I watched a guy last night rail against AI slop in music for twenty minutes. Suno and DistroKid took most of the heat — biggest platforms, biggest targets. His argument was that they should do better.
I think he’s aiming at the wrong people. Quality has always been the operator’s responsibility. If you make crap, the tool didn’t make it crap. You did.
But he’s right that there’s a problem, and I’ve noticed the tell.
The scream that doesn’t belong
Listen to a batch of AI-generated rock tracks and you’ll hear it. The vocalist opens with this big, throaty scream — every time — whether the song earns it or not. It’s out of place. It’s there because the model learned that rock songs have screams, not because anything in the song called for one.
That’s the whole problem in one sound. The machine has the vocabulary of the genre without any of the reasons behind it. It knows what rock looks like. It has no idea why.
And that’s not a music problem. It’s an everything problem.
The rhyme that broke the meaning
I write songs. I use AI to work on them. And the same thing happens to me in a different costume.
I’ll put a lyric in for review and it comes back with suggestions that make no stylistic sense — some weird rhyming component wedged in because the model learned that songs rhyme. And I’ve had to push back hard: it’s fine if this line doesn’t rhyme. What matters is that it means what I meant.
Same false scream. The model was optimizing for the form, because the form is the only thing it can see.
The line I hold
I bring the ideas. I bring the lyrics. I start assembling them. Then I use AI to pressure-test — does this part work, where’s it unclear, where did I get ambiguous.
It comes back with a full pass. Usually about ninety percent of what I wrote, ten percent added or changed. Then I go back through and make the final cut, trimming everything to exactly how I want it.
Then they’re mine.
The rule underneath that: AI never decides what the song is about. It only helps me find out whether I actually said it.
What you lose if you skip the last pass
If you take the AI’s version and let it stand, it can quietly change the meaning of the entire song. It’ll come back sounding fine — maybe technically better — and it’ll be about something slightly different than what you set out to say. You won’t notice unless you go looking.
That’s meaning drift. And it works the same whether you’re writing a song, a sales email, or a piece of code. The output looks finished. It just isn’t yours anymore.
I don’t want Suno filtering on quality — that just gives you whatever their filter thinks is good, which is its own flavor of slop. I don’t want DistroKid refereeing. They’re a pipe, and they’re honest about it.
What I want is more people creating aggressively with AI as a partner, and fewer people handing it the wheel. Some of the most interesting artists on Suno right now are inviting remixes, pulling people into the work. That’s the old folk tradition wearing new clothes. Nobody called it slop when songs got passed hand to hand and changed along the way.
The tool isn’t the problem. The problem is when you stop being the one who knows why.
Everything else is just a scream at the start of a song that didn’t need one.
