Give it a YouTube video id or URL and get the transcript back as rows - one row per caption segment, each with its text, its start time and how long it lasted, plus the whole transcript as a single field. Eight columns. No YouTube account, no residential proxy, no parser to keep alive.
The input is the video. Paste what you already have - the bare id, a watch link, or a short youtu.be link - and the job fetches the caption track behind it.
Every row is tagged with the query it came from and the video id it belongs to - so a run across a playlist's worth of videos still reconciles back to your input list.
All three go in the same box, one per line. A watch URL carrying extra parameters - a playlist, a radio flag - still resolves to the right video, and the id it resolved to comes back in its own column so you can join on it.
Each segment carries the moment it starts and how long it runs, both as decimal seconds. That is what turns a wall of speech into something you can jump to - build a deep link, cut a clip, or line a quote up against the frame it came from.
You get both shapes in one file: a row per segment for anything timed, and the complete transcript as a single field for anything that just needs the text. The data dictionary explains exactly where that full-text field sits, because it is not on every row.
Each row is one caption segment: what was said, when it starts and how long it lasts, tagged with the video it belongs to and the query you submitted.
The column list below is the header row of real run exports, corroborated by the column array the run UI renders; the two agree exactly. Two of the eight behave differently from the rest - they appear once per video rather than once per row - and that is the thing most likely to surprise you, so it is spelled out in the table and again in the note beneath it.
Taken from the header row of real runs. The notes describe what each column holds and how it repeats - not how many rows you will get, because that is a property of the video's caption track, not of the scraper.
en is the only value we have seen - the page does not guess at what else the column can hold.Three things to know before you write code against this. A row is a segment, not a video - one video fills as many rows as its caption track has segments, and query, video_id and language repeat down all of them. segments and transcript break that pattern: they are written on the first row of each video only and are empty on the rest, so group by video_id and take the first row rather than expecting them everywhere. And start and duration come back as real numbers while everything else is text - including segments, which is an integer where it is filled and an empty string where it is not, so a strict parser needs to tolerate both.
This scraper has no limit and no sort order - the transcript is what it is. The only thing you decide is how you hand over the videos.
A few examples of what teams do with spoken words once they are in a spreadsheet.
Pull the transcripts for a channel's videos and grep the text column. Finding every mention of a product, a competitor or a claim takes a filter rather than an afternoon of scrubbing through playback.
A transcript is the first draft of a blog post, a show-notes page or a set of subtitles. The full-text field gives you the whole thing in one cell, and the timed rows give you the timestamps to link back with.
Timed segments chunk cleanly for retrieval, and every chunk already knows which video it came from and where in it. That is the difference between an answer and an answer with a link to the second it was said.
No subscription, no minimum, no recurring bill. Your first 500 rows are on us - after that, pay-as-you-go at the same flat rate as every other scraper here.
Every new account, one-time. No credit card required. File upload and every export format included. Enough to see the shape of a transcript before you plan around it.
The same flat rate as every other scraper on the platform. The pre-flight estimator shows the row count and credit cost before a run starts - no surprise bills, no compute units to translate.
Volume pricing, dedicated workers and an SLA for continuous monitoring or very large historical pulls. Tell us your numbers and we will quote.
LIVESCRAPER10 at checkout.The step before this one. Find the videos worth transcribing by the query people actually search, then feed the ids straight in here.
Where the same topic gets covered in writing. Reading the SERP next to the transcripts shows what a subject sounds like spoken versus how it is written down.
Short answer: captions are published alongside the video - and what you may do with the words afterwards is a copyright question, not a scraping one.
A caption track is served to anyone who opens the video, signed in or not; that is how the subtitles button works. Reading the same public track for research is long-established practice, and nothing here touches a login or a paywall. A video without captions has no transcript to read, which is why the scraper needs one and says so plainly.
The part worth planning around is not access but reuse. A transcript is the creator's words, and the video is their copyrighted work - so quoting, indexing, searching and analysing sit on very different footing from republishing a talk as your own article. If you are training on it or reproducing it at length, that is a licensing question, and it is yours to answer before you scale.
YouTube's terms restrict automated access, so this is a terms question as well as a legal one - if you have a contractual relationship with the platform, check it. We run no third-party trackers on the data layer, and your exports auto-delete after 30 days.
The questions we hear most. Anything else? Talk to us - humans, not bots, write the answers.
500 one-time free rows on every new account - no expiry. After that it is $0.002 per row, pay-as-you-go - no card on file until you say so.
Livescraper's YouTube Transcripts Scraper turns a list of videos into text. You submit YouTube video ids or URLs - typed one per line, or uploaded as a CSV, XLSX, TXT or Parquet file - and download the caption track as a clean CSV, Excel or JSON file. Bare ids, watch links and youtu.be short links all work, and a watch URL carrying playlist parameters still resolves to the right video.
A row is one caption segment. Each carries what was said, the second it starts and how long it lasts, alongside the query you submitted, the video id it resolved to and the language of the track. Two more columns work per video rather than per row: segments, which is how many rows that video contributes, and transcript, which holds the whole thing as one block of text. Both sit on the first row of each video, which is the detail worth knowing before you write a parser.
Researchers pull a channel's back catalogue and search the text for a phrase instead of scrubbing through playback. Content teams turn talks into posts, show notes and subtitles. Teams building retrieval systems use the timed segments as chunks that already know which video they came from and where in it - the difference between an answer and an answer you can link to the second it was said.
Two things are worth knowing before you start. A video with captions disabled has no transcript to return, so captions are a requirement rather than a nice-to-have. And because a row is a segment, a long video costs more than a short one at the same flat $0.002 per row. Start free: your first 500 rows cost nothing and need no credit card.