Photo Credit: Pawel Czerwinski

Spotify has a big AI music slop problem. Third-party tracking databases have stepped in to fill the void caused by the platform’s longtime inaction.

It’s no secret that Spotify has a big AI music problem. Theirs is arguably worse than its rivals, because the platform hasn’t tracked and labeled AI-generated tracks like other streamers such as Deezer or even YouTube. That’s led to a lot of confusion among listeners and frustration among artists. Now, third-party tracking websites are trying to fill the void created through Spotify’s longtime inaction.

One such tracker, SoullessMusic.com, is a self-described database for “AI artists hiding on Spotify. No bands, no studios, no soul, just machines and melody.” There, such viral sensations as AI-assisted R&B act Slime Dot are identified as “almost certainly AI” based on such analysis as a flood of tracks in a short period.

“I didn’t know what was AI and what’s not,” said Graeme Fulton, creator of SoullessMusic, in a statement to 404 Media. “And then I’m on Instagram Reels and I see loads of artists unhappy, saying some artist has popped up with their song and just copied it. And there’s loads of different cases where [AI artists] would copy an artist’s look and appearance, […] sort of just stealing from artists.”

There’s also SlopTracker, where users can copy/paste links to Spotify tracks to run them through a tool that determines, with varying degrees of confidence, whether a song was generated almost entirely with AI. Like SoullessMusic, SlopTracker also found that Slime Dot’s hit “Fully” was “95% likely” to be generated by Suno.

But even relying on third-parties to identify AI and AI-assisted tracks is not infallible, since some AI and AI-assisted artists may take issue with being included on sites like SlopTracker and SoullessMusic. SlopTracker previously identified the openly AI-assisted project Georgia Phantom as likely AI, but that project notably no longer appear on the site after a cease-and-desist sent to SlopTracker.

The person behind the Georgia Phantom project also sent a similar request to Digital Music News for our previous coverage of SlopTracker back in March. At the time, Georgia Phantom was listed in the top ten AI and AI-assisted artists who are draining the Spotify royalty pool. While SlopTracker responded to the legal notice by removing Georgia Phantom from the site “due to an abundance of caution,” SoullessMusic has not. It’s worth pointing out that SoullessMusic still lists Georgia Phantom among their database and identifies them as “almost certainly AI” at 99%.

Photo Credit: Digital Music News

Whether artists choose to use generative AI platforms like Suno or Udio isn’t at issue here—it’s the lack of transparency around it that leads to platforms like Spotify hosting music that has not been properly identified. Listeners deserve to know what they’re listening to, just as artists deserve to not have to worry about an AI-assisted, pitch-shifted version of their song stealing their streaming royalties.

That’s where tracking databases come in. It’s not a flawless process; as 404 Media points out, both SoullessMusic and SlopTracker use AI and other automated tools to detect AI music. Using AI detectors to detect AI content is an inherently flawed process and can lead to false positives.

However, Spotify’s lack of transparency surrounding the AI-generated music on its platform has made it impossible to track just how much money it’s making—and how much money artists are losing—on the platform. Last year, Spotify announced it would add AI disclosures to AI-generated music, but that appears to still be in the testing phases and has not widely rolled out.

That’s particularly embarrassing for the streamer, given that both Udio and Suno, arguably the biggest names on the scene, include digital watermarking in their generation. This makes it theoretically quite simple to identify and tag.

As AI-generated music becomes more common, platforms and labels are investing in ways to tell it apart from human-made tracks. The goal here isn’t just “does this track sound synthetic?” but “can we trace how and possibly from what source” was the track generated. Both SlopTracker and Soulless Music rely on these processes to identify whether a track is solely AI-generated or AI-assisted with human input.

Three key ideas power modern AI music detection including vocoders and their artifacts; fakeprints, which are compact spectral signatures of AI generation; and ‘neural fingerprinting’ for copyright and derivative work detection.

Most advanced text-to-music systems like Suno, Udio, MusicGen, and other similar tools don’t output raw audio directly from a big language or diffusion model. Instead, they generate an internal representation and then pass that representation through a neural vocoder or neural audio codec to reconstruct a waveform at full sample rate.

The vocoders are powerful, but not perfect. They rely on deconvolution and upsampling layers to go from low-rate latents back to high-rate audio. These design choices leave behind predictable, model-specific fingerprints in the frequency such as small regular peaks and ripples in the spectrum caused by deconvolution stride patterns or subtle ‘grid-like’ patterns that don’t typically appear in human recordings and mixes. Research has shown that these vocoder fingerprints are strong enough to distinguish AI from human audio and to identify which vocoder model generated a clip.

If vocoder artifacts are the physical marks of AI-generated music, then a fakeprint is a distilled, machine-readable summary of those marks. A typical fakeprint pipeline works like this: compute an average spectrum over the track, subtract the lower envelope of that spectrum, then focus on the 1-8kHz band, where neural vocoder artifacts tend to be the most prominent. Encode this processed spectrum as a feature vector. That fakeprint is fed into a classifier that has been trained on real human music vs. AI generated music from specific platforms. The resulting score indicates how likely the track is to be AI-generated, sometimes broken down by suspected generator family.

While vocoder-based detection asks “was this made by an AI model?” neural fingerprinting asks a different question: “does this track contain musical material derived from specific human works?” This concept is central to copyright enforcement as labels like UMG and Sony deploy the technology.

A modern AI music detection system may combine several signals including vocoder/fakeprint analysis, neural fingerprinting matching, and statistical and structural analysis. The output is rarely a simple yes or no question to “is this track AI?” Instead, the platforms issue a confidence score from 0.0 = very likely human to 1.0 = very likely AI. For AI-assisted tracks, the breakdown may appear something like 0.6, which means 60% AI generated, 40% human.

Of course, no detection mode is 100% infallible and the pace of AI development means that detection of AI-generated tracks is an arms race. As AI music generators adopt new architecture or vocoders, old detectors may lose accuracy and need retraining. Heavy remastering, extreme EQ, and multi-stage laundering of AI-generated tracks can also weaken fingerprint signals, though robust trackers are trained to handle these common manipulation tracks.

Perhaps the biggest risk is mislabeling human-made music as AI, which can harm artists. Designing reliable thresholds for detecting AI-generated music remains an active research challenge as AI models continuously evolve. At the same time, neural fingerprinting for copyright detection is becoming more sophisticated using music foundation models (like MuQ, MERT) to improve the robustness against small-scale alterations like tempo changes, pitch shifts, compression, and other real-world distortions. But Spotify is left playing catch-up because unlike competitors, it doesn’t explicitly label AI or AI-assisted music.

“We’re employing a layered approach that combines enforcement, artist controls, and greater transparency,” Spotify told 404 Media. “Over the past year, we’ve introduced policies targeting harmful AI-related behaviors like spam, impersonation, and deceptive content, alongside new tools that give artists more control and listeners more context.”

“These include Verified by Spotify, which helps listeners identify authentic artist profiles; AI Credits, where artists disclose when and how AI was used in creating their music […]; and Artist Profile Protection, which gives artists more control over what appears on their profiles. We are continuing to build on these efforts.”

But Artist Profile Protection requires artists to opt-in. AI Credits requires artists to disclose their AI use. Verified by Spotify is, at least seemingly, beholden to a fairly rigorous screening process.

It reads like taking too little action a little too late. Earlier this month, Spotify competitor Deezer, a pioneer in detecting and labeling AI-generated content, said that nearly half of all new music uploaded to its platform is now AI-generated. That number is only increasing, which has prompted Deezer to begin de-monetizing certain tracks.

SlopTracker says that AI-generated music on Spotify that is featured on official Spotify-curated playlists is “draining” $0.1188 per second from real artists. That number is similarly expected to increase.