YouTube transcript downloader

Use the YouTube transcript downloader without handing over the video

Every other tool for this job asks for a video link and fetches the transcript on its own servers. This one does not, and the difference is not decoration. YouTube's terms allow automated access to the service only for public search engines that follow its robots file, or with written permission. The official Data API can download captions only for videos the authenticated account owns, so there is no sanctioned route to an arbitrary video's captions at all. Services that accept a pasted URL reach an undocumented internal endpoint to get around that. This page reads what YouTube has already shown you, in your own browser, in your own session.

That leaves two honest ways to get the text here, and both are offered as equals. You can open the transcript on YouTube, select the lines, copy them and paste them in. Or you can drag the collector to your bookmarks bar once and click it on any watch page afterwards, which opens the transcript panel, reads every line, and passes the result to this tab without a single request leaving your browser.

What you get out, and how exact it is

Five formats: plain text merged into readable paragraphs, plain text with timestamps, SubRip, WebVTT and CSV. Automatic captions arrive as a thousand fragments of four or five words, broken mid-sentence, which is unreadable in a document. The paragraph export joins them and breaks only where a sentence ends. The CSV keeps one row per line with start and end in milliseconds, which is what you want when the transcript is going into a spreadsheet or a script rather than a document.

Precision differs between the two intakes, and the page says which one you have rather than leaving you to guess. The collector reads each line's start offset directly from the page in milliseconds, so subtitle files are exact. A pasted transcript carries only the clock the panel displays, which is whole seconds, so its timings are true to the second and no finer. Some transcript panels publish a real end time for every line and some publish only starts; where a real end exists it is used unchanged, and where none does, a line ends where the next begins. The final line falls back to the video length when the collector knows it. That last value is an assumption, and it is written down rather than hidden.

Three different failures deserve three different answers

A video with no captions is not a broken tool. When YouTube has no transcript for a video the collector says exactly that, because the alternative teaches you to distrust the tool for something that is a property of the video. When the transcript panel exists but cannot be read, the message is different: YouTube changed its page, and the collector sitting in your bookmarks bar is a frozen copy that cannot repair itself. It links back here so you can drag a fresh one.

The third failure is the one worth building for. If only part of the transcript can be read, the result still looks complete — a file arrives, it contains sentences, nothing appears wrong. A transcript that stops twelve minutes into an hour is not slightly short; it is the wrong answer delivered with the confidence of a right one. So the collector compares the last line it read against the video's own length and refuses to send anything below nine tenths coverage, and this page repeats the figure next to the result.

What this page will not do

There is no box for a video URL, no channel or playlist mode, and no server. It does not translate, summarize or run the text through a model. It cannot reach private, age-gated or unlisted videos, because it never reaches anything — it only reads what your browser was already given. The bookmarklet is a desktop mechanism; on a phone, copy the transcript from YouTube's own panel and paste it here, which works the same way and produces the same files.

A downloaded transcript is not a license to republish it. Spoken content in a video is subject to the same rights as the video itself. Use it for notes, search, accessibility, quoting within the limits that apply where you are, or for material you made yourself. If you want to count what was actually said, the word frequency counter takes the exported text directly, and the character counter reports length for platforms with limits. For the video's artwork rather than its words, use the YouTube thumbnail downloader.