YouTube Transcript — Paste a Video, Get the Words

This skill turns YouTube speech into searchable, quotable, translatable text. It works with any public YouTube video in any language, supports plain text / timestamped Markdown / SRT subtitle output, and the format is selected automatically from how you phrase the request — not from a settings page.

Who it's for

• Researchers and analysts turning hour-long interviews, podcasts, and conference talks into searchable text — read in 5 minutes what would have taken 90 to watch
• Students and lifelong learners capturing lectures, tutorials, and educational series for note-taking, review, and study guides — the transcript is your textbook
• Journalists and fact-checkers quoting from YouTube videos, public statements, recorded interviews, and political speeches — the transcript is verbatim, never rewritten, so the quote you cite is the quote that was said
• Content creators repurposing their own long-form YouTube videos into Shorts, Reels, TikToks, blog posts, newsletters, and podcast episodes — pull the spoken script out as text in seconds
• Podcasters turning recorded YouTube episodes into searchable show notes, blog posts, and SEO-indexed content
• Marketers and competitive researchers analyzing competitor YouTube channels, brand videos, sales webinars, and product demos — read the talking points without sitting through hours of video
• Engineers, developers, and technical learners capturing conference talks, tutorial walkthroughs, and code review videos — searchable text beats scrubbing through a 60-minute talk to find the one demo
• Language learners using native-speaker YouTube content as study material — verbatim transcripts paired with translated summaries; an English speaker studying Spanish can read what was actually said and get an English summary alongside
• Translators and localization teams preparing source text for subtitle translation, dubbing scripts, or international versions — accurate verbatim text is the foundation of accurate translation
• Accessibility teams and educators generating SRT subtitle files for hearing-impaired audiences when YouTube's auto-captions aren't accurate enough
• Pastors, coaches, and educators transcribing sermons, talks, and online courses for archives, study materials, and follow-up content
• Knowledge workers and executives consuming long industry talks, keynotes, and earnings calls in skim-and-search format instead of real-time playback

What it solves

YouTube hosts the world's largest archive of spoken knowledge — interviews, podcasts, lectures, conference talks, courses, news, sermons, training videos, expert breakdowns. The platform offers auto-captions and a transcript panel, but turning that into actually-usable text is harder than it should be:

• The transcript panel doesn't paginate, doesn't paragraph, and doesn't export cleanly — copying it produces a wall of timestamped fragments that takes more time to clean than to retype
• Auto-captions are inconsistent on accents, technical vocabulary, proper nouns, and overlapping speech — fine for casual viewing, unusable for citation
• There's no built-in way to get a summary — even on a 90-minute video where the summary is the entire reason you opened it
• There's no SRT export for re-uploading or translating — creators who want to add captions to a different platform have to use third-party tools
• There's no way to search across multiple videos — every video is siloed in its own panel

Third-party transcribers exist but most fail in one of three ways:

• They translate the transcript when they shouldn't. You wanted to quote a Spanish-speaking guest verbatim. The tool returned an English paraphrase. The quote is gone, and you can't cite a paraphrase in journalism, research, or legal work.
• They leave the summary in the source language. You don't read German. The summary came back in German. Now you need a second tool to translate it.
• They hide everything behind a settings menu. You just wanted to paste a link and type a sentence. Instead you're configuring API keys, output paths, and toggles.

The shared problem: these tools optimize for the engineer building the workflow, not the researcher, journalist, student, or creator trying to get a job done in 30 seconds.

This skill takes a link and a sentence. It returns the right thing.

What makes it different

1. Two-language output, by default.

Transcript stays verbatim in the source audio language — your quote is intact, word-for-word, exactly as the speaker said it. Summary and key points come back in your conversation language. A Spanish interview transcribed by an English-speaking user gets a Spanish transcript and an English summary. A Japanese lecture transcribed by a Mandarin-speaking user gets a Japanese transcript and a Mandarin summary. No toggle to flip, no language argument to remember. The dual-language output is the default because it's what the use case actually demands — accurate quotation in one language, scannable comprehension in another.

2. Output format from phrasing, not from a menu.

Say "give me subtitles" → you get SRT, ready to drop into Premiere, DaVinci Resolve, CapCut, or YouTube Studio. Say "with timestamps so I can find the part about pricing" → you get Markdown with [MM:SS] paragraph prefixes, citable down to the second. Say nothing format-specific → you get clean plain text, easiest to read, summarize from, and paste into Notion, Google Docs, or Obsidian. The keyword scan reads any position, any inflection, across English, Spanish, Chinese, Japanese, and Korean — "字幕," "タイムスタンプ," "자막" all trigger the same routing as their English counterparts. There is no settings page. There is no format dropdown. The interface is the sentence you would have written anyway.

3. Fast on long content — even a 3-hour podcast.

Long-form YouTube — Joe Rogan-length podcasts, 2-hour conference keynotes, full university lectures, multi-hour AMAs, earnings calls — is where transcription is most valuable and where most tools get slowest. Most AI transcribers feed the full transcript into the summary model before producing output, which puts the summary 60–120 seconds further away on a long video, and several minutes further away on multi-hour content. This skill writes a sidecar digest — each paragraph's first and last sentence — and summarizes from that. Half the input tokens, the same summary quality, and on long content the user-facing wait drops from "five-minute wait for a wall of text" to "under two minutes for a clean summary plus a saved file." The full verbatim transcript still saves to disk in full; the digest is a behind-the-scenes optimization the user never has to think about.

Supported YouTube content

Examples

Example 1 — Researching from a 2-hour podcast

A researcher is preparing a report and needs to capture the key arguments from a 2-hour podcast with a domain expert.

"What did the guest say about supply chain disruption? https://youtu.be/abc123def"

What comes back:

• The supply-chain section first — verbatim excerpts of the guest's exact phrasing on the topic, with surrounding context for accurate attribution
• A 2–4 sentence summary of the broader episode
• 3–6 key points covering the major themes
• The full verbatim transcript saved to a downloadable file, citable line-by-line for the report

Example 2 — Repurposing a long-form YouTube video into a blog post

A creator just published a 45-minute deep-dive video on YouTube and wants to turn it into a blog post, an email newsletter, and a series of social clips without re-watching.

"Transcribe this video with timestamps https://youtu.be/myvideo123"

What comes back:

• A 2–4 sentence summary capturing the video's argument arc
• 3–6 key points usable as the blog post outline or newsletter bullets
• The full timestamped transcript in Markdown, paragraph-level [MM:SS] prefixes throughout — paste into Notion or Google Docs, edit down, and the timestamps double as cuepoints for picking social clips

Example 3 — Subtitling a conference talk for re-upload

An engineering team wants to re-upload last year's conference talk to their internal learning platform with proper SRT captions for accessibility.

"Give me SRT subtitles for https://www.youtube.com/watch?v=xyz789"

What comes back:

• An .srt file in standard SubRip format with proper timecodes — drop straight into the internal video platform's caption uploader, YouTube Studio, CapCut, or any editor
• A short summary and key points usable as the video description and learning-platform metadata

Output formats

Language support

The transcription engine handles English, Spanish, Portuguese, French, German, Italian, Dutch, Mandarin, Cantonese, Japanese, Korean, Vietnamese, Thai, Indonesian, Tagalog, Arabic, Hindi, Urdu, Russian, Turkish, Polish, Czech, Hungarian, and most major European, Asian, Middle Eastern, and Latin American languages. Mixed-language videos (a host code-switching between English and Spanish, for instance, or a panel with speakers in different languages) are transcribed with the dominant language as the base; the verbatim text preserves both languages as spoken.

You don't have to specify the language. URL signals and channel handle clues drive automatic detection. If the first attempt comes back empty, the skill retries once in your conversation language. After two empty attempts, you'll be asked to confirm — usually the video is private, age-restricted, region-locked, members-only, or removed rather than a language miss.

If you already know the audio language, mentioning it in the prompt ("this Spanish-language podcast," "the audio is in Japanese") skips the detection step and saves a few seconds.

Accuracy and what to expect

The transcript is verbatim — produced by a speech-recognition pipeline, never rewritten or paraphrased. Filler words, repetitions, hesitations, and crosstalk are preserved as spoken. This matters when you're quoting for journalism, research, legal contexts, or fact-checking — exact wording is evidence.

Accuracy is high on clear single-speaker audio in supported languages — most podcasts, lectures, and well-produced YouTube videos transcribe with very few errors. Multi-speaker conversations with overlapping speech, panel discussions with crosstalk, very technical vocabulary or proper nouns the model hasn't seen, strong regional accents, low-quality remote audio in older interviews, and severely compressed audio reduce accuracy — there's no AI fix for an unintelligible source. If a section comes back garbled, it's usually a signal that the audio itself is hard to hear, not that the model failed. For high-stakes citation work (journalism, academic research, legal), spot-check critical quotes against the source video.

The skill processes the public video link only. It does not log into YouTube and does not access private, members-only, or unlisted content.

Out of scope

• Local video files on disk — .mp4 downloads from YouTube, screen recordings, files saved to your computer — use the related audio-to-text skill for those instead
• Active Live broadcasts — only finished recordings work; ongoing Lives return empty until the broadcast ends and is saved to the channel
• Members-only and age-restricted videos — the link must be publicly accessible without login or age verification
• Private and unlisted videos — the skill cannot authenticate into YouTube on your behalf
• Music-only videos with no spoken content — instrumental tracks, performance videos with no narration, and lyric-free music videos have nothing meaningful to transcribe
• Translation of the transcript itself — the transcript intentionally stays in the source language to preserve the verbatim quote; if you need it translated, ask for translation as a follow-up step after receiving the verbatim text