AI Is Making Us All Think the Same — And the Data Proves It

When Everyone’s Co-Pilot Flies the Same Route

Picture a jazz session where every musician secretly studied the same one playlist. The solos might be technically impressive. But after a while, everything starts to sound like the same song.

That’s essentially what new research is telling us about generative AI — the technology behind ChatGPT, Midjourney, and the dozens of tools we now use to write, design, and ideate. And the findings aren’t just theoretical. The homogenization of human creative thought is already happening in the real world.

The Science of Sounding Alike

Researchers Alwin de Rooij (Tilburg University and Avans University of Applied Sciences) and Michael Mose Biskjaer (Aarhus University) conducted a systematic review and meta-analysis — a statistical method that pools findings from multiple independent studies to identify larger patterns — of 19 empirical studies published between 2022 and early 2026.

That window matters. It covers the post-ChatGPT era, when generative AI tools went from niche curiosity to everyday utility for millions of people.

What they found was clear and measurable: when people use AI to help them create, their output becomes more similar to other people’s output. Not slightly similar. Statistically, significantly similar.

The technical term is homogenization — a reduction in the diversity of ideas, designs, and written work across a group of users. Think of it like adding the same seasoning to a thousand different dishes. Each one might taste better individually. But the kitchen loses its flavor profile.

Why AI Pulls Everyone Toward the Same Ideas

Here’s where the mechanism gets important. Large language models (LLMs) — the AI systems powering tools like ChatGPT — are trained on billions of sentences scraped from the internet. They learn to predict what words naturally follow other words, based on patterns in that massive dataset.

Because most of the major tools are trained on heavily overlapping data, they tend to gravitate toward the same statistically common associations. Ask ten thousand people to brainstorm using the same AI, and the AI essentially becomes a semantic anchor — pulling everyone toward the same conceptual territory.

The researchers broke down the effect by task type:

• Divergent thinking tasks (open-ended, like “list unusual uses for a brick”) — lower homogenization
• Idea generation tasks (solution-focused, like “improve public transit”) — highest homogenization
• Creative writing and visual art — significant but moderate convergence

When the task is specific and problem-driven, people lean harder on AI suggestions. And AI suggestions, by design, skew toward the probable. The predictable. The average.

This Isn’t Just a Lab Problem

Skeptics might say: sure, controlled lab experiments show convergence, but what about the real world? The researchers checked. They analyzed quasi-experiments — real-world comparisons of creative output before and after AI adoption, including published essays and visual art. The results held. Diversity of ideas declined.

A parallel study published in the journal PNAS Nexus reinforced the finding. Researchers Emily Wenger and Yoed N. Kenett tested 22 commercial chatbots alongside 102 human participants on verbal creativity tasks including the alternative uses task — a classic divergent thinking exercise where you generate as many creative uses for a common object as possible.

Individual chatbots performed at or above average human levels. Impressive on paper. But when you laid all the AI responses side by side, they were strikingly similar to each other — far more homogeneous than the responses from the human group. The researchers tried adjusting internal randomness settings to force more variety. The result? Incoherent gibberish. The diversity wasn’t accessible without breaking the model.

What This Means for Culture — Not Just Productivity

Here’s the turn that matters for anyone who cares about creative culture: de Rooij and Biskjaer also found preliminary evidence that the homogenization effect carries over after people stop using AI. Interact with these systems, and they may shape how you generate ideas even when the tool is closed.

The researchers compare this to fixation effects in psychology — the well-documented tendency for people to get anchored to examples they’ve seen, limiting their subsequent thinking. Except now the anchor is being dropped simultaneously for millions of people at once. That’s not a bug. That might be a structural feature of how these systems currently work at scale.

Think about what that means for journalism, advertising, product design, music, film, education. Every creative field where AI is being adopted as a co-creator. The individual outputs might improve. The collective creative ecosystem might narrow.

The Takeaway That Deserves More Than a Shrug

The researchers are careful to note what they’re not saying. AI doesn’t eliminate human creativity. Individual quality can rise even as collective diversity falls. These effects are often subtle in single instances.

But subtle at scale is still significant. And the scale here is civilizational.

The question isn’t whether to use these tools — that ship has sailed. The question is how we design workflows, education systems, and creative cultures that preserve the cognitive diversity that makes innovation possible in the first place. Diversity of thought isn’t just an aesthetic preference. It’s how problems get solved in unexpected ways. It’s how cultures stay alive.

The jazz session analogy holds. The musicians aren’t getting worse. But if they all keep practicing from the same AI-generated sheet music, eventually nobody’s playing anything the world hasn’t already heard.

And the music we haven’t heard yet? That’s the stuff worth protecting.