YouTube Transcript Scraper – Captions to Text for RAG & LLMs

Turn any YouTube video's subtitles or captions into clean, joined text, timestamped segments, or ready-to-save SRT/VTT subtitle files — built for feeding video content into RAG pipelines, LLM apps, and summarization workflows. Paste a video URL, a youtu.be link, a Shorts link, or just the 11-character video ID. This actor calls a licensed SERP data provider directly — no proxies, no headless browser, no YouTube rate limits or IP bans.

What data you get

Each result item includes:

• videoId, videoUrl — the video's ID and canonical youtube.com/watch?v= URL
• title — the video's title
• language, translatedTo — the subtitle language returned, and the language it was translated into (if you set translateTo)
• subtitlesCount — how many subtitle lines the video has
• durationSec — video duration inferred from the last subtitle's end time
• text — the full transcript, subtitle lines joined with spaces
• segments — array of {start, end, text} per subtitle line (only when includeTimestamps is on)
• srt, vtt — ready-to-save subtitle file strings (only when outputFormat requests them — see Input)
• status, scrapedAt

Why this scraper

• No bans, ever — this actor never touches youtube.com directly. It calls a licensed SERP data provider's subtitles API, so there's no IP to block and no CAPTCHA to solve.
• Cheap — a flat $0.005 per transcript, and videos without captions cost nothing (see Pricing).
• Fast by default — mode: "auto" picks fast (live, a few seconds per video) for runs of 50 videos or fewer, and standard (queued, cheaper) for bigger batches — see Modes.
• Translation built in — set translateTo to get subtitles translated into another language, upstream, without a separate translation step.
• SRT/VTT out of the box — set outputFormat to get a ready-to-save subtitle file alongside the plain text, no separate conversion step.
• Batch-friendly — up to 1,000 videos per run.
• No silent gaps — every video that doesn't produce a transcript (no captions, wrong language, a stuck task) is listed in the FAILED_VIDEOS record with why, so a batch of 200 URLs never comes back as an unexplained 183 rows.

Use cases

• RAG / summarization pipelines — feed text straight into a vector store or an LLM prompt for question-answering or summarizing video content at scale.
• Content repurposing — turn a video transcript into a blog post, show notes, or social copy without manually re-watching and typing it out.
• Subtitles for clips — set outputFormat: "srt" or "vtt" to get a subtitle file ready to attach to a re-cut clip or upload alongside a repost.

Pricing

This actor uses Apify's pay-per-event pricing, flat across all plans. You are charged for:

Event	What it means	Price
apify-actor-start	Apify's platform-billed "run started" event, charged once per run automatically — no code call needed	$0.00005
apify-default-dataset-item ("Transcript" in Console)	Charged once per video that has a transcript, written to the output dataset	$0.005

Videos without subtitles are never charged — they don't produce a dataset row at all, so you only pay for transcripts you actually get.

Worked example: transcribing 1,000 videos that all have captions costs 1,000 × $0.005 + $0.00005 ≈ $5.00. If 100 of those 1,000 have no subtitles, you pay for the 900 that succeeded: 900 × $0.005 + $0.00005 ≈ $4.50.

Apify's free plan includes monthly platform credit that, on its own, covers roughly the first ~1,000 transcripts with this actor before you'd need to add a payment method.

Modes

Mode	Upstream call	Wall-clock	Cap	When to use
auto (default)	fast for ≤50 videos, standard above that	See below	1,000 videos/run	Picks the right mode for the batch size automatically — no need to think about it for typical runs.
standard	Queued (task_post + polling)	~5 minutes per batch of up to 100 videos	1,000 videos/run	Cheapest upstream cost per video; the right choice once auto would pick it anyway (>50 videos).
fast	Live (synchronous)	A few seconds per video	50 videos/run	Small batches where you need the transcript back immediately (interactive use, a quick one-off check). Priced higher per video upstream to fund the instant response, capped at 50/run so a bad input doesn't run up a live-mode bill.

Expected total wall-clock (with auto, or an explicit mode): 10 videos ≈ a few seconds (fast); 100 videos ≈ ~5 minutes (standard, one batch); 1,000 videos ≈ ~50 minutes (standard, 10 batches processed one after another).

Input

Minimal example:

{
  "videoUrls": [
    "https://www.youtube.com/watch?v=dQw4w9WgXcQ",
    "https://youtu.be/jNQXAC9IVRw"
  ],
  "language": "en"
}

Field	Type	Default	Description
videoUrls	array of strings	(see prefill)	YouTube video URLs — watch, youtu.be, Shorts, or embed links — or bare 11-character video IDs. Max 1,000 per run, 50 when mode is explicitly set to "fast".
language	string	"en"	Language code of the original subtitles/captions to fetch.
translateTo	string	—	Optional language code to translate the subtitles into. Leave empty to keep the original language.
includeTimestamps	boolean	true	Adds the segments array (per-line start/end/text) to each output item, alongside the joined text.
mode	string	"auto"	"auto", "standard", or "fast" — see Modes.
outputFormat	string	"text"	"text", "srt", "vtt", or "all" — adds an srt and/or vtt subtitle-file string to each output item.

Output

A real dataset item, captured from an actual run against "Me at the zoo" (jNQXAC9IVRw, YouTube's first-ever upload) in September 2026:

{
  "videoId": "jNQXAC9IVRw",
  "videoUrl": "https://www.youtube.com/watch?v=jNQXAC9IVRw",
  "title": "Me at the zoo",
  "language": "en",
  "translatedTo": null,
  "subtitlesCount": 6,
  "durationSec": 18.88,
  "text": "All right, so here we are, in front of the\nelephants the cool thing about these guys is that they\nhave really... really really long trunks and that's cool (baaaaaaaaaaahhh!!) and that's pretty much all there is to\nsay",
  "segments": [
    { "start": 1.2, "end": 3.36, "text": "All right, so here we are, in front of the\nelephants" },
    { "start": 5.318, "end": 7.97, "text": "the cool thing about these guys is that they\nhave really..." },
    { "start": 7.974, "end": 12.62, "text": "really really long trunks" },
    { "start": 12.616, "end": 14.37, "text": "and that's cool" },
    { "start": 14.421, "end": 15.73, "text": "(baaaaaaaaaaahhh!!)" },
    { "start": 16.881, "end": 18.88, "text": "and that's pretty much all there is to\nsay" }
  ],
  "status": "ok",
  "scrapedAt": "2026-09-08T05:04:41Z"
}

Field	Type	Description
videoId	string	11-character YouTube video ID
videoUrl	string	Canonical https://www.youtube.com/watch?v=<id> URL
title	string | null	Video title
language	string | null	Subtitle language actually returned (falls back to your requested language if the upstream response doesn't echo one)
translatedTo	string | null	Language the subtitles were translated into, null unless translateTo was set
subtitlesCount	integer	Number of subtitle lines in the video
durationSec	float	Video duration, inferred from the last subtitle's end time
text	string	Full transcript: every subtitle line's text, joined with spaces, empty lines skipped
segments	array of objects | null	null unless includeTimestamps is true, in which case a {start, end, text} object per subtitle line
srt	string | null	null unless outputFormat is "srt" or "all" — the full transcript as an SRT subtitle file
vtt	string | null	null unless outputFormat is "vtt" or "all" — the full transcript as a WebVTT subtitle file
status	string	Always "ok" — videos without subtitles never produce a row (see Limits)
scrapedAt	string	ISO 8601 UTC timestamp

Limits

• Videos with no subtitles/captions in the requested language, or that don't exist, produce no dataset row and are not charged. They're never silently dropped, though: every one is listed in the FAILED_VIDEOS key-value store record (videoId, videoUrl, status — no_subtitles, unsupported_language, timed_out, or error — and a message), and counted in STATS under videosWithoutSubtitles.
• standard mode (and auto above 50 videos): up to 1,000 videos per run, processed in batches of up to 100, polled for up to 15 minutes per batch. A batch still stuck after 15 minutes is skipped (counted in STATS as timedOut, and listed in FAILED_VIDEOS with that status) — this is rare and upstream-side.
• fast mode, whether picked by auto or set explicitly: capped at 50 videos per run (see Modes for why).
• This actor does not distinguish auto-generated captions from manually uploaded ones — the upstream data source returns whichever subtitle track YouTube serves for the requested language, without labeling its origin.
• Private, age-restricted, or region-locked videos may return no subtitles even if they otherwise have captions.

FAQ

Is this legal? The actor retrieves publicly available subtitle/caption data through a licensed SERP data provider, the same category of data source used by SEO and content-research tools. You are responsible for how you use the extracted text under your local regulations and YouTube's own terms of service.

How is this different from other YouTube transcript tools on Apify? Most transcript scrapers drive a browser or an unofficial YouTube endpoint directly, which means occasional blocks and rate limits. This actor calls a licensed SERP data API instead: no proxies, no browser, no bans — the risk of getting blocked is the data provider's problem, not yours.

What languages are supported? Any language YouTube has subtitles/captions for. Set language to the subtitle language code you want (e.g. "en", "es", "pt"), and optionally translateTo to get a machine translation into a different language.

Do I get charged for videos without subtitles? No. A video with no subtitles in the requested language produces no dataset row at all, so apify-default-dataset-item is never charged for it. Videos without captions are listed in the FAILED_VIDEOS record and never billed — check it (alongside STATS.videosWithoutSubtitles) to see exactly which videos didn't come back and why.

Can auto-generated and manually uploaded captions be told apart? No — the underlying data source returns whichever subtitle track YouTube serves for the language you request, without labeling whether it's auto-generated or manually authored. If that distinction matters for your use case, treat all transcripts as potentially auto-generated (which is the overwhelming majority of YouTube's caption tracks in practice).