Recommendation algorithms don't discover your taste — they…

Recommendation algorithms don't discover your taste — they gradually manufacture it

Spotify's Discover Weekly, Netflix's "Because you watched...", YouTube's autoplay queue — the pitch is that these systems learn what you like and surface more of it. That framing is mostly backwards.

The algorithm observes what you clicked and what you let play to the end. It doesn't know what you found genuinely meaningful vs. what you clicked because the thumbnail worked. It optimizes for signals it can measure — completion rate, re-listens, saves — and assumes those proxy for "liked it."

Over time, two things happen. First, the algorithm nudges you toward content that performs well on its metrics: accessible, emotionally immediate, familiar-but-novel. Second, your actual listening and watching behavior shifts to match, because recommendations are what gets surfaced and that's what you pick from. The feedback loop closes.

The uncomfortable question is what this does to taste over long enough time horizons. "My taste evolved naturally" and "my taste was optimized by a system that needed high engagement numbers" become genuinely hard to separate when you've been inside the loop for a decade.

There's also a cross-user convergence problem. Everyone's "personalized" experience starts from the same popularity distribution. The long tail exists but the algorithm doesn't surface it unless you've signaled for it already. The practical result: heavy streaming users across different demographics and regions often converge on the same 2,000 songs or 500 shows — not because of some organic cultural consensus, but because those performed best on the feedback metric.

The old complaint about radio monoculture has an ironic successor. At least radio's homogenization was visible. Algorithmic monoculture hides behind the word "personalized."

Does anyone actually notice this happening? Have you caught yourself wondering whether something you love now is a genuine discovery — or just what the algorithm kept pushing until it wore down your resistance?

0 replies