Free YouTube subtitle extractor: every method that still works

You want the words out of a YouTube video. Not a summary, not a rewrite, the actual caption text with the timings still attached so you can quote it, search it, translate it or drop it into an editor. That job has a name, subtitle extraction, and for most public videos it takes about ten seconds and costs nothing.

The confusing part is that four different things all get called a free YouTube subtitle extractor, and they behave nothing alike. One copies a file that already exists. One shows you text on screen and makes you drag your mouse over it. One is a developer endpoint that mostly refuses strangers. One transcribes the audio from scratch and charges you in minutes of compute. Pick the wrong one for your job and you either lose an hour or you get a file with no timestamps in it.

This page is the long version. It covers all four routes as they behave in August 2026, what each costs in time for a real 20-minute video, where each one quietly fails, and how to read and repair a caption file once you hold it. One disclosure first: YouTubeScribe is our product, so treat the sections about it as the vendor talking. The limits we list are the real ones, including the videos we cannot touch.

What a subtitle extractor actually does

A YouTube video and its captions are two separate things. The picture and sound live in one stream. The captions live in a timed text file that the player fetches on the side and paints over the frame when you press CC. Nothing about the words is burned into the pixels. So getting subtitles out of a YouTube video is not image processing and it is not audio processing. It is asking for a file that is already sitting there, and keeping a copy.

That single fact explains almost every behaviour people find odd about caption downloaders. Why extraction is fast: it is a file transfer, not a computation. Why it is free: no model runs. Why the accuracy is fixed and no tool can beat another on the same video: everyone is copying the identical source. And why some videos return nothing at all, no matter which of the twenty tools you try.

Two kinds of caption track

YouTube holds two species of track and an extractor will happily take either one.

• Creator uploaded tracks. Somebody wrote or corrected these and pushed them to the video. They carry real punctuation, speaker labels where the author bothered, correct spellings of names and products, and sensible line breaks. This is the good stuff.
• Automatic captions. YouTube's speech recognition made these when the video finished processing. No punctuation in many languages, no speaker labels, proper nouns often mangled, and line breaks placed by machine at roughly even intervals rather than at clause boundaries.

You can usually tell them apart in a second. Open the caption menu in the player. A track labelled with a language name and the word auto-generated after it is machine output. A bare language name is human work. Extractors read both, and a good one tells you which kind it handed you, because the difference decides how much editing you are about to do.

Automatic captions do not cover every language. YouTube's own help pages list roughly 70 supported languages for automatic captioning, running alphabetically from Afrikaans to Zulu. That number is worth remembering because it is smaller than most people assume, and it is the reason a perfectly clear video in a less common language sometimes has no track at all.

Extraction is not transcription

These two words get used as if they mean the same thing and they do not. Extraction reads an existing caption track and copies it. Transcription takes the audio and produces new text using speech recognition. Both end with words on your screen. Everything else about them differs.

YouTubeScribe sits firmly on the extraction side. It reads the caption track a public or unlisted video already has. It never invents speech and it does not run speech recognition on the audio. If YouTube never wrote a track, the answer you get back is nothing, and that is the correct answer rather than a bug.

What a caption cue looks like inside

Open any SRT file in a text editor and the structure is obvious within five seconds. Each cue has four parts: a sequence number on its own line, a timing line with a start time, an arrow and an end time, one or two lines of text, then a blank line before the next cue. Times run hours, minutes, seconds, then a comma and three digits of milliseconds.

WebVTT holds the same information with two differences that trip people up. The file must begin with the literal word WEBVTT on the first line, and the milliseconds separator is a full stop rather than a comma. Sequence numbers are optional. That is genuinely most of the difference for ordinary subtitle work, which is why converting between the two is a find and replace rather than a parse and rebuild.

Once you can read a cue, you can audit any extractor's output in about thirty seconds. Check that the first cue starts near the first spoken word. Check that the last cue lands near the end of the video. Scan the middle for a gap of several minutes, which usually means the track itself is partial rather than the download failing.

The four ways to get subtitles out of a video

Here is the honest comparison. Times are for a single 20-minute talk with an existing English caption track, measured start to finish, including the fiddly bits everybody forgets to count like opening a new tab and cleaning up the paste.

Method 1: paste the link into a free extractor

The fastest route by a wide margin. You copy the video URL, paste it, choose a format, and a file lands in your downloads folder. There is nothing to install, nothing to authorise, and no account needed to pull a transcript on the free tier.

What makes one tool better than another here is not accuracy, since everybody reads the same source track. It is the boring operational stuff: whether long videos are refused, how many languages are offered when a video carries several tracks, whether SRT costs money, whether your video URLs are logged, and whether the tool admits it found nothing instead of inventing a transcript.

On YouTubeScribe there is no hard length cap. A three-hour conference recording comes back in about twelve seconds. It reads 125 languages where a track exists, which is not the same claim as YouTube's roughly 70 automatic caption languages: those 70 are the languages the machine will listen in, while the 125 are the languages we can read back out once any track exists, including every human-uploaded translation. The Subtitle Downloader is the page built for this job, and it gives you TXT and SRT free, with VTT on Pro.

Method 2: YouTube's own transcript panel

YouTube ships a reading view for captions and most people have never opened it. On desktop, go under the video, open the description, and look for the button marked Show transcript. On mobile the same control lives in the description sheet. A scrollable panel appears beside the player with the caption text, each line prefixed by its start time, and clicking any line jumps the player there.

It is genuinely useful for reading. There is a three-dot menu inside the panel that toggles the timestamps off, and where a video carries multiple tracks a language picker appears at the top. For finding the one sentence you half remember, this beats everything else because you can scan and jump in the same motion.

The problem is getting the text out. There is no download button. You select the text with the mouse, which on a long video means a scroll-and-drag that keeps snapping to the video player, and what lands on your clipboard is a wall of text with timestamps welded into the same lines. It is not a cue file. Nothing downstream that expects SRT will accept it. For one quote it is fine. For a 90-minute podcast it is a genuinely unpleasant ten minutes.

• Good for: one quote, a fast skim, checking whether a track exists before you bother with anything else.
• Bad for: any file you need to feed to an editor, a translator, a search index or a script.
• Watch out for: the panel showing the auto track when a better human track exists, since the language picker defaults are not always the best track available.

Method 3: download the caption file by hand

There are two versions of this and people conflate them constantly.

The first is YouTube Studio. If the video is on your own channel, open Studio, go to Subtitles, pick the video, open the track and use the download option. You get the real file, exactly as stored, in the format you ask for. This is the cleanest source there is. It is also useless for anybody else's video, which is the case most people are actually in.

The second is the YouTube Data API. There is a captions resource with a list method that enumerates the tracks on a video, and a download method that returns the track body. The list call is broadly available. The download call is the one that stops people: it needs OAuth rather than a plain API key, and it is scoped to captions you own or where the owner has granted third-party contributions. It was built so a creator or their caption vendor can pull their own files, not so anybody can harvest anybody's. Read the terms before you build on it.

Budget an afternoon for the API route the first time: a Google Cloud project, the API enabled, an OAuth consent screen, a client secret, a token exchange, then the actual two calls. If you own the videos and you need this to run nightly on a hundred uploads, it is the right tool. For a handful of videos you do not own, it is the wrong shape entirely.

Method 4: run speech recognition yourself

The last resort, and the only route that works when no caption track exists. You pull the audio and run it through a speech model, locally or through a paid API, and it writes fresh cues with fresh timings.

It is the only method here that can beat the original: a good modern model on clean audio will often out-punctuate and out-spell a lazy auto track. It is also the only method that can be badly wrong in ways you will not notice, because a confident model produces fluent text for mumbled audio and you have no source file to check it against.

• Use it when: the video has no track at all, the auto track is unusable, or you need a language YouTube's automatic captioning does not cover.
• Skip it when: a human-uploaded track already exists, because you will spend forty minutes producing something worse than a free ten second download.
• Always check: the first and last minute, and any section with music or crosstalk. That is where invented text hides.

Step by step with a free extractor

This is the paste-a-link route, written out fully so you can see exactly where the seconds go. The tool is the Subtitle Downloader on YouTubeScribe, but the shape of the process is the same on any honest extractor.

1. Copy the video URL. The share button gives you a short youtu.be link, the address bar gives you the long watch link with a v parameter. Both work. A link with a timestamp on the end also works, the timestamp is simply ignored.
2. Open the Subtitle Downloader and paste the link into the field.
3. Pick the language if the video carries more than one track. Where several exist, prefer the one without the auto-generated label.
4. Pick the format. TXT for reading and writing, SRT for editing and re-uploading, VTT for the web player.
5. Run it. A 20-minute video comes back in a couple of seconds. A three-hour recording takes about twelve.
6. Check the file before you build on it. Open it, confirm the first cue lines up with the first words spoken, and scroll to the end to confirm it did not stop early.

Step six is the one everybody skips and it is the one that saves you. A partial track looks exactly like a complete one until you reach the end of it.

Choosing the output format

Match the format to what happens next, not to what sounds most technical.

• TXT when a person is going to read it, or a language model is going to summarise it, or you want to paste it into a document. No timing noise, no cue numbers, just prose. Free.
• SRT when the text is going back onto video, into a subtitle editor, into a translation workflow, or into anything that needs to know when each line appears. The most widely accepted subtitle format there is. Free.
• VTT when the destination is a web player using the HTML track element, or a platform that specifically asks for WebVTT. Available on Pro.

One habit worth forming: keep the SRT even when you only wanted the text. Going from SRT to TXT is trivial at any point later. Going from TXT back to SRT means re-timing every line by hand, and nobody has ever enjoyed that.

More than one video at a time

One video is easy. Forty is where tools separate. If you are pulling a whole course, a conference series or a competitor's back catalogue, feeding links one by one is the slow way to spend a morning.

The Playlist Export tool takes a playlist or channel URL and works through it, so you queue the list once and collect the files at the end. If what you want is readable prose rather than cue files, the Transcript Generator is the sibling tool pointed at that job: same extraction underneath, output shaped for reading and quoting rather than for a subtitle editor.

Expect misses in any batch. A playlist of sixty will usually contain two or three videos that come back empty, because a member-only upload slipped in, or a very short clip never got a track, or a re-upload is still processing. That is normal. Note them and move on rather than assuming the run failed.

The limits, stated plainly

Since this is our tool, here is what it will not do, in the same place as what it will.

• Private and members-only videos return nothing. There is no track we are permitted to read, and there is no clever workaround being withheld from you.
• It does not run speech recognition. No track means no output, full stop.
• It cannot improve a bad automatic track. Garbled audio in, garbled captions out, because the garbling happened at YouTube months ago.
• The free tier stores nothing. That means privacy, and it also means the file is yours to keep safe, because we cannot re-send you something we never kept.
• The first transcript each day is free. TXT and SRT are free formats. VTT is a Pro format.

The single most common support message we get is that the tool is broken because it returned nothing. In almost every case the video genuinely has no caption track. The fix is not a different extractor, it is speech recognition or a different source video.

YouTubeScribe support notes, June to August 2026

SRT, VTT, TXT: picking and fixing the file

You have a file. Now the questions change: will the thing you are feeding it to accept it, and if not, what do you edit.

SubRip, the default answer

SRT is the plain cotton of subtitle formats. Sequence number, timing line, text, blank line, repeat. It carries no positioning and no styling, and that poverty is precisely why everything accepts it: there is nothing in the file to misinterpret. Video editors, translation platforms, social schedulers and every upload form you will meet take SRT.

YouTube's own upload documentation describes SRT as plain UTF-8 text with no markup, which is the standard everyone else follows too. If you open an SRT and see styling tags in it, some tool has been creative and a strict parser downstream may object.

WebVTT, the web native one

WebVTT is the W3C format that browsers understand natively through the HTML track element. Same cue idea as SRT, plus cue settings for position, alignment and line placement, plus a proper header. YouTube accepts it on upload and supports its positioning, though it limits styling to bold, italic and underline rather than the full range the spec allows.

Convert SRT to VTT by hand in three moves: add a line reading WEBVTT and a blank line at the top, change every comma in the timing lines to a full stop, and save as UTF-8 without a byte order mark. Going the other direction, strip the header and reverse the separator. The cue numbers are optional in VTT so you can leave them or drop them.

Plain text, and what it costs you

TXT is a caption file with the timing thrown away. That is a real loss and a real gain. The gain is that prose reads like prose, so summarising, searching, quoting and pasting all get easier, and a language model handling it wastes no attention on timecodes. The loss is that you can never put it back on video without re-timing every line.

Rule of thumb: if a human or a model is the final reader, take TXT. If a player is the final reader, take SRT or VTT. If you are not sure, take SRT, because it converts down to TXT in one step whenever you decide.

What YouTube accepts if you put captions back

Extraction is often only half the job. People pull a track, fix the spellings, translate it and push it back up. YouTube's supported files list is much wider than most people expect, and worth knowing before you convert something unnecessarily.

The practical reading of that table: if your file is already in one of those shapes, upload it as it is. Converting a TTML file down to SRT to be safe throws away positioning that YouTube would have honoured. If you are authoring from nothing, SRT stays the sensible default because it is the format every other tool in your chain will also accept.

Five file problems and their fixes

1. Accented characters appear as question marks or mojibake. The file is not UTF-8. Re-save it as UTF-8 and the problem disappears.
2. A strict parser rejects a valid looking VTT. Check for a byte order mark before the WEBVTT header. Save without a BOM.
3. Every subtitle shows a fixed amount early or late. That is a constant offset, not a broken file. Any subtitle editor will shift the whole track by a set number of milliseconds in one action.
4. The upload form refuses an SRT. Look for styling tags or HTML inside the cues. SRT is supposed to be plain text and some parsers enforce it.
5. Cues render on one line where you expected two. Line breaks inside a cue are part of the text and survive conversion, so if they vanished, the tool that wrote the file flattened them.

When there is nothing to extract

The hardest part of writing about extraction honestly is this section, because the answer is sometimes that you cannot have the thing you want. No tool on the internet can copy a file that was never written.

Google's own list of reasons

YouTube documents six reasons automatic captions may be unavailable on a video. Every one of them is a genuine dead end for an extractor, not a limitation of the tool you are using.

• The captions are not available yet because the audio is complex and still processing.
• Automatic captions do not support the language spoken in the video.
• The video is too long.
• The sound quality is poor, or the speech is not recognised.
• There is a long period of silence at the beginning of the video.
• Multiple people are speaking over each other, or multiple languages are spoken at once.

Two of those are worth a second look. The processing one is temporary: a video uploaded in the last hour may simply not have a track yet, and coming back tomorrow fixes it. The silence one catches a surprising number of creators, because a long branded intro with only music at the top of the video can be enough for the recogniser to give up before anyone speaks.

The access wall

Separate from whether a track exists is whether you are allowed near it. Public and unlisted videos are readable. Private videos are not, and neither are members-only uploads, because both sit behind an access check that an extractor has no business defeating.

If a video is private and you have a legitimate reason to need its captions, the route is the channel owner: they can pull the file from YouTube Studio in about a minute and send it to you. That is a faster conversation than any amount of tool-hunting.

The language gap

Two numbers matter here and they measure different things. YouTube's automatic captioning covers roughly 70 languages. YouTubeScribe reads 125 languages where a track exists. The second number is larger because it includes every human-uploaded and human-translated track, which can be in a language the machine recogniser has never supported.

So a video in a language outside the automatic 70 will only have captions if a person made them. When creators in those languages do upload their own tracks, extraction works perfectly and the quality is usually better than any auto track, because a human wrote it.

A five-minute triage

Before you conclude a tool is broken, run this. It takes less time than opening a support ticket.

1. Open the video and look at the CC button in the player. Greyed out or missing means there is no track and nothing to extract. Stop here.
2. Open the caption menu and read the language list. Note whether the entry says auto-generated, and whether a better human track sits below it.
3. Open the description and press Show transcript. If the panel appears with text, a track exists and any extraction failure is a tool problem worth reporting.
4. Check the upload date. Uploaded in the last few hours means the track may still be processing. Try again tomorrow.
5. Check the access state. A padlock, a members badge or a video that only plays for you when logged in means the access wall, not a missing file.
6. Only after all five, reach for speech recognition. If the video truly has no track, that is the one route left, and it is a real one.

The summary is short. For a public video with an existing track, paste the link into a free extractor and you are done in seconds with a real SRT. For a fast quote, YouTube's own transcript panel is right there and costs nothing. For your own uploads at scale, Studio or the API gives you the pristine originals. And for a video with no track at all, speech recognition is the only honest answer, with all the cost and all the caution that implies.

Questions people ask

Sources

1. YouTube Help. Use automatic captioning. support.google.com· Checked 17 August 2026.
2. YouTube Help. Supported subtitle and caption files. support.google.com· Checked 17 August 2026.
3. YouTube Help. Add subtitles and captions. support.google.com· Checked 17 August 2026.
4. W3C. WebVTT: The Web Video Text Tracks Format. w3.org· Checked 17 August 2026.
5. Google for Developers. YouTube Data API: Captions. developers.google.com· Checked 17 August 2026.
6. YouTube Help. View video transcripts. support.google.com· Checked 17 August 2026.
7. YouTube Official Blog. Caption my YouTube videos. blog.youtube· Checked 17 August 2026.
8. W3C. TTML2: Timed Text Markup Language 2. w3.org· Checked 17 August 2026.
9. Google for Developers. YouTube Data API: Captions: download. developers.google.com· Checked 17 August 2026.
10. Google for Developers. YouTube Data API: Captions: list. developers.google.com· Checked 17 August 2026.
11. Google for Developers. YouTube API Services Terms of Service. developers.google.com· Checked 17 August 2026.
12. YouTube. Terms of Service. youtube.com· Checked 17 August 2026.
13. AssemblyAI. What is a subtitle file format? assemblyai.com· Checked 17 August 2026.
14. Ditto Transcripts. SRT vs VTT: understanding the difference between subtitle formats for captions. dittotranscripts.com· Checked 17 August 2026.
15. MDN Web Docs. WebVTT API. developer.mozilla.org· Checked 17 August 2026.
16. MDN Web Docs. The Embed Text Track element (track). developer.mozilla.org· Checked 17 August 2026.
17. Wikipedia. SubRip. en.wikipedia.org· Checked 17 August 2026.
18. YouTubeScribe. Extraction test log, June to August 2026. Internal measurements across public and unlisted videos. Checked 17 August 2026.
19. YouTubeScribe. Language coverage audit, August 2026. Internal record of the 125 languages read where a caption track exists. Checked 17 August 2026.

Reviewed on 17 August 2026. If a line is wrong, email support@youtubescribe.com.