How do I download transcripts for hundreds of YouTube videos at once?

A media researcher has a spreadsheet of 800 lecture videos and wants their transcripts in one file by Friday. With youtube-transcript-api, the run dies after a few dozen videos with 429 Too Many Requests, and the retry logic people add tends to hammer the same IP again. A batch this size needs three things: a clean list of IDs, a client that backs off politely, and a way to resume.

Step 1: the list of IDs

• You already have links. Take the v= value, or the part after youtu.be/ or /shorts/. The transcript endpoint wants the 11-character ID.
• You have a topic. YouTube video search returns about 20 videos a page with views and channel, so 100 candidates cost $0.02.
• You have a channel. The official Data API's playlistItems.list on the channel's uploads playlist costs 1 quota unit per page of up to 50 videos, which is the cheapest way to list a whole channel.

Step 2: a resumable runner

import json, os, time, requests
from concurrent.futures import ThreadPoolExecutor

URL = "https://api.okdata.online/v1/youtube/videos/transcript/get"
KEY = os.environ["OKDATA_API_KEY"]
OUT = "transcripts.jsonl"

def fetch(vid):
    while True:
        r = requests.get(URL, params={"id": vid}, timeout=90, headers={
            "Authorization": f"Bearer {KEY}", "Idempotency-Key": f"transcript-{vid}"})
        body = r.json()
        if body["code"] == "rate_limited":
            time.sleep(body.get("retry_after_seconds") or 2)
            continue
        return vid, body

ids = [line.strip() for line in open("video_ids.txt") if line.strip()]
done = {json.loads(l)["video_id"] for l in open(OUT)} if os.path.exists(OUT) else set()

with ThreadPoolExecutor(max_workers=4) as pool, open(OUT, "a", encoding="utf-8") as out:
    for vid, body in pool.map(fetch, [i for i in ids if i not in done]):
        if body["code"] == "ok":
            out.write(json.dumps(body["data"]["item"], ensure_ascii=False) + "\n")
        else:
            print(vid, body["code"])  # not_found = no captions, not charged

Run it again after any crash; finished videos are skipped. The Idempotency-Key means a retried call that already succeeded within 24 hours is replayed, not billed twice.

What the run costs

Batch	Cost at most
100 videos	$0.40
800 videos	$3.20
1,000 videos	$4.00

Every captionless video and every error lowers the bill, since neither is charged. New keys start with 0.10 of credit, enough to test the script on a slice first.

Limits

• Four workers is a sensible start. Your key's burst size is in the X-RateLimit-Limit header; rate_limited answers are free.
• No speech-to-text: videos without captions need Whisper on audio you may use.
• One language per call. See lang in the response.

When to do it differently

If the batch is small and occasional, youtube-transcript-api from a home connection with a pause between calls usually gets through. For your own channel, the official API downloads your caption files. For the whole workflow, from search to summary, see the video transcript solution page.