Use YouTube transcripts with ChatGPT, Claude, and Gemini

Open any YouTube video, swap the domain to ytscribe.io, click Copy as Markdown, paste into your LLM. Two clicks, no extension, no API key — and the title and source URL travel with the transcript so the model gets proper context.

The pasting problem nobody talks about

Most ways of getting a YouTube transcript into ChatGPT are bad. The native YouTube transcript panel has no copy button, so you drag-select the entire panel — and end up with a wall of 0:00 word\n0:02 next word lines that eat half your context window before the model has read a single sentence. Browser extensions exist, but most are abandoned, half are ad-supported, and the trustworthy ones still output cluttered text.

What you actually want is a clean paragraph version with the title and source URL attached, so the model can cite the talk by name and you can ask follow-ups like 'go deeper on the second point' without re-stating the source.

The two-click method that just works

ytscribe gives you a Markdown export tailored for exactly this use case. The export starts with a level-1 title, then a four-line front-matter block (source URL, language, captions kind, snippet count), then the paragraph view, then the timestamped view as a second section.

  1. Open the YouTube video in your browser.
  2. Replace youtube.com with ytscribe.io in the URL bar.
  3. Click the small dropdown arrow next to the Copy button and choose Copy as Markdown.
  4. Paste into ChatGPT, Claude, Gemini, or Perplexity. The LLM will see the title and source URL and cite them correctly in any follow-up.

Picking the right model context

Most modern LLMs handle long transcripts fine. Claude (Sonnet 4.x and above) and Gemini 2.5 Pro both have million-token context windows, so even a 4-hour podcast fits. ChatGPT's GPT-5 series handles transcripts up to several hundred thousand tokens. If you're on a smaller model or running a tight token budget, switch to the Timestamped view in ytscribe before copying — the timestamped format is denser and tends to compress better when models summarise.

For Perplexity, the workflow is the same: paste the Markdown into the chat, then ask the question. Because the front-matter includes the source URL, Perplexity's grounding will treat it as a citable source rather than free-text.

A worked example

Open Karpathy's "Intro to Large Language Models" on ytscribe, click Copy as Markdown, paste into Claude with this prompt:

"Here is the transcript of a one-hour talk on LLMs. Give me a five-bullet summary, then list any specific tools or papers Karpathy mentions, then suggest three follow-up questions I could ask to deepen my understanding of the limitations he describes."

The response will reference Karpathy by name, cite the talk title in the summary, and link follow-up questions back to specific sections — because the front-matter told it who and what the source was. Without the front-matter, you'd get a generic summary of an unattributed talk.

FAQ

Does this work with Claude, Gemini, and Perplexity too?
Yes. The Markdown export is plain UTF-8 Markdown with no LLM-specific formatting, so any model that accepts text input handles it. Claude, Gemini, Perplexity, Llama-based local models, GPT-5 — all of them parse the front-matter correctly.
Do I need a ChatGPT Plus or Claude Pro subscription?
No. Free tiers of ChatGPT and Claude both accept pasted text within their normal context limits. ytscribe gives you the cleanest possible input, which means you fit more of the actual content into the free-tier window.
Can I send a YouTube transcript via the OpenAI or Anthropic API?
Yes. The Markdown output is plain text, so include it as a user message via the API the same way you'd include any other prompt content. The front-matter helps the model cite the source if the response is later quoted.
What if the video is in Spanish or Hindi?
ytscribe sets the Language field in the Markdown front-matter to whatever language the transcript is in. The LLM uses that to decide whether to translate, summarise in the original language, or both. You can also pick a different language from the language picker on the transcript page if YouTube has multiple captions tracks.
Is there a faster way than copy-paste?
Yes — install the bookmarklet on the homepage. One click on any YouTube page opens that video's transcript on ytscribe, then one more click on Copy as Markdown sends it to your clipboard. Total time from YouTube tab to ChatGPT paste is about three seconds.