YouTube Transcript Scraper – Timestamped JSON for RAG

A YouTube transcript scraper built for RAG and LLM pipelines: it turns video, playlist or channel URLs into one flat JSON record per video with the full transcript, cue-level timestamped segments, embedding-sized text chunks with deep-link timestamp URLs, chapters and normalized video metadata. From $0.56 per 1,000 transcripts.

What it does

The Actor resolves each input (video URL, playlist URL, channel URL or bare video id) to its video ids, then reads the same public watch-page payload and caption/timedtext endpoints a logged-out browser uses when you open YouTube's own transcript panel — no login, no API key, no paid subscription. For every video it returns the caption track (human-authored when available, auto-generated as a fallback) as a normalized transcript: full text, per-cue start/end timestamps, and merged RAG chunks sized to a configurable character budget with stable chunk ids and &t=-timestamp deep links for citation. Videos with no captions are returned as a successful, unbilled item with transcriptStatus="no_captions" instead of failing the run, so coverage gaps are explicit rather than hidden.

Who it's for

• RAG/LLM developers who need video transcripts pre-chunked and timestamped for a vector database, with no custom parsing or chunking code to write.
• AI agent and knowledge-base builders who want citation-grade deep links back to the exact second a chunk came from.
• Content, SEO and research teams pulling spoken text and chapter titles from competitor or reference videos for topic analysis.
• Pipeline operators who need explicit per-video status (ok / no captions / unavailable) instead of a run that silently drops failures.

Use cases

• Bulk-extract transcripts for a channel or playlist as embedding-sized chunks for a vector database.
• Build a citeable video knowledge base: every chunk carries videoId, title, channelName and a deep-link timestamp URL.
• Feed timestamped segments into summarization or clip-finding pipelines that map generated text back to a video position.
• Pull spoken text plus chapter titles from competitor videos for content/SEO research.
• Detect which videos in a list have no captions, so coverage gaps are visible instead of silent failures.

Features

• RAG chunks, not just raw captions — chunks[] merges cues into blocks sized by chunkMaxChars, with configurable overlap, stable chunkIds and a deepLinkUrl per chunk.
• Chapter-aware chunking (respectChapters) — chunks never span two chapters when chapters are detected, and each chunk is tagged with its chapterTitle.
• Cue-level segments[] with numeric start/end/duration seconds, kept separate from chunks so you can disable one to keep items small.
• Explicit coverage status — every item has transcriptStatus (ok, no_captions, language_unavailable, video_unavailable, error); videos without captions are not billed.
• Playlist and channel expansion — pass a playlist or channel URL and it expands into its videos (capped by maxResults).
• Language control — preferred language priority list, with strictLanguage to fail explicitly instead of silently falling back.
• Extra renderings on request — SRT, WebVTT and chapter-headed Markdown alongside the JSON (textFormats).
• Normalized video metadata in the same record — title, channel, publish date, duration, view count, description, keywords — droppable in one flag (includeVideoMetadata) if you don't need it.
• Pay only for videos actually returning a transcript — no charge for no_captions or video_unavailable items.

Input

Field	Type	Required	Description
videoUrls	array of strings	No¹	Video, playlist or channel URLs. Playlist/channel URLs expand into their videos (up to maxResults).
videoIds	array of strings	No¹	Bare 11-character video ids, as an alternative to full URLs.
languages	array of strings	No	Preferred caption language codes in priority order. Default ["en"].
strictLanguage	boolean	No	If true, videos without a track in the requested languages get transcriptStatus="language_unavailable" instead of falling back. Default false.
allowAutoGenerated	boolean	No	Allow YouTube's auto-generated (ASR) captions when no human-authored track exists. Default true.
includeSegments	boolean	No	Include cue-level segments[]. Default true.
includeChunks	boolean	No	Include RAG chunks[]. Default true.
chunkMaxChars	integer	No	Target max characters per chunk (200–8000). Default 1200.
chunkOverlapChars	integer	No	Overlap characters carried between chunks (0–2000). Default 100.
respectChapters	boolean	No	Never let a chunk span two chapters. Default true.
includeChapters	boolean	No	Parse chapter markers from the description into chapters[]. Default true.
includeVideoMetadata	boolean	No	Include title, channel, publish date, duration, view count, description, keywords. Default true.
textFormats	array (plain|srt|vtt|markdown)	No	Extra text renderings to attach. Default ["plain"].
maxResults	integer	No	Safety cap on videos processed per run (0 = no limit). Default 25.
maxConcurrency	integer	No	Max parallel requests (1–50). Lower it if YouTube starts rate-limiting. Default 10.
proxyConfiguration	object	No	Apify Proxy settings. Defaults to the Residential group — YouTube blocks the shared datacenter pool with a "Sign in to confirm you're not a bot" challenge.

¹ At least one of videoUrls or videoIds must be supplied.

Output

Each dataset row is one video. Key fields:

Field	Type	Description
videoId / videoUrl	string	11-character id and canonical watch URL.
title / channelName / channelId / channelUrl	string | null	Public video/channel identity.
publishedAt	string | null	ISO 8601 publish date.
durationSeconds / viewCount	number | null	Public video stats.
description / keywords	string/array | null	Public description (source of chapters) and tags.
sourceInput	string	The input URL/id this item came from (a playlist/channel URL for expanded videos).
transcriptStatus	string	ok, no_captions, language_unavailable, video_unavailable or error — always present.
transcriptAvailable	boolean	true when transcriptStatus is ok.
language / isAutoGenerated / availableLanguages	—	Caption track actually used, whether it's ASR, and all tracks offered.
transcriptText	string | null	Full transcript as one normalized string (when textFormats includes plain).
segments	array | null	Cue-level [{index, start, end, duration, text}].
chunks	array | null	RAG chunks [{chunkId, index, text, start, end, charCount, wordCount, chapterTitle, deepLinkUrl}].
chapters	array | null	Parsed [{index, title, start, end}]; null when the video defines none.
srt / vtt / markdown	string | null	Extra renderings (only when requested in textFormats).
wordCount / charCount / chunkCount	number | null	Full-transcript word/char count and number of chunks produced.
error	string | null	Reason when transcriptStatus is error or video_unavailable.

Sample record (trimmed from an actual production run — full segments/chunks arrays and transcriptText normally run much longer; shown here as 2 segments and 1 chunk for readability):

{
  "videoId": "arj7oStGLkU",
  "videoUrl": "https://www.youtube.com/watch?v=arj7oStGLkU",
  "sourceInput": "https://www.youtube.com/watch?v=arj7oStGLkU",
  "scrapedAt": "2026-09-30T15:51:07.117Z",
  "title": "Inside the Mind of a Master Procrastinator | Tim Urban | TED",
  "channelName": "TED",
  "channelId": "UCAuUUnT6oDeKwE6v1NGQxug",
  "channelUrl": "https://www.youtube.com/channel/UCAuUUnT6oDeKwE6v1NGQxug",
  "publishedAt": "2016-04-06T09:59:35-07:00",
  "durationSeconds": 844,
  "viewCount": 61963947,
  "description": "Tim Urban knows that procrastination doesn't make sense, but he's never been able to shake his habit of waiting until the last minute to get things done...",
  "keywords": ["TED Talk", "Tim Urban", "procrastination", "productivity"],
  "transcriptStatus": "ok",
  "transcriptAvailable": true,
  "availableLanguages": ["en", "es", "fr", "de", "ja", "pt-BR"],
  "error": null,
  "language": "en",
  "isAutoGenerated": false,
  "transcriptText": "So in college, I was a government major, which means I had to write a lot of papers...",
  "segments": [
    { "index": 0, "start": 12.645, "end": 14.015, "duration": 1.37, "text": "So in college," },
    { "index": 1, "start": 15.349, "end": 16.913, "duration": 1.564, "text": "I was a government major," }
  ],
  "chunks": [
    {
      "chunkId": "arj7oStGLkU-0",
      "index": 0,
      "text": "So in college, I was a government major, which means I had to write a lot of papers...",
      "start": 12.645,
      "end": 86.235,
      "charCount": 1183,
      "wordCount": 236,
      "chapterTitle": null,
      "deepLinkUrl": "https://www.youtube.com/watch?v=arj7oStGLkU&t=12s"
    }
  ],
  "chapters": null,
  "srt": null,
  "vtt": null,
  "markdown": null,
  "wordCount": 2277,
  "charCount": 12671,
  "chunkCount": 12
}

Legal note: transcript text is third-party copyrighted content owned by the video's creator. This Actor is a data feed for your own analysis, indexing and internal LLM use — you are responsible for how you store, display or republish any transcript text you retain.

How to use it via the API

curl

curl "https://api.apify.com/v2/acts/YOUR_USERNAME~youtube-transcript-rag-feed/run-sync-get-dataset-items?token=YOUR_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "videoUrls": ["https://www.youtube.com/watch?v=arj7oStGLkU"],
    "languages": ["en"],
    "chunkMaxChars": 1200,
    "maxResults": 3
  }'

JavaScript (apify-client)

import { ApifyClient } from "apify-client";

const client = new ApifyClient({ token: "YOUR_API_TOKEN" });

const run = await client.actor("YOUR_USERNAME/youtube-transcript-rag-feed").call({
  videoUrls: ["https://www.youtube.com/watch?v=arj7oStGLkU"],
  languages: ["en"],
  chunkMaxChars: 1200,
  maxResults: 3,
});

const { items } = await client.dataset(run.defaultDatasetId).listItems();
console.log(items);

Python (apify-client)

from apify_client import ApifyClient

client = ApifyClient("YOUR_API_TOKEN")

run = client.actor("YOUR_USERNAME/youtube-transcript-rag-feed").call(run_input={
    "videoUrls": ["https://www.youtube.com/watch?v=arj7oStGLkU"],
    "languages": ["en"],
    "chunkMaxChars": 1200,
    "strictLanguage": False,
    "maxResults": 3,
})

for item in client.dataset(run["defaultDatasetId"]).iterate_items():
    print(item)

Pricing

Pay-per-event: $0.56 per 1,000 transcripts on the GOLD tier (event = one video successfully returning a transcript, transcriptStatus="ok"). That is 20% below the $0.70/1,000 charged by the mid-market transcript Actors with comparable reliability, and well below the $7/1,000 of the largest one. It is not the cheapest in the niche: at least one Actor charges $0.01/1,000 for a plain caption dump — if price is your only criterion, use that one; this Actor is priced for the chunked, chapter-aware, status-explicit output described above.

Tier	Price per 1,000 transcripts
FREE	$0.84
BRONZE	$0.73
SILVER	$0.64
GOLD / PLATINUM / DIAMOND	$0.56

Videos with no captions or that are unavailable are returned for transparency but never billed.

Platform usage you also pay: Apify charges you separately for platform resources this Actor uses while it runs, mainly the Residential proxy traffic needed because YouTube blocks the shared datacenter pool. Based on measured traffic, that adds roughly $11 per 1,000 videos processed (≈1.37 GB/1,000 at Apify's ~$8/GB residential rate) on top of the event price above. To reduce this: keep includeSegments/includeChunks limited to what you need, lower maxConcurrency if you don't need speed, and consider switching proxyConfiguration to the datacenter group for videos where YouTube's bot check isn't a problem for you (at the risk of more video_unavailable/blocked results).

Troubleshooting

• transcriptStatus="no_captions" for a video you know has captions: check availableLanguages — the video may only offer tracks outside your languages list; widen the list or set strictLanguage=false to fall back to the default track.
• transcriptStatus="language_unavailable": you set strictLanguage=true and none of your requested languages matched; either add the language shown in availableLanguages or drop strictLanguage.
• transcriptStatus="video_unavailable": the video is private, deleted, age-restricted or region-blocked; the Actor never attempts a login or age-restriction bypass, so this is expected, not a bug.
• Run returns 0 items: check the run log for input validation errors — malformed or non-YouTube URLs are rejected before any request is made, with a clear message.
• "Sign in to confirm you're not a bot" / unexpectedly high video_unavailable/error rate: this happens when the Residential proxy pool serves an already-flagged exit IP; retry the run, and avoid manually switching to the datacenter proxy group unless you accept a higher block rate.
• Fewer items than expected from a playlist/channel URL: expansion is capped by maxResults; raise it (or set it to 0 for no limit) if you need the whole list.
• Costs higher than expected: most of the per-run cost is Residential proxy traffic, not the event price — see the Pricing section above.

FAQ

Does this require a YouTube account, cookies or API key? No. It only reads the public watch page and YouTube's own public caption endpoints, exactly what a logged-out browser loads when you open the transcript panel.

Can I get transcripts for a whole playlist or channel? Yes — pass the playlist or channel URL in videoUrls; it expands into individual videos up to maxResults.

Am I billed for videos with no captions? No. Only items with transcriptStatus="ok" are billed. no_captions, language_unavailable, video_unavailable and error items are returned for visibility but free.

Can I republish the transcript text? Transcript text is the original creator's copyrighted content. This Actor delivers it to you as a data feed for your own use (indexing, analysis, internal LLM ingestion); you are responsible for how you use or display it further.

Why are chunks and segments both provided? segments are the raw, often short caption cues; chunks merge them into embedding-sized blocks with overlap and stable ids for direct use in a vector database. Disable either with includeSegments/includeChunks if you only need one.

Is this data allowed to be collected? This Actor reads only publicly visible, unauthenticated pages and endpoints, paces requests and never bypasses login, paywalls or access controls. YouTube's Terms of Service restrict automated access without permission — this is a known legal/business risk that is documented, not hidden; you are responsible for your own use of the collected data.

Keywords

youtube transcript scraper, youtube transcript api, youtube transcript json, youtube captions scraper, youtube subtitles scraper, youtube transcript with timestamps, youtube rag data, youtube transcript for llm, youtube playlist transcript, youtube channel transcript, bulk youtube transcripts, video transcript chunks embeddings, youtube chapters extractor, youtube srt vtt export, youtube video metadata scraper

Hashtags

#youtube #transcript #captions #subtitles #rag #llm #ai #embeddings #vectordatabase #dataextraction #apify #developertools