Spotify's Discovery Problem Isn't the Algorithm. It's the Incentives.

I've used Spotify for about eight years, five of them as a paying subscriber, so for this teardown I tried to stop looking at it like a fan and start looking at it like someone who has to defend its roadmap.

Here is the argument up front. Spotify has the best discovery engine in music streaming, but it is quietly working against itself. Over the past year the algorithm has grown more conservative, optimizing for retention by feeding people music they already know, which makes genuine discovery harder for emerging artists. Spotify's main answer to that problem, Discovery Mode, asks artists to give up roughly 30% of their royalties in exchange for algorithmic exposure, a trade that has drawn payola accusations, a class-action lawsuit, and condemnation from the Recording Academy. So the real opportunity is not a better ranking signal. It is an organic emerging-artist discovery surface that does not charge artists a royalty tax or compromise the listener's trust. The rest of this explains why.

Who Spotify is actually for

It is tempting to think of Spotify as a product built for listeners, but that is only half true, because it is really a two-sided marketplace where listeners are one customer and artists are the other. At the end of 2025 Spotify reached 751 million monthly active users and paid out more than 11 billion dollars to the music industry, with independent artists and labels accounting for half of all royalties. The artists are the supply that makes the whole thing work, and almost every discovery decision Spotify makes is really a balancing act between these two customers. Keep that tension in mind, because it is where the whole problem lives.

What works

The recommendation engine is the strongest part of the product, and it is not a close contest.

I wanted to see how much of that quality came from Spotify having eight years of my history to lean on, so I created a brand-new account with nothing attached to it. After only five or six hours of listening, the recommendations were already landing well, and when I skipped past a genre I did not want, the system backed off it quickly. That is a single user's experience rather than a benchmark, so I hold it loosely, but it points at something real. Starting from a blank account is a genuinely hard problem, the so-called cold start, where the system has almost no data to work with, and Spotify gets around it by analyzing the audio of the tracks themselves so it can start making sense of your taste before it really knows you. The engine is very good at reading what you like and giving you more of it.

That last phrase is exactly where the trouble starts.

The real problem is the incentive, not the ranking

When I first sketched this teardown, my instinct was that Spotify should rank emerging artists by engagement intensity, things like save rate and repeat-listens, rather than raw stream counts. Then I checked, and it turns out Spotify already does roughly this. Industry analysts tracking the 2026 algorithm note that it weights retention signals like saves and repeat-listens well above raw stream volume. So "use better signals" is not the insight. They have the signals.

The actual problem is what those signals are being optimized for. Over the past year the algorithm has become noticeably more conservative, prioritizing listener retention above almost everything else, because Spotify has found that people return more often and stay longer when they hear familiar music. The result is that surfaces like Autoplay and the AI DJ recycle tracks a listener already knows rather than introducing unsigned artists. The engine got better at keeping you, and worse at surprising you, and those two goals are in direct tension.

That tension produces two costs from the same root. On one side, emerging artists stay buried, because a system tuned to familiarity keeps routing attention back to the artists who already have it. On the other, long-time listeners quietly get bored, which is why a fair number of them complain their discovery has gone stale and some go as far as starting fresh accounts to break the loop. Popularity bias is not one problem but two wearing the same coat.

Spotify's current answer, and why it backfires

Spotify is aware of the emerging-artist problem, and its main fix is Discovery Mode, which lets artists and labels flag tracks for algorithmic priority in exchange for accepting a royalty cut of around 30%. On paper it is a discovery tool. In practice it has become the most controversial thing the company has shipped in years.

A class-action lawsuit filed in November 2025 alleges Discovery Mode is a modern form of payola, arguing that Spotify markets its recommendations as organic and personalized while quietly selling placement for reduced royalties. The Recording Academy has condemned the practice in the same terms, arguing it raises the barrier of entry for smaller artists and deepens Spotify's extractive relationship with creators. Spotify rejects the characterization, calling it nonsense and stressing that Discovery Mode is optional and is not used in flagship surfaces like Discover Weekly or the AI DJ. The lawsuit itself was sent to arbitration in May 2026, so this is contested rather than settled.

But the strategic point holds regardless of how the litigation ends. Spotify's answer to weak organic discovery is a feature that asks the artists who can least afford it to pay for visibility out of their royalties, and that answer is now a legal and reputational liability. That is the gap worth building in.

What I would do about it

This is a direction I would test, not ship on conviction.

I would build an organic emerging-artist discovery surface that leans on Spotify's strongest asset, its curated and algorithmic playlists, with two hard constraints that separate it from Discovery Mode. It takes no royalty cut, and it involves no pay-for-placement, so nothing about it can be characterized as payola. The selection is purely behavioral.

The signal would be engagement intensity rather than raw volume, but applied specifically to find artists the main algorithm is structurally missing. I would look for the smaller artists whose listeners keep coming back, high repeat-listens and low skip rates inside a committed audience, because that pattern suggests the music is genuinely landing and the only thing missing is reach. A small artist with scattered, high-skip listening would be left alone, because low intensity suggests the market has already made its judgment. The point is to catch the artists whose intensity is high but whose volume is too low for the retention-tuned main engine to ever surface them.

I keep coming back to running into Sleep Token in the early 2020s, when they had five-digit listener numbers, and then watching them blow up a few years later. The signal that they were special, a small but intense listener base, was already in the data well before they broke out, and a surface tuned to intensity over volume would have found them sooner. Plenty of equally good artists never get that break at all, and that gap is the opportunity.

What makes this worth doing is that it serves both sides of the marketplace at once. It gives emerging artists a path to discovery that does not cost them a third of their royalties, and it gives stuck listeners something genuinely new inside the lanes they already like. It will not fully cure the staleness problem, which is broader than any single surface, but it addresses it without the legal and ethical baggage Discovery Mode carries.

Why this is Spotify's problem to win, specifically

Apple Music and YouTube are not in the same bind, and that is what makes this strategic rather than charitable. Spotify's entire identity and its core advantage is discovery and personalization. Apple Music leans on its hardware ecosystem and human curation, and YouTube wins on video and sheer catalog. Spotify wins, when it wins, because it surfaces the right thing at the right moment. So an algorithm that has grown too conservative to do genuine discovery is not a minor product gap for Spotify, it is an erosion of the one thing that differentiates it. And because independent artists already make up half of Spotify's royalty payouts, the long tail is not a rounding error in its catalog. It is half the supply. Protecting and growing organic discovery for those artists defends the moat and the supply base at the same time, which is exactly the kind of thing that should beat a feature nobody is suing over.

The honest tradeoff

A surface that pushes less-familiar music is a riskier engagement bet than one that feeds people their favorites, so it will likely show lower immediate engagement than the core recommendations do. That is why it has to be additive, either opt-in or clearly framed as a discovery lane, so the main retention loop stays protected while the long tail gets served. I will admit this creates a real tension I have not fully resolved, because making it opt-in protects the core but also caps how many listeners the feature reaches, which limits its impact on the artist problem I care about. Where exactly to sit on that line is something I would want real experiment data to answer rather than guess at.

Risks I considered

The clearest one is gaming. If a surface rewards high repeat-listens and low skips from a committed audience, a small artist could try to fake that intensity with a handful of bot accounts, and because you need fewer fake listeners to fake intensity than to fake raw popularity, this signal could be easier to manipulate than stream counts. Spotify already detects and removes fraudulent streams, so this surface would need to sit behind that existing fraud detection rather than trusting the intensity signal on its own. I would treat fraud resistance as a launch requirement, not an afterthought.

How I would know it worked

I would not measure success by whether small artists simply got more streams, since I could manufacture that just by pushing them harder. I would measure whether the discovery was actually good, watching the save rate and repeat-listen rate on the artists surfaced to see whether people genuinely liked them or just heard them once. I would compare the skip rate on this surface against normal discovery to confirm I was not degrading the experience in the name of fairness. Over a longer horizon I would watch whether more small artists were steadily crossing into wider audiences, which would tell me the discovery-to-breakout pipeline was actually widening. The guardrail through all of this would be total session listening time and retention, which cannot be allowed to drop, and if they did, the experiment would be hurting the core and I would pull it. I would run the whole thing as an opt-in experiment against a holdout before putting any serious roadmap behind it.