I stopped watching educational AI videos on YouTube months ago, not because the knowledge stopped mattering but because the time cost stopped adding up. In their place I built an AI video triage system that reads the transcript, measures it against my actual business, and hands back only the parts worth acting on. Here is why the gold nuggets are too few to keep paying full price in hours, and how to build the same system yourself.
My YouTube home page knows me well. Open it on the wrong morning and it serves up a wall of videos I genuinely want to watch, almost all of them about AI, almost all of them wearing titles engineered to make me click. For a while I watched them. Then I did the math on what that habit was actually costing me, and I changed how I deal with the entire category.
Ain’t Nobody Got Time For This
There was a meme a few years ago, “ain’t nobody got time for this.” It has mostly aged out of the Gen Z vocabulary, but it captures my relationship with the AI corner of YouTube exactly. The algorithm worked out that I live in this field and started serving a steady diet of videos with titles like “this tool is going to end every other AI tool,” “this thing is insane,” and “do ten times more with this one trick.” Every one of them promised a breakthrough. My open tabs multiplied faster than I could ever clear them, and a quick scan of how many interesting videos were out there started to feel less like opportunity and more like a standing threat to everything else I was supposed to get done that day.
Admitting you have stopped watching educational videos sounds like a confession. The implication is that you traded them for shorts and an endless scroll of cat videos, which are, in fairness, among the top three things on the internet. That is not what happened. I still wanted the knowledge. I just stopped pretending the format was a good deal.
The Gold Nuggets Are Few and Far Between
Every educational video on YouTube serves two masters. The creator’s purpose is to make money, which means most of the content orbits the narrow funnel of whatever course, tool, or service they are selling. Your purpose is to actually learn something useful. Those two goals overlap far less than the run time implies. The honest answer, the one that took me a while to accept, is that the real insight in a typical seventeen-minute AI video is a thin seam running through a lot of filler. Worse, you are never certain you will find that seam, and even when you do, you are not always sharp enough to recognise it in the moment. You can give a video twenty minutes and come away with nothing, or come away with the single good idea and never notice it went past.
AI Doesn’t Watch the Video. It Reads Around It.
Here is the thing most people get wrong about AI video summaries: the AI is usually not watching anything. The common flow, paste a link and get a summary, works off the transcript, the timestamps, the description, and the title. It reads around the video rather than through it. Some multimodal models can now process the actual footage, but that is slower and it is not what the everyday summarising tools reach for. None of that matters if the output is weak, and early on mine was weak. I would hand a link to whatever model was nearest, get back a tidy paragraph, and feel like a bit of a sucker. I had not watched the video, but I had not really gained anything from the summary either, so I was no closer to knowing whether the thing deserved my time.
A Generic Summary Doesn’t Know You
The reason those summaries fall flat is simple. The AI does not know you. Ask a cold model to “tell me the interesting parts of this video” and it has to guess at what interesting means, because all it has is the handful of facts you once typed into a profile field, if even that. A model with no context about what you do, which tools you already run, and where your real gaps sit cannot tell you which ninety seconds of a sales video are worth your attention. It can only summarise. Summarising is not advising, and advising was the part I actually needed.
I Taught the AI My Business, Then Pointed It at the Video
So I built the context myself. I created a skill, which is less mysterious than it sounds; a skill is a focused set of instructions that tells an AI how to operate and what it is meant to achieve. Into that skill I fed my projects, my documentation, my day-to-day operations, and an honest account of where the biggest gaps in my business sit. I taught the model what I do, how I do it, and what I am trying to fix. You can find ready-made versions of this online, from plugins to shared skill sets, but building my own meant the instructions matched my reality instead of someone else’s. Then I paired the skill with a scraper that pulls the transcript and aimed the whole thing at a specific video. It took several passes to sharpen; the early analysis was blunt, and it got noticeably better the more precisely I told it what I was after.
What the Triage Actually Returns
Now when a seventeen-minute video about some AI tool lands in front of me, I can smell the sales pitch from the thumbnail; nine times out of ten that run time is a course in disguise. I still assume there might be one good idea buried in it, so I let my agents take it apart. What comes back is not a summary. It is an evaluation. It tells me what the video is genuinely about, as opposed to what the title is shouting. It tells me which of its claimed tips I am already doing, and whether I am doing them well or badly. It surfaces the occasional real improvement I had been missing. And it closes with a clear proposition: this specific idea is the strongest candidate for refining my current approach, so adopt it; these few things are worth knowing for context; and here is a second list to revisit later, useful but not a priority today. That sorting, act on it, know it, or park it, is the entire point of the exercise.
How to Build Your Own YouTube Triage System
If you want the same thing, the build is more about discipline than technology. Four steps get you there.
- Write down your context. In plain language, capture what your business does, which tools and processes you already run, and where you honestly think your weak spots are. This is the part most people skip, and it is the part that turns a summary into advice.
- Give the AI a way to read the video. A transcript scraper is enough, since you are feeding it the words, the timestamps, and the description rather than asking it to watch the footage.
- Ask for a triage, not a summary. Tell the model to sort every tip in the video into three buckets: what you already do, what is genuinely worth adopting, and what to skip.
- Demand a bottom line. Make it give you a single clear call on whether anything in this video is worth your time, and what to do about it.
Expect the first few runs to be mediocre. The system sharpens as you correct it on what it misreads about your priorities, and within a couple of weeks it knows your business well enough to be trusted with the filtering.
Who This Is Actually For
This is not for everyone. If you watch three carefully chosen videos a week and enjoy them, keep watching; you do not have the problem this solves. This is for the people drowning in a feed of AI content (or something similar along the lines) they feel obligated to keep up with, who still want the knowledge but have run out of hours to mine it by hand. For that person, and for a long stretch that person was me, routing educational videos through an AI triage skill instead of watching them buys back real time and loses almost nothing worth keeping. The gold nuggets were always few and far between. The only difference now is that something else does the digging, and it already knows what I am looking for.