The Sameness Engine

A meta-analysis published this year reviewed 19 empirical studies and found what the researchers politely call "homogenization effects" in human-AI co-creation. When people use generative AI to write, brainstorm, or make art, their individual performance often improves — but the spread of ideas across the population narrows. You produce more. Everyone produces the same more.

The mechanism isn't hard to follow. The major AI models are trained on massive, overlapping datasets scraped from the internet. They naturally gravitate toward the most statistically common word associations — the conceptual center of mass of everything humans have written and clicked on. When you prompt them, you're consulting the aggregate. The tool doesn't amplify your specific angle; it anchors you toward the mean. You asked for help and got the average.

That would be uncomfortable enough. Then comes the detail that makes it worse: the homogenization persists after the AI interaction ends. Not just while you're using the tool — the effects carry over into subsequent creative tasks. You don't put down the anchor and return to your own orbit. You've moved.

This is what a convergence engine actually does. Not the dramatic version where robots replace you — the quiet version where you start sounding like everyone else and you're the last to notice, because your output feels more polished than ever.

There's a performance happening here. It looks like "using AI as a tool." It feels like efficiency, like getting out of your own way, like removing friction between the idea and the execution. The self-story runs: I'm still the creative one, I'm just removing the grunt work. That story is load-bearing. Notice how hard you'd push back if someone suggested the ideas themselves are changing.

Tools multiply what exists — they don't replace the source, but they scale it. The problem is that everyone picked up the same instrument, tuned to the same frequency, with the factory settings left on. That's not amplification. That's unison. And unison isn't harmony — it's the loss of the distinct voices that give a murmuration its motion.

The researchers found the strongest homogenization effects in constrained idea-generation tasks, weaker in open-ended creative work. Which is to say: when there's a target to hit, AI steers everyone toward the same target. The more defined the problem, the more everyone's solution starts to rhyme.

You probably didn't set out to sound like everyone else. You set out to work faster, to get unstuck, to get something on the page. Those are honest intentions. The outcome is a different question.

The uncomfortable part isn't that the AI is doing something to you. It's that it's doing something to all of you at once, invisibly, while each individual experience is one of expanded capability. The ecosystem contracts while each organism reports flourishing. That's a pattern worth sitting with, even if you can't do anything about it before your next deadline.

The truth you're avoiding isn't going to age into something easier.

i · sources

• Real-world evidence shows generative AI is making human creative output more uniform — PsyPost, 2026-05-16