YouTube Transcript & Subtitles Scraper

Get the transcript of any YouTube video as clean text, ready for an LLM, a RAG pipeline or a spreadsheet. Paste video URLs or IDs, a channel, a playlist or a search, pick a language, and get one row per video with the full transcript, timestamps and video details. Turn on SRT, VTT, plain-text, LLM-ready Markdown or RAG chunks split by YouTube chapters, at no extra cost.

You pay $3 per 1,000 transcripts ($0.003 each), and less on Apify Store discount tiers (Bronze, Silver, Gold), down to $2.10 per 1,000 on Gold. Videos that fail or have no captions are never charged.

At a glance

Topic	Details
Input	YouTube video URLs or IDs (watch, youtu.be, Shorts, embed and live links), channels, playlists and search queries
Output	One JSON row per video: transcript as one plain-text string, segments with start times, plus title, channel, duration and language. Optional: views, likes, comments, tags, chapters, and SRT, VTT, plain-text, LLM-ready and RAG chunk formats
Price	$0.003 per transcript ($3 per 1,000), or $2.70 per 1,000 on Bronze, $2.40 on Silver and $2.10 on Gold. No start fee. Videos without captions and failed videos are free
Tested	Final test run on 16 Oct 2026 with every option on: 56 video links, 4 channels and playlists and 3 searches gave 88 videos, and all 87 that exist returned a transcript with every format, in 73 seconds
Limits	Reads YouTube's own captions and does not run speech-to-text, so a video with no captions returns a free no_transcript row. No cap on videos per run, 10 fetched in parallel by default

Call it from Python

from apify_client import ApifyClient

client = ApifyClient("<YOUR_APIFY_TOKEN>")
run = client.actor("heyitsgautham/youtube-transcript-scraper").call(
    run_input={"videos": ["https://youtu.be/jNQXAC9IVRw"], "includeTimestamps": False}
)
for row in client.dataset(run["defaultDatasetId"]).iterate_items():
    print(row["status"], row.get("title"), row.get("transcript"))

Channels, playlists and search

Besides videos, two optional inputs add videos in bulk.

Input	What it takes
channelsAndPlaylists	Channel links in any form (@handle, /channel/UC..., /c/name, /user/name), playlist links with list=..., and bare @handles, UC... channel IDs or PL... playlist IDs. Channels go newest first
searchQueries	YouTube searches, each giving its top videos in YouTube's order
maxVideosPerSource	How many videos to take from each channel, playlist or search. Default 100, 0 means all, up to 20,000
includeShorts	Channels only. Off: regular videos and past live streams. On: every upload

Rows from these sources also carry sourceType (channel, playlist or search), sourceId, sourceTitle and position, and input holds the channel, playlist or query you gave. A video met twice in one run, in videos, in two sources or twice in one playlist, is fetched and charged once, for the first input that reaches it. Inputs count in order, videos first, then channelsAndPlaylists, then searchQueries. Each later copy is a free duplicate row whose duplicateOf names that first input. Listing a channel, a playlist or a search is free, and you pay only for transcripts. A channel or playlist that does not exist comes back as one free not_found row. A channel or playlist link put in videos comes back as a free invalid_input row whose message points to channelsAndPlaylists.

What you get per video

Field	What it holds
status	ok, no_transcript, unavailable, invalid_input, duplicate or error
title, channel, channelId, durationSeconds	Video details from the same request, at no extra cost
publishedAt	Exact publish time in UTC, such as 2009-10-25T06:57:33Z. Empty in the rare case YouTube leaves it out
language	Language of the returned transcript
isAutoGenerated	true when the captions came from YouTube's speech recognition
requestedLanguageFound	false when the video has no captions in your language, so you got the original ones
transcript	The full transcript as one string
wordCount	Words in the transcript
segments	[{ "start": 12.4, "duration": 3.1, "text": "..." }], when timestamps are on

Video details and output formats

Both are off by default and cost nothing extra. They ride in the same row, so a transcript is still one charge.

includeVideoDetails: true adds:

Field	What it holds
viewCount, likeCount	Exact counts at run time. likeCount can be empty when YouTube leaves it out
commentCount	Exact comment count. Empty when comments are turned off, as on NASA's videos
category	YouTube category, such as Education
description	The video description. Emails and phone numbers in it are replaced with [email removed] and [phone removed]
tags	The uploader's tags
thumbnailUrl	The largest thumbnail YouTube returns
chapters	[{ "title", "start", "end", "url" }] from the description's timestamps, by YouTube's rule: the list starts at 0:00 and has at least three rising entries. Each url opens the video at that chapter

outputFormats takes any of these:

Value	Adds	What it holds
text	plainText	The captions as paragraphs, one per line, split at pauses of 2 seconds or more and after about 120 words
llm	llmText	Markdown for an LLM: title, channel, date, URL, duration and caption type on top, a heading per chapter, a [m:ss] stamp per paragraph, and sound cues such as [Music] removed
srt	srt	The SubRip subtitle file as text
vtt	vtt	The WebVTT subtitle file as text
chunks	chunks	RAG chunks: one per YouTube chapter, with long chapters split evenly, or even pieces when the video has no chapters. Each has index, chapter, start, end, wordCount, text and a url that opens the video at that second

chunkSizeWords sets the chunk size, 300 words by default. Chunks are cut between caption lines, so one can run a little over. In the final test the largest of 1,938 chunks held 319 words. 34 of the 87 videos had chapters.

Input

{
  "videos": [
    "https://www.youtube.com/watch?v=dQw4w9WgXcQ",
    "https://youtu.be/jNQXAC9IVRw",
    "https://www.youtube.com/shorts/aqz-KE-bpKQ",
    "M7lc1UVf-VE"
  ],
  "language": "en",
  "includeTimestamps": true
}

The scraper accepts watch, youtu.be, Shorts, embed and live links, plus bare 11-character IDs.

How language works

1. Human-made captions in your language come first.
2. Then YouTube's auto-generated captions in your language.
3. If neither exists, you get the video's original captions, and the row shows requestedLanguageFound: false.

Output example

This is real output for the input above, from a run on 15 Oct 2026. The segments array is cut to two entries here.

{
  "input": "https://youtu.be/jNQXAC9IVRw",
  "videoId": "jNQXAC9IVRw",
  "url": "https://www.youtube.com/watch?v=jNQXAC9IVRw",
  "status": "ok",
  "publishedAt": "2005-04-24T03:31:52Z",
  "title": "Me at the zoo",
  "channel": "jawed",
  "channelId": "UC4QobU6STFB0P71PMvOGN5A",
  "durationSeconds": 19,
  "language": "en",
  "requestedLanguageFound": true,
  "isAutoGenerated": false,
  "transcript": "All right, so here we are, in front of the elephants the cool thing about these guys is that they have really... really really long trunks and that's cool (baaaaaaaaaaahhh!!) and that's pretty much all there is to say",
  "wordCount": 39,
  "segments": [
    { "start": 1.2, "duration": 2.16, "text": "All right, so here we are, in front of the elephants" },
    { "start": 5.318, "duration": 2.656, "text": "the cool thing about these guys is that they have really..." }
  ]
}

Big Buck Bunny in the same input has no captions, so its row comes back as "status": "no_transcript" and costs nothing.

Output example with every option on

This row comes from the final test run on 16 Oct 2026, with searchQueries: ["how transformers work", ...], maxVideosPerSource: 5, includeVideoDetails: true and outputFormats: ["text", "llm", "srt", "vtt", "chunks"]. Long strings are cut and marked with ..., segments, srt, vtt and llmText are left out, and the arrays are cut. The real row has 17 tags, 4 chapters and 4 chunks.

{
  "input": "how transformers work",
  "sourceType": "search",
  "sourceId": "how transformers work",
  "sourceTitle": "how transformers work",
  "position": 3,
  "videoId": "7RtBUEZbKmI",
  "url": "https://www.youtube.com/watch?v=7RtBUEZbKmI",
  "status": "ok",
  "publishedAt": "2020-04-01T10:00:07Z",
  "title": "GCSE Physics - How Transformers Work (2026/27 exams)",
  "channel": "Cognito",
  "channelId": "UCaGEe4KXZrjou9kQx6ezG2w",
  "durationSeconds": 259,
  "language": "en",
  "requestedLanguageFound": true,
  "isAutoGenerated": true,
  "transcript": "before we look at exactly how transformers work let's very quickly recap how they're used in the nat ...",
  "wordCount": 578,
  "viewCount": 607042,
  "likeCount": 9232,
  "commentCount": 393,
  "category": "Education",
  "description": "⭐️ Check out our website https://www.cognito.org/ ⭐️\n\n*** WHAT'S COVER ...",
  "tags": ["transformers", "step-up", "step-down"],
  "thumbnailUrl": "https://i.ytimg.com/vi/7RtBUEZbKmI/sddefault.jpg",
  "chapters": [
    { "title": "Intro & Role in National Grid", "start": 0, "end": 67, "url": "https://www.youtube.com/watch?v=7RtBUEZbKmI&t=0s" },
    { "title": "Transformer Structure", "start": 67, "end": 84, "url": "https://www.youtube.com/watch?v=7RtBUEZbKmI&t=67s" }
  ],
  "plainText": "before we look at exactly how transformers work let's very quickly recap how they're used in the nat ...",
  "chunks": [
    {
      "index": 0,
      "chapter": "Intro & Role in National Grid",
      "start": 0,
      "end": 65.52,
      "wordCount": 145,
      "text": "before we look at exactly how transformers work let's very quickly recap how they're used  ...",
      "url": "https://www.youtube.com/watch?v=7RtBUEZbKmI&t=0s"
    }
  ]
}

The start of the same row's llmText:

# GCSE Physics - How Transformers Work (2026/27 exams)
Channel: Cognito
Published: 2020-04-01T10:00:07Z
URL: https://www.youtube.com/watch?v=7RtBUEZbKmI
Duration: 4:19
Captions: en, auto-generated

## Intro & Role in National Grid [0:00]

[0:04] before we look at exactly how transformers work let's very quickly recap how they're used in the national grid ...

And of its srt:

1
00:00:04,799 --> 00:00:06,319
before we look at exactly how

2
00:00:06,319 --> 00:00:07,759
transformers work

Use it from AI agents

The scraper works as a tool through the Apify MCP server in Claude, Cursor, VS Code and other MCP clients. Agents can pass a list of video URLs, a channel or a search and get back plain text they can summarise, quote or index, or chapter chunks with links they can cite. Each call charges only for the transcripts it returns.

You can also call it through the Apify API or the Python and JavaScript clients, and schedule it or connect it to Make, Zapier and n8n.

Pricing

Prices are per 1,000 transcripts.

Event	No discount	Bronze	Silver	Gold
Transcript returned	$3	$2.70	$2.40	$2.10
Video unavailable, no captions, invalid input or failed fetch	Free	Free	Free	Free
Listing a channel, playlist or search, video details and output formats	Free	Free	Free	Free

Bronze, Silver and Gold are Apify Store discount tiers, and Apify's Pricing tab shows them as Starter, Scale and Business. Apify decides which tier an account is on.

Set a maximum cost per run in the run options. The scraper stops once that limit is reached.

FAQ

Why does a video show no_transcript? The uploader turned captions off and YouTube made no auto-generated ones. That row isn't charged.

Does it work for Shorts and live streams? Yes, if the video has captions. Live streams get captions after they end.

Does it use the official YouTube API? No. It reads the same caption data the YouTube app shows. Only use transcripts in ways that respect the rights of the video owners.

What about videos with no captions at all? Speech-to-text for those is on the roadmap. Open an issue on the Issues tab if you need it.