How to Find a YouTube Video by Its Transcript (Search Any Quote or Phrase)

Why YouTube search can't find a video by what was said

YouTube ranks videos on metadata and engagement: the title, description, tags, and how viewers respond. The spoken words live in the caption track, and the search box doesn't look up a phrase inside it. That is why typing an exact line from a podcast usually returns videos about the topic, not the episode where it was said.

Captions are the way in. Most public videos have them, either uploaded by the creator or auto-generated by YouTube from the audio, and every caption line carries a start time. Search the captions and you get both the video and the second the phrase was spoken.

Start with a web search if the quote is famous

If the line is widely quoted, someone has probably written it down already. Put the exact phrase in quotation marks in Google or Bing, and add the speaker's name or the word youtube. Articles, show notes, and subtitle sites often name the episode.

This fails for anything that was only said out loud: a remark buried in hour three of a stream, a line from a small channel, or a quote you half-remember. For those you need to search transcripts directly, and it helps to know roughly where to look: which channel, which show, or which playlist.

Search one video's transcript for a phrase

When you already suspect the video and need the exact moment, you have two options.

On a computer, YouTube's own panel works: open the video, click ...more under the title, then Show transcript, and press Ctrl+F (Cmd+F on a Mac). Long transcripts load in pieces, so the browser can miss words further down, and the YouTube mobile app has no transcript search at all.

The quicker route on any device is the free YouTube transcript search tool. Paste the link, type a word, and every match appears with its surrounding text and a timestamp that opens the video at that second. It works on multi-hour streams and lectures, and a few videos a day need no account.

Find which video in a channel or playlist said a quote

This is the real version of the question: you know the creator, but not the video. Checking uploads one by one doesn't scale past a handful, so the approach is to list the videos, pull every transcript, and search them together.

Without code, the free channel transcripts and playlist transcript tools list every video so you can open each transcript and search it. That is fine for a short playlist.

For a channel with hundreds of uploads, the TranscriptYT API has a search option that filters each transcript after it is fetched. A video that never mentions your phrase comes back with an empty segments list, and the ones that do return only the matching lines, so you read a short result instead of scrolling through hundreds of full transcripts. It takes a few calls:

• List the uploads with GET /v1/playlist. It takes a channel link, channel ID, @handle, or playlist link and returns video IDs, titles, and durations, newest first, up to 500 per call.
• Send the IDs to POST /v1/transcripts with a search option. Each video comes back with only the caption lines that contain your text (case-insensitive), each with its start time.
• Set mode to captions so a video without captions is skipped instead of being sent to AI transcription, which is billed per minute.
• Turn every match into a link: youtube.com/watch?v=VIDEO_ID&t=SECONDS opens the video at that moment.

Code: search a whole YouTube channel's transcripts

This Python script lists a channel's latest 500 uploads and prints every line that mentions a word, with the video title and a timestamped link. It sends 25 videos per batch so it runs on the Micro plan. Raise the step to your plan's batch size: 10 on Free, 50 on Starter, 200 on Growth, 500 on Scale. Any video that hits the per-minute rate limit comes back as a RATE_LIMITED item to retry. The full parameter list is in the docs.

import os, requests

API = "https://transcript-yt.com/v1"
HEADERS = {"Authorization": f"Bearer {os.environ['TRANSCRIPTYT_KEY']}"}
CHANNEL = "@veritasium"
PHRASE = "entropy"

videos = requests.get(f"{API}/playlist", params={"url": CHANNEL, "limit": 500},
                      headers=HEADERS, timeout=120).json()["videos"]
ids = [v["videoId"] for v in videos]

for i in range(0, len(ids), 25):
    batch = requests.post(f"{API}/transcripts",
                          json={"urls": ids[i:i + 25], "search": PHRASE, "mode": "captions"},
                          headers=HEADERS, timeout=300).json()
    for item in batch["results"]:
        if not item["ok"]:
            continue
        t = item["transcript"]
        for s in t["segments"]:
            link = f"https://www.youtube.com/watch?v={t['videoId']}&t={int(s['start'])}"
            print(f"{t['title']} | {link} | {s['text']}")

How to make a quote search actually match

Most failed searches don't mean the video is missing. Usually, what you typed doesn't match how the captions were written.

• Search the rarest word, not the whole sentence. Captions are split into short cues of a few words, and a phrase that crosses a cue boundary won't match as one string. Search a distinctive word or two, then read the lines around each hit.
• Expect misheard names. Auto-generated captions spell names and jargon the way they sound, so try a shorter fragment or an alternative spelling.
• Leave punctuation out. Auto-generated captions often have little or none, so a query with commas or quotation marks inside it can miss. Type plain words only.
• Search in the spoken language. A Spanish video's captions are in Spanish. If you only know the English wording, translate the transcript first with the API's translate_to option.
• Remember the quote may be paraphrased in your memory. If an exact phrase finds nothing, search one concrete word from it: a number, a name, or an unusual noun.

Searching the same videos again and again

If you search one creator's catalog often, fetch the transcripts once and keep them. Store the full JSON (it has a start time for every segment) and load it into SQLite full-text search, Postgres, or a spreadsheet. Every later search is then instant and free. To search by meaning instead of exact words, embed the segments in a vector index. Building a searchable video library walks through both.

Cost is predictable. Each transcript is 1 credit, even when a filtered search finds no match, and listing a channel is free on paid plans. The Free plan includes 100 credits, Micro is $3 for 1,000, and Scale is $79 for 50,000 ($1.58 per 1,000). Checking a 500-video channel costs 500 credits. Once a channel is stored locally, searching it again costs nothing.