the five-minute recall?

Not on this evidence. Murphy and colleagues compared rewatching with a single earlier viewing, so the study shows extra exposure helps, not that it beats retrieval; no viewing schedule included a recall attempt between viewings. In Roediger and Karpicke's study the restudy group got more exposure and still recalled less after two days and a week. Answer from memory first, then rewatch only the segments you missed.

## Sources and further reading

- [Test-enhanced learning](https://learninglab.psych.purdue.edu/downloads/2006/2006_Roediger_Karpicke_PsychSci.pdf) — Roediger and Karpicke, Psychological Science, 2006 (author-hosted copy), and [Improving students' learning with effective learning techniques](https://doi.org/10.1177/1529100612453266), Dunlosky et al., Psychological Science in the Public Interest, 2013. Recall figures and confidence ratings; utility ratings.
- [Interpolated memory tests reduce mind wandering and improve learning of online lectures](https://pmc.ncbi.nlm.nih.gov/articles/PMC3631699/) — Szpunar, Khan, and Schacter, PNAS, 2013. The mind-wandering, note, and accuracy figures.
- [The effect of prequestions on learning from video presentations](https://doi.org/10.1016/j.jarmac.2016.07.014) — Carpenter and Toftness, Journal of Applied Research in Memory and Cognition, 2017, and [The limited effects of prequestions on learning from authentic lecture videos](https://doi.org/10.1016/j.jarmac.2018.06.003), Toftness, Carpenter, Lauber, and Mickes, 2018. The short-video result and the lecture-length follow-up.
- [Learning in double time: The effect of lecture video speed on immediate and delayed comprehension](https://doi.org/10.1002/acp.3899) — Murphy, Hoover, Agadzhanyan, Kuehn, and Castel, Applied Cognitive Psychology, 2022. The speed table and rewatching experiments.
- [Spacing effects in learning](https://doi.org/10.1111/j.1467-9280.2008.02209.x) — Cepeda, Vul, Rohrer, Wixted, and Pashler, Psychological Science, 2008, and [Distributed practice in verbal recall tasks](https://doi.org/10.1037/0033-2909.132.3.354), Cepeda et al., Psychological Bulletin, 2006. Gap ratios; the 839-assessment meta-analysis.
- [The pen is mightier than the keyboard](https://doi.org/10.1177/0956797614524581) — Mueller and Oppenheimer, Psychological Science, 2014. The original three studies.
- [How much mightier is the pen than the keyboard for note-taking?](https://link.springer.com/article/10.1007/s10648-019-09468-2) — Morehead, Dunlosky, and Rawson, Educational Psychology Review, 2019, and [Don't ditch the laptop just yet](https://doi.org/10.1177/0956797620965541), Urry et al., Psychological Science, 2021. The two direct replications.
- [Typed versus handwritten lecture notes and college student achievement: a meta-analysis](https://doi.org/10.1007/s10648-024-09914-w) — Flanigan, Wheeler, Colliot, Lu, and Kiewra, Educational Psychology Review, 2024. Both effect sizes. YouTube steps: [Share videos and channels](https://support.google.com/youtube/answer/57741) and [Keyboard shortcuts for YouTube](https://support.google.com/youtube/answer/7631406), YouTube Help.
