How to Extract Captions From a YouTube Video Online

Captions are not burned into a YouTube video. They are separate text assets attached to it — one per language, one per kind — and the player just draws whichever one you switch on. That single fact is what makes extraction possible at all, and it explains why some videos give you nine tracks and others give you none. This is the practical version of extracting captions in a browser: how to see the full track inventory, what each kind is good for, and what to do with the file afterwards.

What “extract” actually means

Every caption you can see on YouTube lives in a subtitle track: an ordered list of cues, each with text, a start time and an end time, stored per language. Google’s own captions resource describes them exactly that way — separate objects with their own IDs, languages and kinds. Extracting means reading those cues out of the track and writing them into a file you can keep: SRT, VTT, or plain text with the timing thrown away.

Two consequences worth internalising. First, extraction is cheap and fast — a caption track is text, so a three-hour lecture extracts in seconds and length barely affects the wait. Second, you can only extract what exists. A video with no track has nothing to extract, and no website, extension or command-line tool changes that. YouTube’s caption documentation is the reference for how tracks are created and exposed.

Extract vs download vs transcribe

Three phrases, two real activities:

• Extract and download are the same job seen from different ends — pull an existing track, convert it, save it. We use them interchangeably here, and the tool we use labels both its buttons for that reason.
• Transcribe is a different job entirely: running speech recognition on audio to create text that didn’t exist. If a video has no captions, transcription is the only path — and it’s slower, less accurate and usually metered or paid.
• Repurpose is what people mean five minutes later: the article, notes or clips you build from the text. Extraction is step one of that, not the deliverable.

Knowing which of the three you need stops you picking the wrong tool: an extractor when you want the creator’s exact wording, a transcription service when there are no captions, an editor when the text is fine and the timing isn’t.

The extraction workflow, start to finish

1. Copy the video URL — watch link, youtu.be link, Shorts link or YouTube Music link, one video at a time.
2. Paste it into an extractor — we use YouTube Caption Downloader because it lists every track rather than guessing which one you want, needs no signup, and installs nothing. Its homepage states the terms: manual, auto-generated and translated captions, output as SRT, VTT or plain text, free to use.
3. Read the inventory — language, kind, and (in most tools) a preview. Don’t just take the first row.
4. Choose the output format by destination: TXT to read, SRT for an editor, VTT for a web player. The video tools we rate cover both burning-in and editing.
5. Open the file and check three lines — a name, a number, and a sentence with punctuation. Two seconds, and it tells you immediately whether you got the manual track or the automatic one.

For the click-by-click version with troubleshooting, see our companion guide to downloading captions from a YouTube video for free.

Reading the track inventory

A typical extraction returns something like this, and the tiers matter more than the languages:

Track kind	Where it comes from	Use it for
Manual / uploaded	The creator or someone they hired	Anything published, quoted or used for accessibility
Auto-generated	YouTube’s speech recognition	Research, notes, searching a long video, first drafts
Translated	Machine translation of a source track	Understanding a video you don’t speak in
Community	Volunteer contributions (largely retired)	Rare, and often stale when present

The order of preference never changes: manual first, auto-generated second, translated third. The W3C’s captioning guidance is blunt about why — captions that misrepresent speech are worse than no captions for a deaf viewer, which is exactly the case for shipping an unedited auto track.

Where auto-generated tracks go wrong

Automatic tracks are good enough to be useful and bad enough to be dangerous. The failures we see over and over in extraction work:

• Proper nouns. Product names, people, places, and especially anything newer than the model’s training data. “Whisper” becomes “whispered”, a company becomes a common noun.
• Punctuation and sentence boundaries. Auto tracks frequently run clauses together, which is why a raw TXT export reads like a wall of text.
• Speaker changes. Nobody gets labelled, so a two-person interview extracts as a monologue.
• Overlap and music. Two voices at once, or a music bed under dialogue, cost you whole sentences.
• Duplicated rolling cues. Auto captioning re-sends the tail of a segment, so the same words appear twice in a row — the thing to strip before you paste anywhere.

All five are fixable in a text editor in a few minutes; the technique is in our subtitle text extractor guide.

Extracting languages the creator never published

Translation in YouTube’s system is derived: it runs from whichever source track exists, so a video captioned only in English can list Spanish, Portuguese, Hindi, Japanese and a long tail more. That makes extraction a surprisingly good cross-language research shortcut — a Japanese conference talk becomes readable English in one click — with one caveat. If the video has no source track, there is nothing to translate, so the translated rows vanish too. And machine translation inherits every error in the automatic track underneath it: a mistranslated word is often a misheard one.

Where a tool exposes 150+ language variants rather than the handful YouTube’s own player shows, the multi-platform extractor in our subtitle extractor guide covers that shape of the market.

What to do with the extracted file

• Video to article. TXT out, paste into a doc, cut the filler, add headings and a pull quote. Search engines can finally read the thing.
• Video to short clips. The SRT is a search index: find the sentence worth cutting, note its timecode, cut there. Our timestamps walkthrough covers importing the same SRT into an editor.
• Meeting and lecture notes. Searchable text beats rewatching; pair it with a summariser if the video is long, and see our best free AI tools roundup for that step.
• Own-channel cleanup. Extract your own auto track, fix the names, upload it as the manual captions — the cheapest accessibility win available.
• Competitive research. Extract a rival’s most-viewed video and count the phrases they repeat in the first five minutes.

Limits nobody mentions

Extraction is one-video-at-a-time in every browser tool we’ve used, including this one; playlists need yt-dlp or a script. Extractors don’t edit: mistimed or typo-ridden cues come out exactly as they went in. Private and age-restricted videos are unreachable without being the rightful viewer, which no anonymous tool can be. And nothing here changes the fact that the words belong to whoever said and published them — extracting a transcript to read is ordinary, republishing it as your own article is not.

Frequently asked questions

What is the difference between extracting and downloading YouTube captions?

No real difference in outcome. Extraction means pulling the caption track out of the video’s subtitle list and converting it to a file you keep; downloading is the same act described from the browser’s point of view. Both produce an SRT, VTT or TXT file rather than the video itself.

Can I extract captions from a video in another language?

Yes, whenever YouTube has a translated track for it. Translation is derived from an existing source track, so a video with English captions can list dozens of translated languages even though the creator never published them. Quality is machine-grade, so treat it as comprehension help rather than a publishable file.

Does extraction create captions if the video has none?

No. An extractor reads tracks that already exist. If the creator disabled captions and YouTube has not produced an automatic track, there is nothing to extract and you need a speech-to-text transcription instead. For the one-click version of the extraction itself, this tool is what we use.

How long does it take to extract captions from a long video?

Seconds. A caption track is text with timecodes, so even a three-hour lecture extracts far faster than the video plays. Length affects the size of the file, not the wait, in a browser tool that reads the existing track.