Summarize YouTube videos without leaving the page

The summary follows a fixed structure so you can scan it:

• TL;DR: two or three sentences on what the video is about and its main takeaway.
• Key points: five to eight bullets in the order they come up, each starting with the timestamp where it is discussed, so you can jump to that moment.
• Notable quotes: two to four short verbatim quotes with timestamps. If nothing is worth quoting, the section is left out.

The instruction to the model is to use only what the transcript says, and to quietly correct obvious caption slips, such as a misheard name. It is a summary of the words spoken. It does not see the picture, so a tutorial that relies on what is shown on screen will be summarised from what the speaker says about it.

Copilotly reads the caption track from the YouTube page you are on, using your own browser session. That includes auto-generated captions, and the sidebar notes when the transcript is auto-generated, because those can mishear names and technical terms.

It reads up to 45,000 characters of transcript. A very long video, such as a multi-hour talk, is trimmed to fit, and the chip says trimmed so you know the summary covers only the part that was read. For long videos, ask about specific sections instead of the whole.

If a video has no captions at all, the chip says No transcript - the link will be sent instead. The assistant is then told to say plainly that no transcript was available and to give only what the title and description support, not to invent what the video says. Videos with captions turned off by the uploader, and some live streams, fall in this group.

On an ordinary page the Summarize action reads the page text instead, and the composer shows a chip with the site and how many characters were read. That makes it useful for long articles and documentation. The page action menu also has Reply, Email, Explain code and Write post, depending on the site.

If you would rather summarize text you pasted, use the free text summarizer. If your source is a document rather than a video, chat with a PDF, Word, Excel or CSV file works the same way.

A video summary is a normal chat message. There is no separate daily video limit; what counts is the model you pick. Fast models are not limited, premium models such as Claude, Gemini or GPT use your 10 premium messages a day on Free, and Pro removes that cap. With your own API key nothing is metered.

Dedicated video summarizers can have extras we do not: summary panels built into the YouTube page, export to notes apps and playlists in one go. We are an assistant that happens to summarize videos, so the summary lives in a chat, which is better for follow-up questions and worse for one-click bulk work. We have not compared output quality with specific tools and will not claim one.

Summaries are most useful when the video is long and you have a specific need: deciding whether to watch, finding the part that matters, or revising what you saw. They are a way to choose what to watch, not a replacement for the parts that matter.

What a transcript cannot tell you

A transcript tells you what was said. It does not tell you whether the speaker is right, whether a demo worked, or what was shown on screen. A confident summary of a flawed argument is still a summary of a flawed argument.

Verify with the timestamps

Every key point starts with a timestamp. Click through to that moment before you repeat a number or a claim. For anything you will act on, such as health, money or legal advice from a video, check it against a reliable source with web search.

Auto-captions and accuracy

Automatic captions are generated by YouTube and can mishear names and terms; YouTube describes how they work and their limits on its automatic captions help page. The sidebar tells you when the transcript is auto-generated, and the summary is instructed to fix obvious slips quietly, which means a slip it cannot spot will remain.

The summary is one message. The transcript stays in the chat, so the real value is in the follow-ups.

Ask for structure

• List the steps as a numbered list, with timestamps.
• What tools, settings or prices does the speaker mention?
• What are the speaker's three main claims, and what evidence do they give for each?

Ask for a specific stretch

For long videos that hit the cap, ask about named sections: What do they say about pricing? beats summarize everything. If the chip says trimmed, the later part of the video is not in the chat, so ask for the early part and watch the end yourself.

Ask for a different output

Turn the video into revision notes, a checklist, an email to a teammate or a tweet thread. If you want the result as a file, see how chat creates spreadsheets and documents. For text you already have, the text summarizer does the same job on a pasted transcript.

You want...	Best option	Why
To decide whether a video is worth watching	Summary	Gives the structure in seconds
To find where a topic is discussed	Follow-up question	Answers come with timestamps
To learn a skill shown on screen	Watch, with the summary as a map	The summary only has the words
Exact wording of a quote	Open the timestamp and listen	Auto-captions can be wrong
A video without captions	Another source or captions from the uploader	Nothing to read

Uploaders can turn captions on or off; YouTube's guide on adding subtitles and captions explains how, which is useful if you are the one making the video. If you are reading in another language, combine this with page translation for the description and comments.

What the extension reads

On a watch page it reads the caption track and basic details: the video title, channel, length and the first part of the description. It does this using your own browser session on the page, without playing the video.

What is sent to the model

The transcript, the title and the description are included in the message to the model that writes the summary, together with your question. If there is no transcript, the link is sent instead so the model can try to read the video; the chat is told to say plainly when it has nothing and not to invent content. Do not use it on private or unlisted videos you are not allowed to share with an AI provider.

Choosing a model for long transcripts

A long transcript is a lot of text. If one model struggles with detail, switch the picker to another in the same chat and ask again. The Gemini and Claude pages describe what each is like for long inputs.

Before you watch

Summarize first and read the TL;DR and key points. Decide whether it deserves your time, and note the timestamp of the one part you most need.

While you watch

Jump straight to that timestamp. Keep the chat open and ask about terms you do not know. Ask for the steps as a list if it is a how-to.

After you watch

Ask for notes in your own format, such as flashcards or a checklist, and check them against what you saw. For deeper research on the topic, run a web search or a deep research report and compare with what the video claimed.