Every “Recommended for you” is trained on one thing: what you already clicked. Feed it enough plays and it narrows, quietly, track by track, until the feed mostly reflects itself back at you.

Nobody chose that on purpose. It's just what optimizing for “keep listening” does, given enough time.

Recommendation systems are popularity machines first. A track that's already been played a million times has a million data points behind it. A track from three days ago has none. So the system plays it safe, it keeps surfacing what's already big, because that's the surest bet.

That's not a conspiracy, it's math. But it means the system meant to help you find something new is structurally biased against anything actually new.

A curator picks a track because it's good, full stop. Not because it statistically keeps you around three minutes longer. There's no test behind “heard this once, on the radio, and it stuck.”

“New to me” is a small, deliberate exercise in the opposite direction: listen first, guess, then find out, no feed shaping the answer before you've even heard the question.