How I built Silas AI

I knew I wanted to do a channel where I actually try.

I've made a lot of channels before this, shorts and long-form. A few got past a thousand subscribers. A lot of them I posted a few videos on and gave up, because I saw no progress and I knew there wasn't much potential there. I was creating for fun. A lot of them were pretty bad, but over time I learned how to make better videos, and eventually some of them actually became good videos. It's just that there was no niche or strategy behind them, so they weren't the type of thing that would actually work as a channel on YouTube.

I've also worked for a few YouTubers, and I know a decent amount of creators, so I've talked strategy with people running channels way bigger than mine. Between that and my own stuff, I had a pretty well-rounded picture of how YouTube actually works, and how to actually have a channel that gets views, because it's not just about having “good videos”.

Then this summer, before my senior year, I had time off and figured: why not lock in and apply all of that to my own channel? Instead of making videos just for fun or working for someone else, make videos for fun that also go toward my own successful channel, one I could maybe actually make a living off of.

Coming into the summer I knew I wanted to do this, but the first big decision is the niche. What type of content do you actually make.

I had a few different ideas, all from stuff I already watched. All types of content, from IRL to faceless channels, across a bunch of different genres. Anything I had an interest in that was also a proven concept. After listing out and thinking through a bunch of them, I realized this format, entertaining videos where AIs compete, was the best, for a few reasons.

The videos are decently hard to make. That sounds like a downside, but it means the space doesn't get polluted. It's not an easy trend channel. So there's long-term potential, and I want this to last for years. The AI space is growing, the demand is growing, the audience is huge.

And I saw a gap. There's a ton of AI content, but it's either too educational, like AI news and AI tutorials, or it's just broad entertainment. There weren't many channels that actually had entertaining videos with AI. But there were a few doing something similar. There was this popular format: ChatGPT vs Gemini vs Claude recreate a popular game from scratch. Fortnite, Minecraft, a horror game, Subway Surfers, anything like that. Some of those channels had around 100,000 subscribers and were relatively new, and most importantly, multiple of their videos had over a million views.

So I watched the videos, and they were kind of disappointing. The creators were bland. You never saw them. Any time they talked, they filled the screen with a black background and random brain rot memes. Good, watchable videos (I watched them, that's how I found this), but something was missing. Personality. Better editing. Better sound, better music, less repetitive pacing.

And this is something that applies to all of YouTube and even business: you don't need to reinvent the wheel. Take what's working and make it 5% better, or slightly change something that already works. Same proven format, but with what I personally think makes a better video: more personality, and a little more uniqueness.

I used an old channel I'd started about a year earlier with a few random shorts on it that got no views. It's good to have an aged channel so YouTube doesn't think you're a bot account. I named it BYTE at first, with a pixelated toast with a bite out of it as the logo. I changed it later to Silas AI so it had more personality and fit better into the branding ecosystem I want to build.

I also wanted a weekly upload schedule. Each video only took a few days to make, so I could do it, and uploading weekly can help you grow really fast on YouTube.

Now for the first video. After some quick research I went with ChatGPT vs Claude make Geometry Dash. Two popular AIs, a popular game, and one of those creators had about three million views on ChatGPT vs Gemini make Geometry Dash. So take that video, make it slightly better and a bit different, since Claude is a different, better, and more newly relevant model than Gemini. It got about 2,000 views in the first week, which was good. YouTube was clearly pushing it a bit and testing it. Not a lot of views, but that doesn't matter much, because it takes time for YouTube to find the right audience. As long as it's testing it and getting some views, that's fine.

Second video: ChatGPT vs Gemini vs Claude make Minecraft. I was hesitant about this one, because one of the channels doing this format had already made pretty much that exact video, and it didn't do too well for them. I didn't want to just copy it, but their video was kind of lame and I knew I could do it slightly better. For about two weeks it sat around 4,000 views. A bit better. YouTube's pushing it a little, trying to find the audience. I'll take it. You don't take off the second you start.

Third video: ChatGPT vs Claude make Super Mario Bros from scratch. This one grew a bit differently. It didn't flatline after a few days like the first two. I remember it being at about 1,000 or 2,000 the first day or two, then it kept slowly climbing, slowed down somewhere around 5,000 or 6,000, and pretty much flatlined around 8,000 after about a week. Better, and about what I was expecting.

Then came what I'd actually been waiting for. The video before it, the Minecraft one, suddenly really started getting pushed. It went from about 4,000 to 10,000 in a day, then quickly up to 50,000, then 100,000, then 200,000, and finally slowed down after a few weeks. It's currently sitting at almost 400,000 views and is still slightly growing.

That one video woke the whole channel up. The Super Mario one went to 30,000, Geometry Dash went to about 10,000, and while it was still getting views, the videos after it did decent too. On the best day the whole channel did 60,000 views in a day.

The Minecraft one was doing extremely well at that point, right at its peak, so I figured, why not do another one around Minecraft? There's already a popular outlier format on YouTube, 1 minute vs 1 hour vs 1 day, so why not have Claude make Minecraft in 1 minute vs 1 hour vs 1 day. That got 100,000 views in the first few days and was my second most popular video for a while, though it recently got passed, which I'll get to later. It's sitting around 130,000 views now. Not getting many at the moment, but it could always pick up later.

I technically hit the monetization requirements in about 21 days. YouTube Studio showed 4,000 watch hours, but that number runs about five days behind because they verify the hours are real. Then, since I'd never made an AdSense account, I had to set all of that up and ran into a bunch of random issues. If I'd already had AdSense I would've been monetized in around 30 days. Instead it was 39.

So it took 39 days from a brand new long-form channel from scratch, with no shoutouts or paid promotion of any kind. Just one long-form video every week. By the time I actually got monetized I already had about 20,000 watch hours, so I got a huge push of views right before I could actually start earning from it, which is unfortunate. But there's a lot more to come.

This started out as an experiment to test my YouTube knowledge. It went about how I expected, and I'm happy with the results, but I also happen to actually like making these videos, and I'm happy with how much it's currently earning.

At the time of writing this, it's been three months since the first video, and I just hit a million long-form views. My last two videos were sponsored brand deals. I can't say exactly how much, but between everything I'm currently making around the amount of a modest salary. Most of it is from sponsors, because the AI niche is really good for that. AdSense doesn't pay too well since the videos are short. Keep in mind it fluctuates, so it could easily go down or way up.

And the old videos keep going. The Super Mario one that flatlined at 8,000 got a second push, flatlined again at 30,000, and is now my second most viewed video at around 160,000. The fifth video, right after the Minecraft ones, flatlined at 18,000, which I was kind of disappointed by at the time. It's about to hit 100,000. A more recent one did a bit above average, then flatlined, then got another bump, and is still getting more views than some of the newer ones. So some of the videos I'm releasing right now that look like they're flatlining may just need one more push for YouTube to find the right audience. Sometimes you have to let it marinate and trust the process.

I started this niche right around the same time a few other creators seemed to find it too. Since then a ton of channels have copied it, and they're all pretty much replicating the exact same videos, just copying each other. Short term, it's working for a lot of them. But I don't think there's much long-term potential in just trend-jacking. And personally, I like having personality in the videos, having me be interesting, and having fun with formats and editing, instead of running a repetitive channel purely to get views.

At the time of writing this, I'm working on a video where I make my own AI, and I'm editing one called I Tested Terrible AI Sponsors in My Inbox. So it's getting slightly broader: taking the interest that me and a lot of other people have in AI and making entertaining videos around it. I have a lot of other ideas too. AIs competing to day trade, AIs running a YouTube channel, and maybe even going back to the OG format every now and then.

Everything I learned from years of failed channels went into one that worked.

And if you read this far, go watch a video. That's the whole point of the channel anyway.