Launched a YouTube channel a month ago: AI-generated instrumental music for people building something — synthwave for mornings, dark electronic for late-night coding, ambient for deep focus. No voiceover, no face, just long sessions meant to sit in the background while you work.
The biggest lesson so far wasn't about the music — it was about language. One track used "warm reverb pads" in the AI generation prompt, thinking it'd sound cozier. Instead the video got algorithmically routed into YouTube's meditation/wellness discovery bucket instead of deep-work, and the impression budget for that whole content line collapsed by over 90% on the next upload. The descriptive language you feed a generation model doesn't just shape the output — it shapes which audience the platform decides you're for.
Meanwhile the coldest, darkest, least "pleasant" track in the catalog is the one actually pulling subscribers. Builders apparently don't want warm and inviting at 11pm — they want something that matches the grind.
Still tiny — six subscribers, this is very much day one — but the system's holding up: three distinct modes for three different parts of a builder's day, each one teaching me something different about how discovery actually works underneath.
Today's upload is the morning one — 4 hours of deep-work synthwave for whoever's already at their desk before the rest of the day starts: https://youtu.be/qXLV5ywIy6w
The "warm reverb pads routed you into meditation instead of deep-work" story is genuinely sharp, and it generalizes past music: the words you use to describe your thing decide which audience the algorithm assigns you, before a human ever judges it. True for landing pages and app store listings too, not just prompts.
The more interesting finding is buried: the coldest, darkest track pulls the most subscribers. That's your positioning talking. "Music for builders" is the category, but "matches the grind at 11pm" is the hook. Warm and pleasant is what every focus-music channel offers. The dark, grind-matching angle is the wedge nobody leans into because it feels counterintuitive. Lean into what's working, even though it feels wrong.
Which mode has the best subscriber-per-view rate? That ratio, not raw views, tells you which to build around.