YouTube Transcript Scraper - Subtitles & Captions to Text

Scrape YouTube transcripts as text, timed segments, SRT and VTT: paste watch, Shorts, live, embed or youtu.be links or bare 11 character ids, or just type a search, and each video comes back as one row holding the full plain text, the timed segments, a ready-to-save SRT and a ready-to-save WebVTT. No API key, no cookies, no login.

The honest limit: it reads caption tracks the video already has. It does not listen to the audio, so a video nobody captioned and that YouTube never auto-captioned comes back as NO_TRANSCRIPT, and no setting changes that.

Input	YouTube links, video ids, or search keywords
Output	One row per video: full text, timed segments, SRT, WebVTT, the language used and every language available, title, channel, views and length
Ceiling	400 videos per run
Speed	Roughly 20 to 30 transcripts a minute on measured cloud runs
Account needed	None from you
Price	$0.80 per 1,000 transcripts, flat on every plan. The free plan's $5 a month covers about 6,250

🔍 What YouTube Transcript Scraper does

For each video it finds the caption tracks, picks one using languagePreferences in order, and pulls the cues. An exact match wins, then a base-language match, so pt-BR falls back to pt. If none of your preferences exist it takes another available track rather than failing the video, and detectedLanguage tells you what it actually used.

Manual captions are preferred over auto-generated ones when both exist. isAutoGenerated says which you got, and availableLanguages lists everything the video had, so you can re-run for a different one.

Type something into searchKeywords instead and it searches YouTube, takes the top maxVideosPerSearch videos from each search, and handles them like pasted links. Shorts, channels and playlists are left out of the results. Each row it finds carries searchKeyword and searchRank.

One charge covers the whole row: the plain text, the segments, the SRT, the VTT, the language list and the metadata. There is no per-format or per-language extra. A video that turns up in more than one place, links or searches, is processed once.

📋 What data you get from each YouTube transcript

What you get	Field
The whole transcript as one string	text
Every cue with its start, duration and end in seconds	segments, segmentCount
Subtitle files, ready to save	srt, vtt
The language you got and the one you asked for first	detectedLanguage, requestedLanguage
Machine-made track or a person's	isAutoGenerated
Every caption track the video carried	availableLanguages
Video metadata	videoId, url, title, author, channelId, viewCount, durationSeconds, thumbnailUrl
Which search found it, and where it ranked	searchKeyword, searchRank

▶️ How to scrape YouTube transcripts

1. Open YouTube Transcript Scraper and click Try for free.
2. Press Start with the input empty to see the free sample row.
3. Paste links or ids into Video URLs or IDs, or type what you want into Search keywords.
4. Set Preferred transcript languages if English is not what you want.
5. Download the dataset as JSON, or pull srt out of a row and save it as a subtitle file.

💰 How much does it cost to scrape YouTube transcripts?

$0.80 per 1,000 transcripts, the same on every Apify plan. On the free plan, the $5 Apify gives you each month covers about 6,250 transcripts.

You pay per transcript returned. The sample row, diagnostic rows and duplicate ids are never billed, so a batch of 100 videos where 12 have no captions bills 88 transcripts, not 100.

📥 What you give it

{
  "videoUrls": [
    "https://www.youtube.com/watch?v=M7lc1UVf-VE",
    "https://youtu.be/dQw4w9WgXcQ",
    "aqz-KE-bpKQ"
  ],
  "languagePreferences": ["fr-CA", "fr", "en"]
}

Or search instead of pasting:

{
  "searchKeywords": ["sourdough for beginners", "no knead bread recipe"],
  "maxVideosPerSearch": 10
}

Field	Default	What it is
videoUrls	[]	Watch, Shorts, live, embed and youtu.be links, or bare ids. Up to 200. The form opens with one example link.
videoIds	[]	A second list for raw ids, handy when your pipeline stores them separately. Up to 200.
searchKeywords	[]	Searches to run, one per line. Up to 20. Works on its own or next to links.
maxVideosPerSearch	10	How many top results to take from each search, 1 to 100.
languagePreferences	["en"]	BCP 47 codes in priority order, such as en, es, pt-BR. Up to 20.
maxConcurrency	2	Videos handled at once, 1 to 10. Low on purpose: it keeps the run steady and cheap.
retries	2	Retries for a video that did not answer, 0 to 5.
requestTimeoutSecs	45	Total time allowed per video, 10 to 180 seconds.
proxyConfiguration	off	Leave it off. A normal run needs nothing here.

📤 What you get back

A real row from a real run. text, segments, srt and vtt are cut short here, because the full row is about 20 KB of the same transcript in four shapes:

{
  "ok": true,
  "_sample": false,
  "_diagnostic": false,
  "videoId": "M7lc1UVf-VE",
  "url": "https://www.youtube.com/watch?v=M7lc1UVf-VE",
  "title": "YouTube Developers Live: Embedded Web Player Customization",
  "author": "Google for Developers",
  "channelId": "UC_x5XG1OV2P6uZZ5FSM9Ttw",
  "viewCount": 1656845,
  "durationSeconds": 1344,
  "detectedLanguage": "en",
  "requestedLanguage": "en",
  "isAutoGenerated": false,
  "segmentCount": 466,
  "text": "JEFF POSNICK: Hey, everybody. Welcome to this week's show of YouTube Developers Live. ...",
  "segments": [
    { "text": "JEFF POSNICK: Hey, everybody.", "start": 10.349, "duration": 1, "end": 11.349 },
    { "text": "Welcome to this week's show of YouTube Developers Live.", "start": 11.349, "duration": 2.681, "end": 14.03 }
  ],
  "srt": "1\n00:00:10,349 --> 00:00:11,349\nJEFF POSNICK: Hey, everybody. ...",
  "vtt": "WEBVTT\n\n00:00:10.349 --> 00:00:11.349\nJEFF POSNICK: Hey, everybody. ...",
  "availableLanguages": [
    { "languageCode": "en", "languageName": "English", "isAutoGenerated": false },
    { "languageCode": "en", "languageName": "English (auto-generated)", "isAutoGenerated": true }
  ]
}

Field	How to read it
segments	start, duration and end in seconds, for anything timestamp driven.
srt, vtt	Save the string straight to .srt or .vtt.
detectedLanguage, requestedLanguage	Compare them to spot a fallback.
isAutoGenerated	true when the track is machine-made, which is where the punctuation and names get loose.
title, author, channelId, viewCount, thumbnailUrl	Any of them can be null when YouTube did not return it.
searchKeyword, searchRank	Only on rows that came from a search.

🧾 Reading the output

Three kinds of row can land in your dataset, and the flags are how you tell them apart.

Row	How to spot it	Billed
A transcript	ok: true and _sample: false	yes
The sample row	_sample: true, videoId SAMPLE00000	no
A diagnostic	ok: false, _diagnostic: true, an errorCode	no

The Overview table has no _sample column, so the free sample row reads there as a perfectly good transcript. Filter on _sample == false before counting anything.

The error codes are the useful part of a diagnostic row, and each carries a hint:

errorCode	What happened
NO_TRANSCRIPT	The video has no caption track in any language.
VIDEO_UNAVAILABLE	Private, removed, or the id does not exist.
LOGIN_REQUIRED	Age-gated or members-only, so a signed-out reader cannot open it.
INVALID_VIDEO	The string you passed is not a YouTube link or an 11 character id.
RATE_LIMITED	That video would not answer this run. Re-running usually clears it.
NO_SEARCH_RESULTS	A search found no videos. Try fewer or different words.
SEARCH_FAILED	A search could not be run. Try again in a few minutes.
EXTRACTION_FAILED	Tracks were found but no cues came back.
TIME_LIMIT	The run got close to its time limit and stopped early. Every transcript delivered before that is in your dataset; give the run more time, or ask for fewer videos, to get the rest. If none came back at all, the run is marked as failed.
RUN_ERROR	The run itself failed, written as one row so the reason survives.

A batch where every video fails still finishes as a successful run, with diagnostics and nothing else. Count ok: true rows rather than trusting the run status.

💡 What people use it for

• Turning a talk or interview into text you can search, quote and paste into notes.
• Feeding transcripts to a model for summaries, without paying a speech-to-text bill.
• Making .srt files for re-uploads, or as the starting point for a translation pass.
• Finding the exact moment a phrase was said, using segments to jump to the second.

From a topic to what the top videos say and what viewers ask, in two runs:

1. Type a few phrases into Search keywords here.
2. Sort the rows on viewCount and read the first segments of the top videos to see how they open.
3. Paste their url into the Video URLs box of YouTube Comments Scraper to see the questions viewers left.

🚧 What it does not do

• No speech to text. Published caption tracks only, auto-generated ones included. Nothing captioned means nothing to return.
• No translation. It fetches tracks that exist. If the video has no French track, asking for French gets you another language, not a translation.
• One video per entry. Channel and playlist links are not expanded here. YouTube Transcript Bulk Scraper does that.
• durationSeconds can be the transcript's length, not the video's, when YouTube returned no metadata for that video. Sanity check it against the last segment's end if it matters.
• Metadata can be null. Title, author, view count and thumbnail come from YouTube's own response and are sometimes absent, while the transcript itself is fine.
• Auto-generated tracks have no reliable punctuation or speaker names, and get proper nouns wrong. isAutoGenerated is how you know which you are reading.
• Rows are large, roughly 18 to 24 KB each, because text, segments, SRT and VTT hold the same words four ways. A 400 video run is a chunky dataset.
• Search takes YouTube's own top results. There is no filter for upload date, length or views, and a video without captions still uses up one of the places you asked for.
• 400 videos per run in total, links and search results together.

🧭 Which YouTube scraper do you need?

If you want	Use
What a video says, as text, SRT and VTT, for links you already have	This one
Transcripts for a whole channel or playlist in one run	YouTube Transcript Bulk Scraper
The video, its audio or its subtitles as a file	YouTube Downloader Pro
Comments under a video	YouTube Comments Scraper
Videos to run this on, by keyword or a channel's newest uploads	YouTube Scraper
Speech from a video file that has no captions, with your own OpenAI key	Video & Audio Transcriber
The same job on TikTok	TikTok Transcript Scraper

❓ Questions people ask

Do I need a YouTube Data API key?

No. No key, no cookies, no OAuth, no quota.

What if the video has no captions?

You get an uncharged NO_TRANSCRIPT row and the batch carries on. It does not transcribe the audio.

Which language do I get?

The first one in languagePreferences that exists, an exact match before a base-language one. Otherwise another available track, named in detectedLanguage.

Can I get the SRT file directly?

The srt field is the file's contents. Write the string to a .srt and it plays.

Why is one video missing from my results?

Look for its diagnostic row. LOGIN_REQUIRED and VIDEO_UNAVAILABLE are permanent for a signed-out reader, RATE_LIMITED is worth retrying.

Can I call it from code or connect it to an AI assistant?

Yes. The API tab has ready-made code for Python, JavaScript and the command line. For Claude, ChatGPT or another MCP client, connect https://mcp.apify.com/?tools=fetch-actor-details,dami_studio/youtube-transcript-scraper. Either way the run happens on your Apify account at the same price.

Is scraping YouTube transcripts legal?

This reads publicly available captions. What you then do with somebody else's words is your call, and copyright still applies. Apify's write-up on the legality of web scraping is a good starting point, and we are not lawyers.

🆘 If something breaks

Open the Issues tab on the actor page. Include the video link and the run ID, and the errorCode on the diagnostic row if you got one.