YouTube Transcript Bulk Scraper

YouTube Transcript Bulk Scraper: full text and timed cues from a list, a channel or a playlist

Paste video links, a channel or a playlist and get one row per video with the whole transcript as plain text, the per-cue timings, the language, and the video's own details next to it. Channels and playlists are expanded into their videos for you, and that expansion costs nothing.

It reads the caption tracks YouTube already holds. There is no speech recognition here, so a video that was uploaded without captions has nothing to return and comes back as an uncharged diagnostic row instead. You pay for transcripts, not for the videos a channel turned out to have.

🔍 What YouTube Transcript Bulk Scraper does

Mix whatever you have in one list. Watch links, Shorts links, youtu.be links, embed links, bare video ids, a channel URL, an @handle, a playlist link. Channels are read newest first, playlists in their own order, and a video that shows up twice across two sources is fetched once.

You get the transcript two ways on the same row. text is the whole thing as one readable string, already joined up, which is what you want if it is going into a model or a search index. segments keeps every cue with its start, duration and end, which is what you want for jumping to a moment in the video or building captions.

Language is a preference list, not a filter. Put en, es, pt-BR in and the first one the video actually has is used; pt matches pt-BR. A human-written track wins over the machine one unless you say otherwise, because it has punctuation and speaker turns. Whichever track was used is named on the row, and availableLanguages lists everything the video offered.

📥 What you give it

{
  "videoUrls": [
    "https://www.youtube.com/watch?v=aircAruvnKk",
    "https://www.youtube.com/@3blue1brown"
  ],
  "maxItems": 10,
  "languages": ["en"]
}

Run it with the input empty and you get one labelled sample row, uncharged, so you can look at the shape before you queue a real list.

📤 What you get back

A real row from a real run. text and segments are cut short here; on the real row text runs to 822 characters and segments has all 13 cues:

{
  "ok": true,
  "charged": true,
  "recordType": "transcript",
  "videoId": "6XPlmCDNLNc",
  "url": "https://www.youtube.com/watch?v=6XPlmCDNLNc",
  "title": "The 64 sugar cubes puzzle",
  "channelName": "3Blue1Brown",
  "channelId": "UCYO_jab_esuFRV4b17AJtAw",
  "channelUrl": "https://www.youtube.com/channel/UCYO_jab_esuFRV4b17AJtAw",
  "durationSeconds": 55,
  "viewCount": 322213,
  "thumbnailUrl": "https://i.ytimg.com/vi/6XPlmCDNLNc/sddefault.jpg?sqp=...",
  "language": "en",
  "languageName": "English",
  "isAutoGenerated": false,
  "segmentCount": 13,
  "wordCount": 160,
  "characterCount": 822,
  "text": "It's time for a new puzzle of the month. This one comes from a problem on the 2024 British Math Olympiad. ...",
  "segments": [
    { "start": 0, "duration": 1.7, "end": 1.7, "text": "It's time for a new puzzle of the month." },
    { "start": 2.08, "duration": 3.56, "end": 5.64, "text": "This one comes from a problem on the 2024 British Math Olympiad." }
  ],
  "availableLanguages": [
    { "languageCode": "en", "languageName": "English", "isAutoGenerated": false },
    { "languageCode": "en", "languageName": "English (auto-generated)", "isAutoGenerated": true }
  ],
  "sourceType": "channel",
  "sourceUrl": "https://www.youtube.com/channel/UCYO_jab_esuFRV4b17AJtAw",
  "playlistId": "UUYO_jab_esuFRV4b17AJtAw",
  "scrapedAt": "2026-09-21T01:27:33.286Z"
}

🧾 Reading the output

Three kinds of row land in your dataset, and recordType separates them.

Filter on recordType == "transcript" and you have your transcripts. The charged flag is written onto the row as it is built, so treat it as the marker that tells a real row from a sample or a diagnostic, not as a receipt for what was billed.

▶️ How to run it

1. Open YouTube Transcript Bulk Scraper and click Try for free. 2. Paste links into Videos, channels or playlists, one per line. 3. Set Maximum transcripts low for the first run, since that is what you pay for. 4. Put your language codes into Preferred languages in the order you want them tried. 5. Click Start, then download the dataset as JSON, CSV or Excel, or pull it from the Apify API.

💰 How much does it cost?

$0.40 per 1,000 transcripts. Flat on every Apify plan, no volume tiers.

Only a delivered transcript counts. Expanding a channel or a playlist into its videos adds nothing, however many it finds. A video with no captions, a rejected entry, a video that would not open and the sample row are all uncharged rows.

A video that appears in two of your sources is fetched once and counted once.

💡 What people use it for

• Feeding a model. Pull a whole channel, keep text, and you have a corpus that needs no audio

handling at all.

• Making a video searchable. Keep segments and a hit gives you the second to jump to.
• Repurposing. Long-form episodes become newsletter drafts, show notes and clip candidates faster

from text than from video.

• Research across creators. Run ten channels and count who talks about what, with dates and view

counts already on the row.

🚧 What it does not do

• Captions only. There is no speech recognition. If the uploader published no caption track, no

amount of retrying will produce text.

• No translation. You get the tracks the video already carries. Asking for fr on a video with

only English gets you the English track, not a translation.

• Channels come from the uploads list, newest first. That misses unlisted videos and sometimes

misses Shorts, and there is no date filter, so reach older videos with maxVideosPerSource rather than a date range.

• A watch link carrying list= is treated as one video, not as the playlist. Paste the

playlist link itself if you want the playlist.

• The 5,000 ceiling is the schema's, not a measured promise. One run was measured carrying

about 830 transcripts before memory became the limit. Long videos make big rows: a 3 hour 13 minute podcast came back as a single 160 KB row.

• Only the first 25 rejected entries get a BAD_INPUT row. Beyond that they are dropped

quietly, so check your list if the count looks short.

• Private and age-restricted videos are reported as a failure to open rather than as a distinct

reason of their own.

🧭 Which YouTube actor do you need?

❓ Questions people ask

Do I need a YouTube API key? No. No key, no OAuth, no quota to manage.

What if the video has no captions? You get a NO_RESULTS row naming that video and the run moves on. That row is not charged.

Auto-generated or written captions, which do I get? The written one when the video has it, because it is punctuated. Tick Prefer auto-generated captions to flip that.

Can I get the timings? Yes, that is segments, and it is on by default. Turn it off when you want smaller rows and only the plain text.

Is scraping YouTube legal? This reads publicly available caption tracks. Transcripts are somebody's work, so what you may do with them afterwards is a copyright question rather than a scraping one. Apify's write-up on the legality of web scraping is a reasonable starting point, and we are not lawyers.

Can I run it on a schedule? Yes. Point it at a channel with a small maxVideosPerSource and each run picks up what is new at the top.

🆘 If something breaks

Open the Issues tab on the actor page. Include the links you used and the run id. The diagnostic rows in your dataset usually name the reason already.