--- title: "How to get AI to summarise a YouTube video" url: https://howtogetaito.com/summarise/a-video/ published: 2026-09-15 updated: 2026-10-04 author: "Jack Stovell" author_profile: https://i.hilyt.it/jack publisher: "Adapt Progress Evolve" publisher_legal_name: "Adapt Progress Evolve Limited" publisher_company_number: "10047328" publisher_vat_number: "GB362654684" summary: "Get the substance of a long YouTube video from its transcript — chapters, claims, what's being sold — and know what a summary can't see." tools: ["Google NotebookLM","Gemini","YouTube","ChatGPT","Claude"] language: en-GB --- # How to get AI to summarise a YouTube video > Get the substance of a long YouTube video from its transcript — chapters, claims, what's being sold — and know what a summary can't see. Published 2026-09-15 · Last updated 2026-10-04 · By Jack Stovell (https://i.hilyt.it/jack) · Canonical page: https://howtogetaito.com/summarise/a-video/ Author profiles: X https://x.com/StovBuilds · Instagram https://www.instagram.com/stovbuilds · LinkedIn https://www.linkedin.com/in/jackstovell/ Operated by Adapt Progress Evolve Limited. Registered in England and Wales, company no. 10047328, registered office 5 Brayford Square, London E1 0SG. VAT GB362654684. ## What you'll be able to do You'll be able to take a 40 minute lecture, a rambling podcast upload, or one of those product review videos that takes 12 minutes to say "it's good, buy it", and get the actual substance out of it in under a minute. Chapters, key claims, the bit where they tell you what they're actually selling. All without watching the whole thing. That's the promise, anyway. It works well, most of the time, for most videos. Not always. More on that later. ## Before you start Right, so here's what's actually happening when you ask an AI to "summarise this video". In almost every case, it isn't watching the video. It's reading the transcript, the same captions you can turn on yourself, and summarising that text. Which means anything that isn't spoken, the demo, the chart on screen, the facial expression that tells you the presenter doesn't quite believe what they're saying, gets missed entirely. This matters more than you'd think. A cooking video where someone says "and now we fold it in gently" is not the same as watching them fold it in gently. A tutorial where someone says "as you can see here" is summarising nothing you can see. The AI just skips past it, or fudges something generic. So a summary is genuinely useful for getting the shape of an argument, the main points, whether a video is worth your time at all. It's less useful for anything visual, anything demonstrated rather than said, and anything where tone matters. Sarcasm doesn't survive a transcript. Neither does nuance, most of the time. You get the flattened version. Keep that in mind and you'll use these tools well. Forget it and you'll end up confidently repeating something the presenter didn't quite mean. ## Get set up There are a few routes in here, and they're not all equally reliable. **NotebookLM** (from Google) lets you add a YouTube link directly as a source, as long as the video is public and has captions. It'll pull the transcript in behind the scenes and let you ask questions, get summaries, generate notes, all from that one link. This is probably the smoothest option if the video qualifies. If you're thinking of going the other way and making a clip rather than shrinking one, that's a different job entirely. I've put together a guide on [making a short video with AI](https://howtogetaito.com/make/a-video/), from the script and footage through to the voice and the edit. It also covers the bits it still can't manage, which saves you finding that out halfway through. **Gemini** can also summarise a YouTube video if you give it the link and ask, though this depends on your region and your plan, so it might not work for you the way it works for someone else. Worth trying, just don't be surprised if it doesn't play ball. **YouTube itself** shows AI-generated summaries, or an "Ask" feature, on some videos for some users. If you see it, great, it's built in and free. If you don't see it on a video you're watching, that's normal too. It's not rolled out everywhere. Then there's the route that always works, or close enough to always: **grab the transcript yourself and paste it into ChatGPT or Claude**. Under most YouTube videos, click the three dots (or "more") below the description, and choose "Show transcript". It'll open a panel with the whole thing, timestamped. Click the three dots on that panel and you can usually turn off timestamps to get cleaner text, or just copy the lot as is. Paste that into your AI of choice. This is the boring, reliable option. It doesn't care about your region, your plan, or whether the video happens to be one YouTube has decided to bless with a summary feature. If the video has captions (auto-generated ones count), you can get a transcript, and if you can get a transcript, you can get a summary. If the video has no captions at all, none of this works directly. Your move there is to download the audio and run it through a transcription tool first. There's a separate guide on this site for that, worth reading if you hit this wall often. ## Try it yourself Once you've got a transcript, or a link that one of the tools above will accept, here's how to actually ask for what you want. Don't just say "summarise this". You'll get something bland. Ask for structure instead. For a lecture or long talk, paste the transcript and try this: ``` Here is the transcript of a video. Give me: 1. A chapter breakdown with rough timestamps if they're in the transcript 2. The 5 main claims or arguments made, one line each 3. Anything stated as fact that sounded like an opinion Keep it tight, no fluff. ``` For a product review or anything with a sales angle, this pulls out the bit that actually matters: ``` This is a transcript of a product review video. Tell me: - What the presenter is actually recommending, and any caveats they gave - What they're selling or promoting (affiliate links, sponsorships, their own product) - Anything they demonstrated that I should go watch myself rather than take their word for ``` For something dense, like a technical talk or an interview, you might just want the plain gist first: ``` Summarise this transcript in plain English, as if explaining it to someone with no background in the topic. Then list any technical terms I'd need to look up to understand it properly. ``` If you're using NotebookLM with the video added as a source, you don't need to paste anything, just ask it directly in the chat: "What are the main points of this video?" or "What does the presenter recommend and why?" It'll pull from the transcript it's already ingested. ## Check the result Don't just trust the summary because it sounds confident. AI summaries always sound confident, that's sort of the problem. A few checks worth doing. Skim the actual transcript for a section the summary claims to cover, and see if the summary matches what was actually said, not a softer or more generic version of it. Pick one claim from the summary and search the transcript for the exact sentence it came from; if you can't find anything close, that's a red flag. Watch the bit of the video the summary says matters most. If it was a demo, a chart, a moment where the presenter changes their tone, go and actually look at it. You'll often find the summary missed something the visuals were doing all the work on. And check the claims against what you already know, if you can. Summaries strip out hedging. "This might work in some cases" can quietly become "this works" once it's been through a summariser a couple of times. Worth a sanity check, especially for anything medical, financial, or where someone's trying to sell you something. ## If it doesn't work If the summary is thin, vague, or clearly wrong, the transcript itself might be the problem. Auto-generated captions on YouTube can be rough, especially with accents, technical jargon, or background music. Garbage in, garbage out, as they say. Try a different video, or check the transcript quality before blaming the AI. If NotebookLM or Gemini won't accept the video link, fall back to the manual transcript-and-paste method. It's slower but it basically always works if captions exist. On privacy: don't paste transcripts of anything private, confidential, or belonging to someone who hasn't agreed to have it shared, into a public AI tool. That includes recordings of meetings, unlisted internal videos, or anything with personal information in it, names, addresses, health details, that sort of thing. Treat a video transcript the same way you'd treat any other document you're handing over to a company's servers. If in doubt, don't paste it. Also worth remembering: these tools don't watch the video. So if a summary seems to describe something visual with real confidence, that's worth double checking, it might be guessing based on context rather than actually knowing. ## Questions people ask ### Does AI actually watch the YouTube video to summarise it? No. In almost every case it isn't watching the video, it's reading the transcript, the same captions you can turn on yourself, and summarising that text. Anything not spoken, a demo, a chart on screen, a facial expression, gets missed entirely. ### Which AI tool is best for summarising a YouTube video? NotebookLM lets you add a YouTube link directly as a source if the video is public and has captions, and it's probably the smoothest option. Gemini can also do it from a link, though that depends on your region and plan. YouTube itself shows AI summaries or an "Ask" feature on some videos for some users. ### How do I get a YouTube transcript to paste into ChatGPT or Claude? Under most YouTube videos, click the three dots (or "more") below the description and choose "Show transcript". Click the three dots on that panel to turn off timestamps for cleaner text, or copy the lot as is, then paste it into your AI of choice. ### What should I ask for instead of just "summarise this video"? Don't just say "summarise this" or you'll get something bland. Ask for structure: a chapter breakdown with rough timestamps, the five main claims one line each, and anything stated as fact that sounded like an opinion. ### Why is my AI video summary vague or wrong? The transcript itself might be the problem. Auto-generated captions can be rough, especially with accents, technical jargon, or background music, so check the transcript quality before blaming the AI. If NotebookLM or Gemini won't accept the link, fall back to the manual transcript-and-paste method. ## Keep exploring If this was useful, the same approach works well beyond video. There's a guide on this site for how to get AI to summarise a document, which covers PDFs, reports, and long written pieces using much the same logic. If you've got audio without captions, whether that's a podcast, a voice memo, or a recorded meeting, the guide on how to get AI to transcribe an audio recording will get you a transcript to work from. And once you've got notes from any of this, whether pulled from a video or written yourself, there's a guide on how to get AI to turn my notes into flashcards, handy if you're trying to actually remember any of it rather than just skim past it once. ## Sources and review notes - Google NotebookLM Help: https://support.google.com/notebooklm/ - YouTube Help: https://support.google.com/youtube/ - Gemini Apps Help: https://support.google.com/gemini/ Review date: 15 September 2026 — every source above was opened and checked against the text on that date.