As audio-focused generative tools and platforms have grown more sophisticated, the internet has filled with AI-generated music whose melodies and vocals are algorithmically derived from the work of human artists. Some of the people producing this content own up to using AI immediately. Others deny it until mounting public scrutiny forces them to tell the truth.
For musicians — especially those working in technologically focused spaces like electronic dance music — the question of what is "real" has grown steadily more complicated and more personal. Professional courtesy makes it harder still: it is one thing for a listener to call out a performer they suspect, and quite another for an artist to raise that suspicion about a peer.
Some do it anyway. This guide unpacks the resulting detective culture in six parts: what they look for, why they cannot prove it, and where this ends.
1. Who is doing the calling out
One of the people The Verge spoke to is Max "H4RRIS" Harris, a 26-year-old EDM producer who makes videos calling out tracks — and the people behind them — that he believes were made with AI.
Harris's framing is blunt: he sees AI-generated music as "a kind of decoy art form" and thinks the people making it have nothing meaningful to say. To him, true art in any medium is "created by people that are trying to express feelings and messages."
"I don't consider AI-generated material to be art, and I don't think this technology is really advancing art in any meaningful way," he says. "It's a technological advancement that's giving people a way to steal real art and pass it off as their own."
Another is Nihil Young, a 39-year-old Italian producer. Harris started making his videos because of what Young posted on Threads about people using Suno. Young does not usually get involved in online discourse, but felt the sheer volume of AI tracks flooding the scene made it necessary.
2. What they look for: the marks in the audio
This is the practical part. The tells producers describe come out of how the models work.
- A constant hiss. Harris describes a sharp hissing across the track. He attributes it to how the models operate: starting with a big block of white noise and then guessing at waveforms by referencing data taken from real songs.
- Layers that stutter in unison. In Suno-generated tracks, vocals and other melodic elements start stuttering at the exact same moment. The reason: models still struggle to fully separate the parts of the songs they were trained on, and treat those elements as one big instrument.
- Similar vocals. A vocal character that recurs across tracks and sounds a little too alike.
Why Harris finds these convincing matters: as a producer, all of it reads to him like "choices that a human just wouldn't make with their compositions" — because they do not make sense. The diagnosis comes less from the flaw itself than from the kind of flaw.
3. The visuals give clues too
Sometimes what gets caught is not the track but the video beside it. You may not hear the model behind an account's music, but you can clearly see a character's fingers vanishing in and out of existence. On other accounts, an overly glossy, uniform visual quality makes the content look like the familiar AI slop.
It is evidence that complements the uncertainty in the audio — but it is indirect. Someone whose visuals are generated may have made the music themselves.
4. Why dance music first
It is no accident that this argument erupted in electronic dance music first. The genre is entangled with technology from the start: production already happens in software, synthesised sound is already legitimate, and listeners carry no expectation of "real instruments." The line between a model's output and a human's is far blurrier here than on a track played on a guitar.
The second reason is the low barrier. Imitating an orchestral recording is hard; imitating a four-minute house track is easy. AI takes the most easily imitated format first — exactly as it did with short-form drama.
The third is the shape of the scene. Young's observation: over recent months several newcomers' stars rose after posting music he believes was generated with tools like Suno's. Recognition in dance music is largely online and algorithmic; an account does not need a label or a touring history to climb fast. That structure both opens doors for real talent and rewards imitation.
5. The proof problem
This is the crux. The producers behind the tracks Harris points to have not publicly commented on whether they use generative tools, and there is no definitive evidence that any of them was made with AI.
The Verge states it plainly: Harris could well be wrong about these songs. But his insistence reflects a broader culture of mistrust that has developed in response to the growing presence of AI content online.
That is the knot. The tells are real and diagnosable, yet none of them is proof on its own. A human-made track recorded on a bad microphone, over-compressed and mixed in a hurry, can hiss too.
So the person calling it out may be right and may be wrong — and either way it is the accused who pays.
6. The platform being blamed
Harris holds Suno responsible for the recent spike in this kind of music. But there is an interesting contradiction: the same platform's particular quirks are what taught him which tracks are worth calling out.
The model both enables the imitation and leaves the signature that gives it away. That signature is not permanent — as models improve, the stutter and the hiss will disappear. A detection method that works today may not work twelve months from now.
Harris also describes a blunter use: "People are just straight up uploading original, copyrighted tracks of artists like Madonna and asking AI to remix them." That is no longer a debate about style; it is a copyright matter.
7. Why it is so personal
Harris's account of his own process explains where the anger comes from. He always begins with a core concept, then keeps "throwing more ideas down until I find something that sticks." His setup runs on Ableton Live, MIDI controllers, analogue synthesisers and software samplers. But his description of the process is not technical: it is "really just me making creative decisions."
And then: "Each of those decisions — and there's hundreds of them to go into each song — brings me closer to evoking a specific kind of idea or emotion."
For Harris, being a real artist means being able to make experimental but informed decisions out of a studied understanding of a medium. That is where his objection sits: the technology often eliminates the user's need to think about how a song comes together at all.
8. What a listener can do
Everything above is the producers' view. For an ordinary listener the picture is simpler but more uncomfortable: you have no diagnostic tool, and you should not need one.
Three things still help. Look at the source: a live recording on the artist's own channel, studio footage, or an archive spanning years says more than any single track. Look at the pace: a new account shipping several tracks a week is not moving at a human tempo. Ask for a declaration: the answer to "did you use it" is itself information, and an evasive answer is an answer too.
But the main point is this: a listener is under no obligation to know how a track was made. That burden sits with the producer and the platform. Pushing detection onto the audience does not solve the problem, it transfers the responsibility.
What to watch
This argument will not stay inside music; the same pattern repeats in text, images and video. Three things will decide it.
First, the lifespan of the tells. Every flaw that is diagnosable today may be closed in the next model release. A defence built on catching things by ear is temporary by definition.
Second, disclosure. The durable answer is not catching but declaring. The difference between someone who says up front that they used a tool and someone who denies it until caught will decide where the culture settles. We apply the same principle on this site: every page states that it was produced with AI.
Third, the cost of a wrong accusation. In a field without proof, a callout culture hits innocent people too. Nobody has measured that cost and probably nobody will — but an artist's reputation outlives a post.