Why your Discover Weekly stopped surprising you
July 16, 2026 · 6 min read
There's a specific disappointment in opening a playlist built just for you and recognizing every name on it. Not liking it — recognizing it. The machine has read you back to yourself, and it turns out you're a bit predictable.
If that's where you are, here's the reassuring part: your Discover Weekly isn't broken. It's doing precisely what it was designed to do. Understanding why is the first step to getting out.
The loop that closes around you
Recommendation engines learn from behavior. You listen to something, that signal goes in, and the next batch leans a little further in that direction. Listen again, and it leans further still. Each round is a small, sensible step — and after enough of them, the system has quietly drawn a fence around your taste and started mowing the same lawn.
The mechanism isn't malicious. It's optimizing for a very reasonable goal: don't waste your time. A recommendation you'll probably enjoy is a safer bet than one you might not. But "probably enjoy" and "might change your life" are different targets, and only one of them is being aimed at.
This isn't a fringe complaint. Spend five minutes in Spotify's own community forums and you'll find people describing the identical experience — the algorithm fixating on one genre and staying there for weeks, even as they skip, dislike, and actively listen to other things. The common advice is to keep feeding it better signals and wait it out. Which works, sort of, eventually. But notice what's being asked of you: you are supposed to do the work of managing the algorithm that was supposed to do the work for you.
Why "just listen to more stuff" is harder than it sounds
The standard fix is to diversify your listening so the system learns you're broader than it thinks. Sound advice, with two catches.
First, there's lag. It commonly takes one to three weeks for a meaningful change in your habits to show up in your personalized playlists. That's a long feedback loop for something you're trying to steer.
Second — and this is the real trap — to listen to something new, you first have to find it. And the tool you'd normally use to find it is the very thing that's stuck. You're asked to escape a room using a key that's inside the room.
Five ways to break the loop
Practical things that actually work, roughly in order of effort:
1. Seed it deliberately
Don't just passively skip what you dislike — actively save music that's off your usual axis. Liked Songs and saved artists are heavier signals than a passive play. If you want the system to believe you're wider than it thinks, give it evidence, not just objections.
2. Then actually wait
Give it those one to three weeks before you judge whether it worked. Most people change their behavior for four days, see no difference, and conclude it's hopeless. The lag is real; plan for it.
3. Get out of the recommendation lane
Artist radio and playlist radio pull from a different well than your personalized mixes. So does simply playing an album by someone you've never heard, all the way through. Anything that isn't the algorithm's main highway is worth a detour.
4. Let humans back in
Critics, blogs, Bandcamp Daily, a subreddit, a friend with aggressively specific taste — these sources have something no engine has: they'll recommend something because it's good, not because it's similar. Human curation is the oldest discovery technology and it still outperforms on surprise.
5. Change the unit from song to album
This is the underrated one. Most recommendation feeds deal in tracks, and tracks are frictionless — they slide past. An album asks for forty minutes and gives you a shape in return: a sequence, an arc, a mood that develops. You remember albums. You rarely remember the 300 tracks that autoplayed at you last month.
What we'd argue is the actual problem
It's tempting to blame "the algorithm," but that's lazy — we build with algorithms too, and they're genuinely good at reading taste. The problem isn't that software is doing the recommending. It's what the software is pointed at.
Most discovery is tuned for engagement: keep the session going, minimize skips, serve the safe adjacent thing. Optimize that hard enough for long enough and you get exactly what you're experiencing — a comfortable, tightening circle. Nobody chose that outcome. It's just what happens when the goal is "don't get skipped" instead of "find something worth keeping."
B-Side is our answer to this. Once a week it hands you five albums you don't already know — pulled from the edges of your real listening on Spotify, Apple Music, or Last.fm.
Three things we do on purpose:
- Albums, not tracks. The unit is the record, because that's the thing you actually remember.
- It never repeats. Once an album has been on your radar, it's spent. The pool has to keep moving outward — the loop can't close.
- It tells you why. Every pick comes with the reason it's there and how it connects to your taste. No black box, no shrug.
Five a week. Small enough to actually listen to, which is rather the point.
Try B-Side freeThe good news
A closing circle isn't evidence that you've heard everything worth hearing. It's evidence of a system doing its job a little too well. The catalogue didn't shrink — the window onto it did.
Whether you fix that with our app, a good critic, a stubborn friend, or just the discipline of playing one unfamiliar record all the way through this week: widen the window. Your favorite music is still out there. It just isn't going to come find you.