How to get a YouTube transcript for ChatGPT or Claude (and what 23 real videos looked like)
Short answer: run a YouTube transcript scraper on the video URL and paste the transcript (or ready-made Markdown) field into ChatGPT or Claude. No YouTube API key or login is needed, because it reads the public captions. On 2 November 2026 we ran the YouTube Transcript Scraper on 24 videos found through three searches: 23 returned a transcript (326,648 words, 31.6 hours of video) and 1 had captions disabled. A 5-minute explainer is about 1,000 words; a 5-hour podcast is about 60,000 words, so long videos need a long-context model or chunking.
What 23 real transcripts looked like
Three searches ("how to use ChatGPT", "AI agents explained", "Lex Fridman podcast"), top 8 videos each, English preferred. Every transcript was returned as plain text plus timestamped segments.
| Video | Channel | Length | Words | Captions |
|---|---|---|---|---|
| ChatGPT Tutorial: How to Use Chat GPT For Beginners | Charlie Chang | 28 min | 6,253 | auto-generated |
| 23 Ways to Use ChatGPT So Well it Feels Like Cheating | Dan Martell | 17 min | 3,283 | auto-generated |
| Learn 95% of ChatGPT Work in Under 20 Minutes | Jeff Su | 20 min | 3,375 | auto-generated |
| AI Agents, Clearly Explained | Jeff Su | 10 min | 1,698 | auto-generated |
| AI Agents Explained - What Is an AI Agent and how to build one | Tech With Tim | 22 min | 5,101 | auto-generated |
| AI Agents explained in 3 steps | Jeff Su | 52 sec | 147 | auto-generated |
| AI Crisis in Math: Navier-Stokes, Riemann Hypothesis & More | Lex Fridman | 3 h 34 min | 38,095 | manual |
| DHH: Future of Programming, AI, Agentic Engineering | Lex Fridman | 5 h 16 min | 54,046 | manual |
| Dario Amodei: Anthropic CEO on Claude, AGI & the Future | Lex Fridman | 5 h 15 min | 59,948 | manual |
Totals across the 23 videos with a transcript: 326,648 words, 36,008 timestamped caption segments, 31.6 hours of video, an average of 172 spoken words per minute. 16 had auto-generated captions and 7 had manually uploaded ones (all seven from Lex Fridman's channels). One was in Hindi (an Intellipaat video, hi), and one video ("30 ChatGPT Secret Codes & Prompts", 7 seconds) had transcripts disabled by its uploader and returned an error instead of text.
What it shows
- Rule of thumb: about 170 words per minute of video. Individual videos ranged from 128 to 273 words per minute in this run, so use roughly 10,000 words per hour of video to size a job.
- Long podcasts are big. The three Lex Fridman episodes above are 38,000 to 60,000 words each. At the common rough estimate of about 1.3 tokens per English word, that is roughly 50,000 to 80,000 tokens per episode (an estimate, not a tokenizer count), which is why bulk summarising usually means chunking.
- Short videos are cheap. A 1-minute clip is about 150 words, so you can feed hundreds of them into one prompt.
- Auto-captions are the norm. 16 of 23 videos relied on auto-generated captions, so expect occasional mis-heard names and no speaker labels. Manual captions were rarer and only on Lex Fridman's channels.
- Not every video has a transcript. Uploaders can disable captions, and the scraper reports that instead of guessing.
How this was collected
- Source: public YouTube captions, no login and no API key.
- Date and method: 2 November 2026, run locally with the YouTube Transcript Scraper, search queries
how to use ChatGPT,AI agents explainedandLex Fridman podcast, 8 videos per query, language preferenceen, timestamps on. - Caveats: the sample is the top search results on one day from one connection, not a random sample of YouTube. Word counts are of the caption text. Token numbers are estimates. View counts and rankings change daily. For larger runs on the Apify platform, YouTube blocks many datacenter IPs, so a residential proxy is recommended (the actor's default).
Get the same data yourself
Single videos, playlists, channels and searches all work:
{
"videoUrls": ["https://www.youtube.com/watch?v=FwOTs4UxQS4"],
"searchQueries": ["AI agents explained"],
"maxVideosPerSource": 8,
"languages": ["en"],
"includeTimestamps": true,
"includeMarkdown": true
}
Each result has transcript, segments (start, duration, text), markdown (title, channel, link and transcript, ready to paste), language, isGenerated, title, channel, viewCount, publishDate and wordCount. Set translateTo (for example en) to auto-translate when YouTube offers it.
Ask an AI assistant instead: connect the scraper through the Apify MCP server (https://mcp.apify.com/?tools=get_anything/youtube-transcript-scraper) and ask "Get the transcript of this video and give me five key takeaways with timestamps." If you also need video and channel statistics rather than text, see the YouTube Scraper.
FAQ
How do I get a YouTube transcript to paste into ChatGPT or Claude?
Give the scraper the video URL, then copy the markdown or transcript field from the result. For long videos, paste it in sections or upload it as a file.
Do I need a YouTube API key or login?
No. The scraper reads publicly available captions, so no API key, login or cookies are needed. You only need a free Apify account to run it on the platform.
Does it work with auto-generated captions and other languages?
Yes. In our run 16 of 23 transcripts were auto-generated, and one was Hindi. You can set preferred languages, and optionally auto-translate where YouTube allows.
Can I get transcripts for a whole playlist or channel?
Yes. Pass playlist or channel URLs (or a search query) and set maxVideosPerSource to cap how many videos are taken from each.
Why did one video return no transcript?
Some uploaders disable captions. In our run, 1 of 24 videos had transcripts disabled; it was reported as an error rather than returned as empty text.
