How to summarise a YouTube video with ChatGPT in 4 steps
Working with documentsLast checked
Short answer
Use the transcript rather than the video. YouTube provides one for most videos: open the description and click Show transcript, then copy it. Paste that into ChatGPT and you can summarise it, pull the main claims out, or find the part you actually needed, with timestamps.
Getting the transcript
Open the video description
Click the description area below the video to expand it.
Find Show transcript
Usually near the bottom of the expanded description. Most videos with captions have one.
Copy it
The panel opens beside the video. Select the text and copy. There is a toggle to hide timestamps if you prefer, though keeping them is usually more useful.
Paste it into ChatGPT with your question
No upload needed, which also means it does not touch your upload allowance at all.
Keep the timestamps
They cost you nothing and they earn their place. Ask for timestamps in the answer and you can jump straight to the two minutes you actually needed, instead of scrubbing through forty.
Asking for something better than a summary
A summary of a video is usually as unsatisfying as a summary of a meeting. More useful requests:
For a tutorial:
From this transcript, give me the steps in order, as a numbered list, with the timestamp each one starts at. Leave out the introduction and the sponsorship section.
For a talk or interview:
What is the main claim, what evidence is given for it, and what is asserted without support? Quote the line each point came from.
For deciding whether to watch it at all:
In five bullets, what does this video actually cover? Then tell me whether it addresses [your specific question], and at what timestamp.
That last one is the highest value use of this by a distance. Three minutes to find out a forty minute video does not answer your question is a good trade.
What a transcript misses
Speech only. Anything demonstrated on screen and not narrated is absent.
That matters most for tutorials, where "and then you click here" is the whole instruction and the transcript records it as exactly that. For a talking-head video, an interview or a lecture, a transcript loses very little.
If the visuals carry the content, screenshot the important moments and upload them as images alongside the transcript. Images are capped at 20 MB each with a ceiling of about 100 per conversation.
Long videos
Auto-generated transcripts are wordy, since speech is repetitive. A ninety minute video can produce a transcript long enough to matter.
If it approaches 2 million tokens, it gets truncated silently. Check by asking what the final thing discussed was. If it cannot say, it did not reach the end, which for a video with a conclusion is the part you wanted.
Several videos at once
The same two-pass approach that works everywhere else. Summarise each transcript separately into the same structure, then compare:
Here are structured notes from six videos on the same topic. Where do they agree, where do they conflict, and which one actually explains the mechanism rather than describing the outcome?
Genuinely useful for working through a playlist, and it keeps attribution intact, which uploading six transcripts at once does not.
Common questions
Can I just give ChatGPT the YouTube link?
Not reliably. Whether it can open a link depends on browsing being available in your session, and even then the result is inconsistent. Pasting the transcript is the approach that works every time.
Where do I get the transcript?
YouTube itself. Under the video, open the description and look for "Show transcript". Most videos with captions have one, and you can copy it straight out.
The transcript is a mess of timestamps. Does that matter?
Not much. ChatGPT handles them fine, and they are useful, because you can ask for timestamps in the output and jump straight to the part you care about.
Will it understand what was shown on screen?
No. A transcript is speech only. For a tutorial where the value is in what is demonstrated rather than said, screenshot the key moments and upload those as images alongside.
Keep reading
Can you upload an MP4 or other video to ChatGPT?
Video is not among the 4 file families OpenAI lists for uploads. Transcribe it and upload the .txt instead, plus what to do when the visuals are the point.
How to get something useful out of a meeting transcript
Do not ask for a summary. Ask for 4 things: decisions, commitments with owners, open questions, disagreements. Keep the speaker labels or you lose every name.
How to summarise long documents with ChatGPT
Anything past 2 million tokens is cut off silently, so run the last section test before you read the summary. Then the prompts that stop it coming back bland.