Getting the transcript
Reading the captions from YouTube. A video nobody has opened here before takes 10 to 30 seconds; this page fills in on its own.
Getting the transcript
Reading the captions from YouTube. A video nobody has opened here before takes 10 to 30 seconds; this page fills in on its own.

Sharbel A. · @sharbelxyz
This video has no Most replayed graph yet: YouTube shows one only once a video has enough views. These are the moments viewers replayed most in Sharbel A.'s most watched videos.
Most replayed moment at 13:32
3.9x that video's typical replay level
built-in memory, which is great. Here is how I would think about memory providers. Memo is interesting if you want a dedicated memory layer for personalized AI agents. It focuses on extracting, storing, linking, and retrieving memories efficiently. Their
Said at 13:25
Most replayed moment at 6:37
5.8x that video's typical replay level
trading strategy. Tests it. If it's better, it keeps it. If it's worse, it discards it and tries again. So, that's exactly what I built. Okay, here's the system we have at play. I gave it two years of crypto data,
Said at 6:31
Most replayed moment at 1:53
3.7x that video's typical replay level
let's install it together. Okay, so for step one, we need to actually start by installing Bullpen's CLI so that we can actually do everything that we want to do. And for that, let's first open Claude and put in the dangerously skip
Said at 1:47
The graph counts replays. It does not show where viewers stopped watching.
Words
2,354
Runtime
14:57
Speaking pace
157wpm
Reading time
10min
157 words per minute, below the 160 25th percentile of 349 measured videos. That distribution comes from the 349-video hook study.
Opening (first 30 seconds)
I built an AI agent whose only job is to make viral YouTube videos. This agent browses YouTube, competitor videos, sources transcripts, posted videos, and my channel history before it is allowed to recommend anything. >> [snorts] >> And today, I'm going to let it pick what I should film next and show you exactly how you can use it to grow a YouTube channel as well. By the end of this video, I will be giving you the
79 words, the words spoken in the first 30 seconds at 157 words per minute.
Free, no signup. See how the first 30 seconds hold attention, with rewrites.
Sentence shape
| Measure | This transcript |
|---|---|
| Sentences | 149 |
| Average words per sentence | 15.8 |
| Longest sentence | 46 words |
| Questions asked | 10 |
| Sentences containing a number | 9 |
Most used terms
Filler phrases
33 in total: like 10 · literally 8 · uh 7 · actually 6 · basically 1 · you know 1.
A literal whole-word count of the same phrase list the Prepublish browser extension uses, so a phrase inside another word is not counted and a phrase used in its ordinary sense still is. It is a count and not a judgement.
Free, no account. See where attention is likely to drop, with a rewrite for each weak line. The free check shows the scores and the one issue costing the most. Or run it on the words above first.
Free · No login · See a sample audit first if you prefer.
What this transcript is
Every word below is the caption track YouTube publishes for this video, pulled from the video itself and reproduced unchanged. It is not Prepublish's writing, not a summary, and not a re-transcription: it is the video's own published captions. English captions, generated automatically by YouTube, in the video’s original language. Source: the video on YouTube. A channel that would rather this page did not exist can ask for its removal through the contact page, and it is removed.
I built an AI agent whose only job is to make viral YouTube videos. This agent browses YouTube, competitor videos, sources transcripts, posted videos, and my channel history before it is allowed to recommend anything. >> [snorts] >> And today, I'm going to let it pick what I should film next and show you exactly how you can use it to grow a YouTube channel as well. By the end of this video, I will be giving you the same exact tool for free for you to download and install.
Let's get started. The idea here is pretty simple. AI can help you grow on YouTube. That is exactly why videos about AI YouTube automation, faceless channels, and 7-day channel challenges get so many clicks. But, most of those videos skip the part that actually matters. They make it look like the magic is the tool. Use this AI video generator. Use this script writer. Use this shorts workflow. But, the real leverage is not the tool.
It is choosing the right video before you waste an entire filming day. Because if the idea is wrong, everything that comes after it doesn't really matter. You can have the best script in the world, the best edit in the world, the best thumbnail, the best title. But, if the idea is bad, the video ends up fighting an uphill battle from the start. So, I wanted to build an AI agent for the part I always struggled with myself as a creator.
Finding the best YouTube ideas. But, the real question I had wasn't can AI generate YouTube ideas because of course it can. >> [snorts] >> The real question I had was can an AI agent filter ideas well enough that I would actually trust it with my next filming state. To answer that, I needed to turn it from a title generator into a research employee. But, the way I define that job is very strict. Nova, which is my YouTube agent, is not just allowed to just brainstorm.
If I ask for 50 ideas, I will still have the hard job. Now, I have to sort them, validate them, reject the weak ones, check if I already filmed them, and figure out which ones are actually worth betting on. So, Nova, my YouTube agent, has one job. Bring me the strongest ideas with receipts. That means before it recommends anything, it has to check what I already posted. It has to check what is already in my Notion pipeline.
It has to look at my approved and rejected ideas, so it understands what I personally like and dislike. It has to check competitor videos, YouTube validation, VidIQ when available, and source transcripts before calling something proven. And the most important part is that Nova has to label the evidence. If something is VidIQ backed, say that. If something is competitor backed, say that. If something is just judgment, say that, too.
That is the difference between an idea generator and a strategist. A generator gives you more options. A strategist removes the bad one. So, before Nova is allowed to find anything viral, it has to learn how to say no. So, when I ask Nova for video ideas, it does not just start brainstorming. There is an actual sequence of events that happens. First, it checks my channel memory. It looks at what I already posted, what worked, what flopped, what topics I'm repeating, and what my audience actually responded to.
Then, it looks at what is already in my Notion database, what is already filmed and posted, what is already in to film or write, what has been approved, what has been rejected. Then, it looks for demand. It searches YouTube, it checks VidIQ when available, it looks at competitor videos, and it tries to find real signals instead of just nice-sounding ideas. Then, it finds what I like to call the viral pond, the viral cluster.
That means it looks for the bigger audience desire behind the idea. Not what tool did I build, but what are people already clicking on when it comes to AI agents, YouTube automation, make money with AI, one-person businesses, whatever pond is actually moving that is within my niche. Then, it pulls source videos. It looks for videos with similar promises, strong titles, strong thumbnails, high views, and or clear outlier behavior.
Then, it reads the transcripts for all those videos. Really, this is where it tries to understand what are these videos about behind that title. You can you can't always understand what a video is just by looking at its thumbnail and title. Sometimes, there's a lot more context inside those videos. Uh Nova also looks at the hooks, the structure, the mechanism to try to understand, you know, what really differentiated this one video that did massively versus all of those other ones.
Then, it translates the pattern to my channel. It does not copy the source video directly. It asks, "How does this become a Charbel video? What proof can I show? What system does Charbel already have that fits well within within this idea?" Then, step eight, it's scores the ideas. It looks at demand, channel fit, proof, freshness, packaging strength, and whether the idea would force me to overclaim, for example. And then, step nine, it labels the evidence.
Is this idea with the IQ backed? Is there numbers that back that this is a strong idea? Uh is the idea competitor backed? Is the idea transcript backed? So, it collects and labels all of the evidence and data that it collects. And finally, number 10, it returns only the top picks. Not 50 random titles. Not a giant brainstorm. Just two to three recommendations with receipts, risks, and camera proof, just like it's written over here.
Camera proof is basically what would be nice for me to show on camera, just like I literally am right now. And that is the difference between asking AI for ideas and building an AI strategist. If you want to use Nova yourself, as I promised, I put the repo link in my video description below, but here it is on screen as well. And the easiest way to install it is not to manually figure out every file. You can literally just take this link it, send it to your coding agent.
You can use Hermes agent, you can use Cloud Code, Codex, Cursor, whatever you already personally use to code and write code, send it to that agent. And you can say something like this, "Install and set up this repo for me with the repo link. Read the read me and code base first, then walk me through the setup step by step." This is what's going to help you onboard that agent to your personal workflow. Before changing anything, tell me what this repo does, what dependencies it needs, what accounts or APIs I need, what files I need to configure, how to run it locally, and how to test that it works after setup.
Help me connect it to my own YouTube workflow. That right here is the simplest way to install it. You're not trying to become a developer for 3 hours. You're using your AI coding assistant to install the YouTube strategy agent and have it adapted to your setup. I've already put instructions for your AI agent on how to install it and how to onboard you when you go to install it. So, literally just ask it to help you with that and it will read everything my agent has written to your agent and it will just help you.
And if you're using Hermes, by the way, you can make this even more native. You can ask Hermes to clone the repo, inspect the instructions, and configure the environments, then help you connect Nova to your own files. Say, if you work on Notion, like I do, if that's where you manage a lot of your videos, you can ask it to connect to your YouTube research process, your channel memory, whatever files you currently already use.
The important thing is this. Do not just run random commands blindly. Have your coding agent explain what it is installing, what keys it needs, what data it can access, and how to verify it's working. Because Nova is only useful if it is connected to your real channel context. By the way, last note I'll give on this. There are a few dependencies I use for my YouTube research, like having a logged-in session to YouTube on my desktop that runs Hermes agent and having a logged-in vidIQ session.
Make sure you have those as well if you want just as good results as the ones I get. If it does not know your posted videos, your pipeline, your rules, your audience, It's just going to end up being another idea generator. Like random ideas. This setup is what turns it into an actual useful employee. But, once Nova is installed, using it is mostly about prompting it correctly. And I made this as simple as possible. You can literally ask it, "Nova, what two videos should I film next week?" And behind the scenes, Nova knows the job, it checks your channel memory, it checks your current pipeline, what you already posted, what you filmed, what you approved, rejected, what your audience has responded to, as in what has performed well for you and what hasn't.
Then it looks for demand, it checks YouTube, source videos, competitors, and transcripts when it needs to understand why a video works. Then it translates those patterns into your personal channel. Not, "Here are 50 generic AI ideas." Two to three recommendations. Each one should come back with the title, the source proof, the pattern it is adapting, what you should show on camera, and why it thinks this is worth filming.
Let me show. Literally, here I was asking it to give me the page I was just showing you, but I'm going to run the prompt right here with Nova myself. Nova, what two videos should I film next week? And you're going to see when I ask this, Hermes, since I'm using it on a Hermes agent, it's going to load up my Nova skill to answer that question. It's not just going to guess or just use ChatGPT, it's going to load up Nova.
Uh I'll give it a few minutes to set up and come back to us with the answers. Okay, here you go. This is exactly what I wanted you to see. First, it loaded up the Nova skill, then it started searching our files, it activated the vidIQ bridge to start collecting data. It read our posted videos history, our approved ideas, rejected ideas, performance log, our configurations, everything it needs, and then it started running code to fetch us the best videos.
Now, I literally had two videos in my pipeline, so it gave us the same exact ones, to which I told it no, I want two new videos. I already filmed those two just today day two two new videos just corrected it, and boom, this is what it gave us. The first one is how I build a million-dollar business with Hermes Agent, which is amazing because one, it builds on a video of mine that has found success, five ways I make money with Hermes Agent, which is now at 51,000 views.
Uh so, this is a clear demand that it found in my channel that you guys enjoyed. It also also found uh a lot of our competitors had similar successes with ideas, hence it suggested this as a video idea. The second one is my $8 a month Hermes Agent just got an AI team, which it got from uh one of our competitors. I would maybe switch up the title so that it's literally not a copy-paste of one of our competitors, but the idea's solid.
Its premise is how can we set up a low-cost Hermes Agent workflow or AI team, an entire AI team working for you for just $8 a month, which is such a solid idea. And here you can see, it's telling us what not to make next, uh which are these, which didn't really perform well. But, that being said, if I enjoyed any of them, for example, this one, I really like it, then I can take it and send it this. Hey Nova, I'd love to film this video idea next week, how I build a million-dollar business with Hermes agent, interview me, and let's make the script together.
And that is literally something Nova can help you do as well, from writing hooks to writing scripts, bodies, help you brainstorm what is actually valuable for your audience and your channel context. The best AI YouTube system is not the one that makes videos for you. It is the one that helps you choose the right videos before you waste an entire filming day. That is where the leverage is. If the idea is wrong, the script, edit, thumbnail, and upload package are all just fighting uphill battles.
But if the idea is right, everything after it comes and gets a lot easier. So, if you want to copy this, do not start by asking AI for titles. Start by giving it your channel history, your posted videos, your pipeline, every single piece of data it'll need. Give it Nova. Have it install it for you, and have it help you find the right videos for your channel. Comment Nova if you've watched this far, and if you've enjoyed this video, make sure to give it a thumbs up.
Or, if you're new to my channel, make sure to subscribe, because I have a ton more content coming your way. Oh, would you look at that? The algorithm gods are predicting that you are very likely to enjoy this next video. So, maybe click it, and maybe I'll see you there.
The words are the caption track's own and nothing is reworded or re-transcribed. Paragraph breaks are placed between sentences so the text reads as prose.
Free tools for your own script: paste a draft and see where it stands before you record it.
Paste your draft and see where viewers are likely to drop off, with a rewrite for each weak line.
Paste the first 30 seconds of your own draft for a hook score and rewrites.
Check your draft against YouTube's advertiser-friendly guidelines before you record it.
Read this channel's public videos and transcripts, and download a writing brief for it.