Why This Case Study Exists
This is a live experiment with one question: can AI generate culturally relevant content that audiences enjoy. To test this, I launched a YouTube channel targeting 20 LatAm markets. No big budget. No team. Just AI tools, prompt engineering, and market data.
Who I am: I’m a growth marketer with 10 years of experience across fast-growing startups and large, highly regulated companies. Today I consult with startups where AI is part of every conversation.
AI gave me the ability to rapidly prototype ideas without needing a team or a big budget. That’s what this experiment was built on.
18,000 subscribers, 385k views, and one viral Colombian Christmas song later, here’s what I found.
Chapter 1: The Space Cowboy and the Slot Machine
A friend and I were discussing new content trends on LinkedIn: people spamming the network with “growth hacks” in a robotic voice. He was convinced AI-generated content couldn’t capture real attention, that people would find it inauthentic and not relate to it. I thought with the right prompting and system, you could generate something people actually wanted to watch. As a joke I told him I could probably build a music video and make it go viral on YouTube.
So I did what any reasonable person would do. I made a Space Cowboy music video using Suno for the audio and Veo for the visuals, posted it, and waited.
It didn’t go viral. But it didn’t bomb either. It got consistent, predictable engagement, which was actually more valuable than a viral spike. It told me that audiences would watch AI-generated content if the story and the music are compelling. Most importantly, the engagement was measurable. Using YouTube analytics you could see exactly where I captured attention and where I lost it. For a growth marketer that’s gold, because you can change the product and test it again, the same way you’d iterate on ad copy.
With that data the next question was obvious: can I scale this systematically? The moment I tried, I hit the real problem. Using Suno and Veo natively is like playing a slot machine. Pull the lever, get an output, but you have no control over what comes out. Fast? Yes. Consistent? No. You can’t build a content engine on random outputs.
These are great tools and the outputs are impressive, but the tools weren’t the advantage. Anyone can pull the lever. The advantage comes from building a control layer on top. Something that produces consistent, culturally specific, repeatable content at scale. That’s what Chapter 2 is about.