How it works

How the YouTube Summarizer Works

Paste a link, and a minute or so later you're reading the summary, free and without an account. Here is exactly what happens in between, and what the tool can't do.

Try it now See the features

How does a YouTube summarizer work?

It reads the video's captions, not the pictures. The summarizer fetches the transcript with a timestamp on every line, and an AI model turns it into a TL;DR, key takeaways, chapters and notes. Each chapter keeps the second it starts, so you can jump straight to that moment in the video.

Five stages from link to summary: your link, a video check for title, channel and thumbnail, captions with a timestamp on every line, the AI summary with TL;DR, takeaways, chapters and notes, and a final check; no captions means you are told, and results are kept for 30 days.
The five stages behind the Summarize button, and the two places the process can stop early or skip ahead.
  1. 1

    Step 1: Paste the YouTube link

    The link is read the moment you paste it. Watch links, youtu.be short links, Shorts, embeds, live replays and bare video IDs all resolve to the same eleven-character ID.

    That ID is checked against YouTube first, and the video's real title, channel and thumbnail are shown, so you know it's the right video before anything else runs. If YouTube says the video is private or removed, you're told there and then.

  2. 2

    Step 2: The captions are fetched

    The summary is built from the video's captions, the same text YouTube shows in its transcript panel. Human-written captions are used where they exist, otherwise YouTube's auto-generated ones. Every line keeps its start time.

    The timings matter as much as the words. They're what lets every chapter point back to the second it begins, and they're carried through the whole process rather than thrown away after this step.

    If a video has no captions at all, you're told plainly. You will never get a confident summary of a video nobody could read. There are other ways to handle a video without a transcript.

  3. 3

    Step 3: The AI reads it, rather than skimming

    The transcript goes to a large language model with the timings intact and a deliberately narrow brief: find the actual argument, keep what's specific, drop the filler, keep the speaker's hedges, and carry names, numbers and dates across exactly as spoken. The summary is written in the video's own language.

    It comes back as structured data rather than loose prose: a TL;DR, a list of takeaways, chapters with start times, and headed notes. That's why switching between the views is instant; they were all made in the same pass.

    Before you see anything, the result is checked. Every part must be present, chapters are put in order, and no start time may fall past the end of the video. An incomplete answer is rejected, so you're asked to try again rather than shown half a summary.

  4. 4

    Step 4: Copy, download or share the summary

    Copy the summary as Markdown or download it as a .md file. The Markdown has real heading levels and working timestamp links, so it survives a paste into Notion, Obsidian or a document; see exporting a summary to Markdown. To share one, send youtubesummarizer.com/?v= followed by the video's ID: it opens the same summary for whoever clicks it.

    Your last six videos stay a click away under Recent. Without an account that list lives in your browser's local storage and nowhere else, and one click clears it.

How the timestamps survive

The AI never sees a wall of text. It reads the transcript in short lines, each tagged with the second it starts, and it's asked to use those tags when it marks where a chapter begins. The chapter times you click on are copied from those tags, not estimated by the model.

Transcript lines tagged with start times, such as [125s|2:05] an arch turns the weight above, become chapters with the same times: 0:00, 2:05 and 3:09, each opening the video at that second.
The line that starts at 125 seconds becomes the chapter that opens the video at 2:05.

That's what makes a summary checkable: every chapter is one click from the words it came from. More on why that matters in timestamped YouTube summaries.

What happens with very long videos

Most videos, including talks and podcasts of a few hours, are read in one pass. Only the very longest transcripts are split into parts at whole lines. Each part is summarized on its own, then one more pass merges the part summaries into a single summary of the whole video.

A very long transcript is split into three parts, each part is summarized, and the three part summaries are merged into one summary of the whole video.
Splitting keeps every part within what the model reads well, and the chapter times still refer to the full video.

Each part keeps its original timestamps, so a chapter three hours in still links to three hours in. The guide to summarizing long YouTube videos has tips for getting the most from multi-hour streams and lectures.

Why the same video comes back instantly

A finished summary is kept on our server for 30 days, filed under the video, not under you. If you or anyone else asks for the same video again, it comes straight back without re-reading the transcript, and it doesn't count towards your five free summaries a day. Failed attempts don't count either, and the allowance resets at midnight UTC.

What leaves your browser

The link you paste goes to our server. From there the video's public details come from YouTube, the captions from a transcript service, and the caption text goes to an AI model provider to write the summary. No account is needed. The daily allowance is tracked by a scrambled one-way code made from your connection, not your name or address. The guide to summarizer privacy covers what to consider before pasting internal or unlisted videos, and the privacy policy has the formal detail.

Honest limits

What a YouTube summarizer can't do, and other questions

Worth knowing before you rely on it.

Can it summarize a video with no speech?

No. Music videos, silent timelapses and footage where the meaning is entirely visual have nothing for a caption-based tool to work with. You'll be told, rather than given a summary invented from the title.

Can it read private or age-restricted videos?

Private videos, no: we can only reach what a public link can reach. Unlisted videos usually work if you have the link. The guide to private and age-restricted videos covers the edge cases.

Can it tell whether the video is right?

No. A summary reflects what was said, not whether it's true. If a video is confidently wrong, the summary will be a faithful account of a confidently wrong argument. The timestamps are there so you can judge for yourself.

How accurate are summaries made from auto-captions?

They're as good as the captions. Where YouTube's automatic captions mishear a name or a technical term, that error flows through. Heavy accents, poor audio and specialist vocabulary are where it shows most; see how accurate YouTube's auto-captions are. If a summary looks off, here's how to tell why.

How long does a summary take?

Usually a minute or so; the page shows which stage it's on while you wait. Very long videos take longer, and a video that's already been summarized in the last 30 days comes back instantly.

Does it watch the video?

No. It reads what was said, via the captions. Anything shown on screen but never spoken, such as a chart, a code listing or a demonstration, isn't in the summary.

That's the whole process.

It takes longer to describe than to run. Try it on something from your watch-later.

Summarize a video — free