AI Music Slop: New Research Confirms Most Generated Tracks Are Ignored by Listeners

A major new study has delivered hard data on something many music fans and artists have long suspected: the vast majority of AI-generated music is simply not wanted by real listeners.

The paper, An Empirical Analysis of AI Slop in Music Streaming, published on arXiv in June 2026 by researchers from the University of Chicago, analyzed Spotify’s massive catalog and recommendation systems. Their findings paint a clear picture: AI music is growing explosively in volume, but it is overwhelmingly failing to connect with audiences.

---

Explosive Growth, Tiny Engagement

Using Spotify’s metadata covering hundreds of millions of tracks and a recommendation graph of 33 million tracks, the researchers found that AI-generated music now makes up 5.1% of Spotify’s entire catalog. In late 2025, AI tracks accounted for more than 40% of new weekly releases on the platform — up from just 1% in early 2024.

Yet this flood has produced almost no meaningful listener interest:

• 93% of AI-generated tracks receive fewer than 1,000 plays — the minimum threshold Spotify uses for monetization.
• Between January and May 2026, 92.7% of AI tracks received negligible total plays.
• Only 0.27% of AI tracks (1,632 in total) earned more than $1,000 in royalties during that period.
• AI music as a whole contributed just 0.3% of Spotify’s monthly royalty payouts.

In short, the overwhelming majority of AI tracks are effectively invisible to listeners and the recommendation algorithm.

---

The “Spray and Pray” Strategy

AI “artists” (or more accurately, operators) have adopted a classic low-effort, high-volume approach the researchers call “spray and pray.”

On average, these accounts release 27 tracks since 2024 — double the output of human musicians. They spread releases across many different genres in the hope that the algorithm will eventually surface one of them somewhere. The conversion rate is terrible, but the cost of generating and uploading another track is so close to zero that even rare, tiny successes can justify the entire operation.

This is not “democratization of creativity.” It is old-fashioned spam, now dressed up in the language of artistic empowerment.

Human-made music still performs dramatically better. Only 67.5% of human tracks receive negligible plays, compared to 92.7% for AI. When a human artist has a hit, their other tracks see a much stronger boost in plays than AI tracks do. Recommendation systems also show clear separation: AI tracks mostly recommend other AI tracks, while the majority of human tracks avoid recommending AI content.

---

Easy to Upload, Hard to Detect

The researchers didn’t just analyze existing data — they tested the system themselves. They generated tracks using popular tools (Suno, Udio, and others) and successfully uploaded them through 11 different indie music distributors. Policies on AI content were inconsistent and poorly enforced. In many cases, it was surprisingly easy to get mass-produced AI songs onto Spotify.

Current AI music detectors, meanwhile, proved unreliable, especially after simple audio manipulations like compression or pitch shifting.

---

Not Completely Useless — But Very Niche

There are narrow cases where AI music can still find an audience:

• Background and functional playlists (focus music, sleep sounds, workout mixes, lo-fi-style instrumental tracks).
• One-off or disposable tracks created for specific short-term needs (YouTube videos, social media content, ads, games).

In these low-attention contexts, listeners often don’t care (or even notice) whether the music is human-made. The bar for “good enough” is much lower.

But for active, intentional listening — discovering new artists, following genres, building personal libraries — the data is unambiguous. The public is not embracing AI music at scale.

Also read:

---

The Real Cost of “Free” Creativity

The core problem is economic. When generation and distribution costs approach zero, even a 1-in-10,000 success rate can be profitable. This creates a powerful incentive to flood the system with low-quality content, which in turn makes discovery harder for everyone — including the small number of genuinely good AI-assisted projects.

The University of Chicago researchers conclude that without meaningful intervention (better detection, clearer labeling, distributor policies, or changes to recommendation algorithms), AI music slop is on track to become a self-sustaining shadow industry — much like email spam or low-quality content farms in other domains.

The study doesn’t claim all AI music is bad. It simply shows that, at current scale and quality levels, the market has already voted with its ears: the vast majority of it is unwanted noise.

My initial concerns weren’t paranoia. The data has arrived, and it’s worse than many expected.