How to get a transcript from a YouTube video

You want the words out of a video. Maybe it is a two hour lecture you need to quote accurately, a product review you would rather skim in thirty seconds than watch in twenty minutes, or a client's own upload that has to become a blog post by Friday. The useful news is that the words usually already exist. YouTube wrote them down when the video was published, and they sit next to the video as a caption track.

That one fact changes how you should think about the whole task. Getting a transcript from a YouTube video is mostly a retrieval problem. You are not asking a machine to listen to audio and guess at words. You are asking it to fetch a file that is already there and hand it to you in a shape you can use. Retrieval takes seconds. Listening takes as long as the video does.

This guide covers the four routes people actually use, the exact steps for each, the three YouTube link shapes and what to do about the junk that gets stuck to them, and the documented reasons a transcript can be missing entirely. One disclosure up front: YouTubeScribe is our product, so read the parts about it as a vendor describing their own tool, and check them against the free routes below.

What a YouTube transcript actually is

A transcript is the spoken content of a video written down. On YouTube it comes from timed text: a list of cues, where each cue carries a start time, an end time, and a line or two of words. Play the video with captions on and you see one cue at a time. Open the transcript panel and you see all of them stacked in a column. Same underlying data, two different views of it.

Those cues have exactly two possible origins, and which one you got decides how much cleanup work is waiting for you.

Uploaded captions versus automatic captions

Uploaded captions are written by a person, or by a paid captioning service, and pushed to the video by whoever owns the channel. They carry punctuation, sentence case, speaker labels when the creator bothered, and sound cues in square brackets. News channels, university courses, and anything made with an accessibility budget tend to have them. When you pull one of these, the transcript needs almost no work.

Automatic captions are produced by YouTube's own speech recognition when the video is processed after upload. They are free, they arrive without anyone asking, and they cover roughly seventy languages. Quality tracks the audio. Clear studio speech in a supported language comes back close to perfect. Two people talking over each other in a noisy room comes back as a mess, and sometimes does not come back at all.

You can usually tell which kind you have from the first ten lines. Punctuation and capital letters at the start of sentences mean a human was involved somewhere. A wall of lowercase words with no full stops means the machine did it alone. Older automatic tracks in particular carry no punctuation whatsoever.

Captions, subtitles and transcripts are three different words

The three terms get used interchangeably and they should not be. Captions are the timed cues drawn over the player, meant for people who cannot hear the audio, which is why they include sound cues like [Applause]. Subtitles are translations of the speech for people who can hear the audio but do not speak the language. A transcript is the running text, with the timing either shown as a column beside it or discarded.

That distinction matters when you pick a file format later. A caption file keeps the timings in a structure a player can read. A transcript in plain text usually throws them away. Ask what the text has to do next before you choose.

The track exists before you ask for it

Nothing about fetching a transcript makes YouTube listen to the video again. The track was written once, at publish time or when the creator uploaded a caption file, and it is served alongside the page. That is why a three hour podcast and a three minute clip take roughly the same amount of time to pull. You are downloading a text file, and text files are small.

Four routes to a transcript, compared

Four routes cover essentially every case. They differ in effort, in whether you end up holding a file, and in whether they work at all when the video is a Short.

Route one: YouTube's own Show transcript panel

This is the free route with no third party involved, and it should always be your first try. It sits in plain sight and most people have never noticed it.

1. Open the video on youtube.com in a desktop browser.
2. Look below the player, in the row under the title and description, and find the three dot More menu.
3. Choose Show transcript.
4. A side panel opens to the right of the video with the full text, one cue per line, each with a timestamp.
5. Click any line to jump the player to that moment.
6. To copy, select the text inside the panel and copy it in the normal way.

Two limits catch people out. First, YouTube gives you no download button. There is no Save as TXT, no export, no menu item hiding under a submenu. You select text with the mouse and copy it, and on a long video that means a long drag through a scrolling panel. Second, the panel is not available on every video, and when it is not available the menu item simply is not there.

The panel has its own small menu that turns the timestamp column off, which is worth doing before you copy if you want readable prose. There is also a language picker on videos that carry more than one caption track.

Transcripts are not available for all videos.

YouTube Help, View video transcripts

On phones the panel exists in the YouTube app, tucked under the description rather than beside the player. Selecting several thousand words on a touchscreen is miserable work. If you are on a phone and you want the whole thing, use a link based route instead and save yourself the fight.

Route two: paste the link into a transcript generator

A transcript generator asks YouTube for the same track the panel shows, then hands you a file instead of a scrolling column. The whole interaction is: copy link, paste, press the button. YouTubeScribe is ours, and the page built for this job is the Transcript Generator at /tools/transcript-generator.

1. Copy the video link from the address bar or the Share button. Any of the three shapes works.
2. Paste it into the box and press the button.
3. Wait. A three hour video takes about twelve seconds, and there is no hard length cap.
4. Toggle timestamps on or off. The text rewrites in place, so you can look at both and decide.
5. Copy the text, or download it. TXT and SRT are free. VTT is on the Pro tier.

A few practical notes on the free tier. The first transcript each day needs no account. Nothing is stored, so the text is gone when you close the tab, which is either a privacy feature or an annoyance depending on why you came. Language coverage runs to about a hundred and twenty five languages, but only where a track already exists, because the tool reads rather than listens.

Two sibling pages handle the neighbouring jobs. The Subtitle Downloader is the one to reach for when you want a caption file for a video editor rather than text to read. The Shorts Transcript Downloader accepts /shorts/ links directly, if you would rather not think about link shapes at all.

Route three: a browser extension

Extensions inject a button into the YouTube page itself, usually near the subscribe row, and pull the transcript without you leaving the tab. When one works properly it is the fastest option on this list, because nothing gets copied between windows.

The costs are real, though. An extension needs permission to read and change pages on youtube.com, which is a broad grant to hand to a developer you have never heard of. YouTube ships layout changes to the watch page often, and every change breaks a handful of extensions until their authors catch up. You install per browser and per machine, so the workflow does not follow you to a phone or a borrowed laptop. And an extension whose last update was eighteen months ago is a security question, not just a stale button.

Use one if you pull transcripts from ten videos a day and the friction actually costs you time. For occasional work, the link route costs less trouble overall.

Route four: run speech recognition yourself

This is the only route that produces text when YouTube never wrote a track. You play the audio into a speech to text engine, either by pointing a local model at a downloaded audio file or by running a dictation tool while the video plays.

It is slow, and the slowness is structural rather than a tuning problem. Real time transcription takes as long as the video. Even a fast local model wants a meaningful share of the runtime, plus setup you only pay once but do have to pay. Accuracy depends on your audio chain, and a laptop microphone picking up laptop speakers is the worst version of that chain. Timings, if you need them, come out approximate.

Keep this route in your back pocket for the videos where nothing else works, and treat the effort as the price of text that does not otherwise exist anywhere.

One line summary of that table: the panel is free and immediate but gives you no file, the link route gives you a file in two steps, the extension trades permissions for speed, and speech recognition is the fallback that always works and always costs the most.

Every YouTube link shape, and what to paste

Everything in this section reduces to one idea. A YouTube video is identified by an eleven character video ID, and every link shape is a different envelope around that ID. Get the ID across intact and the rest is decoration.

The ID uses upper and lower case letters, digits, hyphen and underscore. Case matters, and so do the hyphens and underscores. dQw4w9WgXcQ and dqw4w9wgxcq are not the same video, and one of them does not exist. This is the single best reason to copy links rather than retype them from a screenshot or a slide.

Shape one: youtube.com/watch?v=VIDEOID

This is the desktop address bar form, and it is what you get when you open a video from search, the home feed, a channel page or a subscription. The ID lives in the v query parameter. Everything after it is context YouTube adds for its own purposes: where you came from, what should play next, which moment to start at.

Paste it whole. Any route worth using reads the v parameter and ignores the remainder. If you want a tidy link for a document, keep from https:// through the eleven characters after v= and delete everything from the next ampersand onward.

Shape two: youtu.be/VIDEOID

This is the short link the Share button produces, on desktop and inside the mobile app. There is no v parameter here at all. The ID is the path segment, sitting directly after the slash, and youtu.be is YouTube's own redirect domain rather than a third party shortener.

It resolves to exactly the same video and nothing is lost in the conversion. If a tool rejects a youtu.be link, that tool is doing naive string matching for 'watch?v=' somewhere in its code, and you should expect other rough edges from it.

Shape three: youtube.com/shorts/VIDEOID

Shorts use a path form, like the short link but on the main domain. The ID is still eleven characters and it is still an ordinary video ID. A Short and a regular upload are the same kind of object wearing a different player.

Which leads to a trick worth knowing. Swap /shorts/ for /watch?v= in the address bar and the same video opens in the standard player, three dot menu and transcript panel included. If the transcript panel refuses to appear on a Short, try that before you conclude there is no track.

Shorts do fail more often than long videos, and the reasons are structural rather than mysterious. They are short, so there is less speech for the recogniser to work with. They are frequently music first with very little dialogue. And a Short that opens with two seconds of a logo sting can trip the long silence at the start condition. The Shorts Transcript Downloader handles the link shape without complaint, but no tool can produce a track that was never written.

Links carrying extra query parameters

This is where most of the confusion lives, so here is the full list of what you are likely to see hanging off a link and what to do about each one.

• &t=125s or ?t=125. A start time, added when you share from a specific moment. It changes where the player begins playing. It does not trim the transcript, and every route returns the full text regardless. Leave it or strip it, the result is identical.
• &list=PLxxxxxxxx. Playlist context. The link is still a watch link with a v parameter in it, so it works exactly as pasted. The playlist only tells the player what to queue next.
• &index=7. The position of this video inside that playlist. Harmless, ignored.
• ?si=xxxxxxxx. A share identifier appended by the mobile share sheet and the desktop Share dialog. It carries attribution information back to YouTube. It does no damage to a transcript request, but strip it before pasting the link into a public document.
• &pp=xxxxxxxx and other opaque parameters. YouTube adds these for internal reasons and changes them without notice. Ignore them.
• &feature=share, &ab_channel=Name and similar legacy tags. Old, harmless, ignorable.

The general rule: nothing in the query string after the video ID affects which text comes back. The only thing that would is a link with no video ID inside it at all.

Links that will not work, and what to use instead

• youtube.com/playlist?list=PLxxxx. A playlist page contains no single video, so there is nothing to transcribe. Open a video from the playlist and use that link. If you want the whole playlist, work through it one video at a time.
• youtube.com/@handle or youtube.com/channel/UCxxxx. A channel page, not a video. Open any video on it first.
• youtube.com/clip/xxxxx. Clip identifiers are not video IDs. Open the source video from the clip page and use its link, then find your section by timestamp.
• youtube.com/embed/VIDEOID. This one does contain a real video ID and most routes accept it, but the plainer forms are safer to hand around.
• youtube.com/live/VIDEOID. Live streams and premieres. Once the broadcast has finished and processed, these behave like normal videos. During a live broadcast the caption track is still being written, so wait until it ends.
• Links from music.youtube.com. Music tracks usually carry no caption track at all, and the ones that do are lyric videos.

If you do need to clean a link by hand, for a document or a bug report rather than for a tool, the process takes four steps.

1. Copy the link from the address bar or the Share button. Never retype it.
2. Find the video ID: it follows v= on a watch link, and it follows the last slash on a youtu.be or /shorts/ link.
3. Confirm it is eleven characters, and confirm the upper and lower case letters survived the copy exactly.
4. Rebuild it as https://www.youtube.com/watch?v= plus those eleven characters, and delete everything else.

Timestamps, file formats, and cleaning the text

You have the text. What it has to do next decides which switches you should have flipped on the way out.

With timestamps or without

Timestamps are the times attached to each line of cues. Keep them when you will need to point back at the video: research notes, quotes you may have to defend, chapter markers, subtitle work, anything where somebody will eventually ask where a claim came from. Drop them when the text will be read as prose, pasted into a document, summarised, or fed to a language model that will only trip over the numbers.

In YouTube's own panel, timestamps are on by default and the panel menu turns them off. In a link based tool it is usually a toggle that rewrites the text in place, so you can look at both versions before you copy anything.

TXT, SRT and VTT, and when each one is right

Three formats cover practically the whole field.

• TXT is plain running text, with or without a timestamp column. Use it for reading, quoting, pasting into a document, or handing to a summariser. It carries no timing structure, so no video player can use it.
• SRT is the SubRip format: numbered cues, a start and an end time, then the text. Nearly every video editor imports it, which makes it the right pick for burning subtitles into a cut or building a caption track for a re-upload.
• VTT is WebVTT, the W3C format designed for the web. It is what the HTML track element expects, so use it when the captions have to play inside a browser video player. It also supports styling and positioning that SRT does not. On YouTubeScribe, TXT and SRT are free and VTT sits on the Pro tier.

If you cannot decide, take TXT for reading and SRT for anything that has to sit alongside video. VTT is a specific answer to a specific web problem, and you will know when you have that problem.

Copying cleanly out of the YouTube panel

If you are staying on the free panel route, the copy step is where all the work hides. Four things make it less painful.

1. Open the panel menu and turn timestamps off first. Removing them afterwards means a find and replace with a regular expression, which is a worse afternoon than one click.
2. Click at the start of the first line, scroll to the bottom of the panel, then shift click at the end of the last line. Dragging through a scrolling panel tends to select the whole page instead.
3. Paste into a plain text editor rather than a word processor, so you drop YouTube's formatting rather than importing it.
4. Expect one line break per cue. Cues break on timing, not on sentences, so a single paragraph of speech arrives as a dozen short lines and you will have to join them by hand.

That last point is the honest argument for the link route. A downloaded TXT file arrives already joined into readable paragraphs. The panel gives you cue fragments and leaves the joining to you, which on a ninety minute video is genuinely half an hour of dull work.

Cleaning automatic caption text

Automatic tracks come with predictable rough edges. Fixing them is a ten minute job on a long video once you know exactly what you are looking for.

• No punctuation, or punctuation only at obvious pauses. Older tracks have none at all, so you will be inserting full stops by ear.
• Everything in lowercase, including names and places. Proper nouns need fixing by hand or with a prepared find and replace list.
• Bracketed sound cues such as [Music] and [Applause]. Search and delete them in a single pass before you do anything else.
• Doubled words at cue boundaries, where the recogniser repeated a word across the seam between two cues.
• Filler kept verbatim. Um, uh and you know are transcribed faithfully, which is correct behaviour for a caption and wrong for a printed quote. Cut them if you must, and say in your piece that you did.
• Homophone errors that read as real words, which is exactly why they survive a spell check. Read anything you plan to publish rather than trusting the file.

Uploaded human captions skip almost all of that, which is a good reason to check which kind of track you have before you budget the time. Ten lines of reading tells you.

When no transcript exists, and how to tell

Sometimes there is nothing to fetch. This is the part most guides skip, and it is the part that costs people the most time, because they keep trying new tools against a video that has no track for any of them to read.

The six documented reasons automatic captions are missing

Google publishes the list, and it is short enough to memorise.

• The captions are still processing. Complex audio takes time after upload, and a very new video sometimes just needs another hour.
• The language is not supported by automatic captions. Coverage runs to roughly seventy languages, which is a lot but not all of them.
• The video is too long.
• The sound quality is poor, or the speech is not recognised.
• There is a long silence at the start of the video.
• Multiple people are speaking over each other, or several languages are spoken at once.

Hold that list next to a video that returned nothing and the cause is usually obvious inside a few seconds. A silent cold open. A noisy live recording from the back of a room. A panel discussion where four people talk at once. A language outside the supported set. None of these are tool failures, and switching tools will not fix any of them.

The other reasons, which are about access rather than audio

• Private videos. Not visible without specific permission, so no route can read them. Unlisted is a different matter and works normally, because unlisted still means anyone holding the link can watch it.
• Members only videos sitting behind a channel paywall. Same answer as private.
• Deleted videos, and videos blocked in your region.
• Age restricted videos, which behave inconsistently depending on the route and the session.
• Creators who turned captions off for that video deliberately.

In every one of those cases the correct output is nothing at all. A tool that returns text for a genuinely private video is either wrong about which video it fetched or is doing something it should not be doing.

If YouTube never wrote a caption track, no link based tool can hand you one. Anything that returns text in that case is running speech recognition on audio, which is a different job with a different bill.

YouTubeScribe, extraction test notes

A triage order that saves time

1. Open the video in a desktop browser and check the three dot menu for Show transcript. If it is there, a track exists and any route will find it.
2. If the menu item is missing, turn captions on with the CC button in the player. Captions playing means a track exists and something else is wrong with the route you tried.
3. If it is a Short and it refuses, swap /shorts/ for /watch?v= in the address bar and check the menu again.
4. Check the upload time. A video published in the last hour may still be processing, and a retry later the same day often succeeds.
5. Check the language and the audio against the six reasons above. Overlapping speakers, heavy background music, a silent opening: each one is documented.
6. If all of that says there is no track, stop shopping for tools. Your only remaining option is running speech recognition on the audio yourself.

What we see in practice

The following is illustrative, drawn from our own extraction testing between June and August 2026, and it is not an audited study. Across a mixed sample of public videos, misses clustered hard rather than spreading evenly. Long form talking head content and conference recordings almost always had tracks. Shorts missed far more often than anything else in the sample. Music heavy uploads with little spoken dialogue were the next worst group. Videos published within the previous hour sometimes missed and then succeeded on a retry later the same day, which matches the still processing reason exactly.

The practical lesson from all of that is simple. Check for a track before you shop for a tool. It takes fifteen seconds in the three dot menu, and it tells you whether your problem is the video or the route you picked.

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. Google for Developers. YouTube Data API: Captions: list. developers.google.com· 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: Videos. developers.google.com· Checked 17 August 2026.
11. Google for Developers. YouTube Embedded Players and Player Parameters. developers.google.com· Checked 17 August 2026.
12. YouTube Help. Embed videos and playlists. support.google.com· Checked 17 August 2026.
13. MDN Web Docs. WebVTT API. developer.mozilla.org· Checked 17 August 2026.
14. MDN Web Docs. Web Speech API. developer.mozilla.org· Checked 17 August 2026.
15. W3C. Timed Text Markup Language 2 (TTML2). w3.org· Checked 17 August 2026.
16. W3C Web Accessibility Initiative. Captions and Subtitles. w3.org· Checked 17 August 2026.
17. YouTubeScribe. Extraction test log, June to August 2026.

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