The internet is filling up with music nobody really made

The scariest thing about AI music is not that it might make something brilliant. That would at least be interesting. The more depressing possibility is that it makes millions of songs that are just fine: clean enough to pass, empty enough to forget, cheap enough to flood every corner of the internet and convincing enough that nobody notices how little of it needed to exist.

This is the problem with almost-music. Not bad music, because bad music can still have a pulse, a person, a wrong decision, an ego, a stupid lyric, a strange accident, or the kind of failure that makes it human. Almost-music is worse because it has no real risk in it. It sounds like something you have heard before because that is basically the point. It arrives polished, harmless and emotionally unfurnished, as if a playlist brief had been put through a blender with stock photography and a minor chord.

The music industry has already been trying to put numbers around the dread. APRA AMCOS’ AI and Music report found that by 2028, 23 percent of music creators’ revenues could be at risk due to generative AI, with an estimated cumulative damage of NZ$572 million across Australia and New Zealand. The same research found that 54 percent of surveyed songwriters and composers agree AI can assist the human creative process, while 82 percent are concerned AI could stop them making a living from their work.

That contradiction is the whole story. Musicians are not simply scared of a tool because it is new. Plenty of artists have always used tools, cheats, machines, samples, plug-ins, loops, presets, borrowed sounds and whatever else helped get the thing out of their head and into the room. The fear is not that AI can help make music. The fear is that AI can help make too much music, too quickly, too cheaply, with too little care, and then make actual artists compete with the synthetic landfill their own work may have helped train.

The word “slop” sounds ugly because it is meant to. A recent research paper on AI slop in music streaming looked at how AI-generated tracks move through streaming platforms and found that 93 percent of AI music receives few, if any, listener plays and is rarely recommended. On one level, that sounds reassuring. Most of the slop is not winning. But on another level, it proves the point: the business model does not need every track to matter. It just needs the cost of making and uploading them to become low enough that spraying the system with almost-music starts to look rational.

That is where the music conversation gets less romantic and more grim. Streaming already made music feel abundant to the point of invisibility. AI just makes the abundance cheaper. If anyone can generate, package and distribute hundreds of tracks under fake names, across fake genres, for fake moods and algorithmic niches, the issue is no longer whether the songs are good. The issue is whether the system can tell the difference between culture and content pretending to have a pulse.

It is easy to mock this stuff until you remember how much music is already treated like wallpaper. Study beats, sleep sounds, lo-fi café loops, focus playlists, corporate ambience, gym music, fake jazz, fake folk, fake cinematic tension for fake productivity. A lot of the market does not ask music to mean anything. It asks music to sit politely in the background while someone works, buys, scrolls, stretches, sleeps or pretends to be the kind of person who wakes up calmly. AI is extremely well suited to that kind of nothing.

That does not mean background music is worthless. Some of it is beautiful, and a lot of working musicians have made real craft out of music that does not need to dominate a room. But there is a difference between music designed with restraint and music generated because silence has not yet been monetised properly. One has authorship. The other has optimisation.

This is where trust starts to matter. When a song appears on a platform now, what exactly is the listener being asked to believe? That someone made it, chose it, recorded it, shaped it, meant something by it, or at least stood behind it as a piece of work? Or that it was generated, named, uploaded and distributed by someone running a catalogue strategy with all the warmth of a parking meter? Music has always had commercial machinery around it, but the machinery used to need a person somewhere near the centre. AI makes it easier to build the machinery and skip the person.

The real artists will not vanish, obviously. People will still want voices with bodies behind them, live shows with sweat in them, songs attached to stories, scenes with texture, records that arrive with a sense of life around them. The human part of music is stubborn because music is social before it is technical. A machine can generate a track, but it cannot make a crowd believe they were there before everyone else, or turn a bedroom song into a scene, or carry the weird emotional authority of someone singing a line that cost them something.

But discovery could get uglier. That is the real risk for emerging artists, especially in smaller markets like New Zealand where the path from bedroom to audience is already narrow. If platforms are full of synthetic almost-music, then getting heard becomes less about being good and more about surviving the sludge. The problem is not that AI music will replace Lorde, or whoever the next obvious example is. The problem is that it may make it harder for the next interesting person to be found before the system has buried them under a thousand tracks called “Midnight Focus Bloom.”

This is why disclosure, rights and platform responsibility cannot be treated like boring admin. If AI music is going to exist at scale, listeners should know what they are listening to, artists should know whether their work helped train it, and platforms should not be allowed to shrug while synthetic catalogues quietly distort royalties, discovery and trust. The industry has spent years telling artists to adapt, but adaptation is not the same as accepting a rigged flood.

There is a useful version of AI in music, and pretending otherwise just makes the argument weaker. A producer using AI to test an idea, an artist using tools to sketch a sound, a small team using technology to move faster without losing their point of view; none of that is the enemy. The enemy is the empty catalogue with no author, no audience and no reason to exist beyond extracting pennies from a system that already undervalues the people who give it meaning.

The internet is filling up with music nobody really made, and the worst part is not that it sounds terrible. The worst part is that some of it sounds good enough. Good enough for the playlist, good enough for the café, good enough for the ad, good enough for the platform, good enough to blur the line until listeners stop asking who made the thing and artists stop trusting the room they are releasing into.

Music does not need to be pure to matter. It does need someone behind it who cares whether it exists.