How to Turn a Video Into Notes
A transcript is not notes. Here is how to get from one to the other without reading an hour of text, and why the machine cannot do the last step for you.
The Scriptorium team ·
Transcribing a video gives you every word that was said. Notes are the opposite: the few things worth keeping, in your own words. Getting from one to the other is an editing job, and it is the part that actually produces understanding.
Why the transcript is not the notes
A one-hour talk is roughly nine thousand words. Reading that takes about as long as watching the talk did, which means a raw transcript has saved you nothing except the ability to search it. That last part is worth a lot, but it is not notes.
Notes are shorter than the source, organised by idea rather than by chronology, and written in your own words. The last property is the one that matters: rewriting something in your own words is what tests whether you understood it.
The workflow
- Transcribe the video. video to text takes the file directly.
- Skim the headings and timestamps to get the shape of it. You are looking for where the topic changes, not for content yet.
- Mark the parts that matter. Usually a small fraction of the whole.
- Write the notes from the marked sections, in your own words, keeping the timestamp beside anything you might need to check.
- Keep the transcript. It is your index back into the video.
What to actually write down
The useful test for any sentence in a transcript is whether you would want it again in six months. That usually leaves:
- Definitions and the distinctions between similar things.
- Claims with reasons attached, and the reasons.
- Anything that contradicted what you expected.
- Numbers, with what they measure and where they came from.
- Things to look up.
- Anything the speaker repeated, slowed down for, or said twice differently.
What to leave: recaps, examples you already understood, housekeeping, and anything you are copying because it sounded good rather than because it was useful.
Using timestamps properly
Keep the timestamp next to any note you might need to verify. Written explanations lose tone, emphasis and the pause before an important qualifier, and sometimes the only way to recover what someone meant is to hear them say it again.
It also makes your notes checkable by someone else, which matters if you are taking notes on behalf of a team.
On asking a model to summarise it
You can paste a transcript into a language model and get a summary. It will be fluent, roughly correct, and will silently drop the qualifications — which is exactly where the meaning often lives. It also cannot know which three sentences matter to you, because that depends on what you already know.
Used as a first pass to find the sections worth reading, it is genuinely helpful. Used as the notes themselves, you end up with a document you have never actually read, about a video you never watched.
Building an archive
The compounding benefit arrives after you have done this a few times. Individually a transcript is a convenience; twenty of them, searchable, is a reference library of everything you have watched — and searching across all of them is a genuinely different capability from searching one.