"Everything made with AI looks the same" has been a design-industry complaint for a while now — usually dismissed as an aesthetic gripe, or a phase the tools would grow out of. A 2026 academic meta-analysis suggests it's neither. It's a measurable, statistically significant effect, and it doesn't go away as the tools improve.
The homogenization problem is no longer just a vibe
Researchers Alwin de Rooij (Tilburg University and Avans University of Applied Sciences) and Michael Mose Biskjaer (Aarhus University) pooled 19 empirical studies and 61 effect sizes on human-AI co-creation, published between 2022 and early 2026, spanning ideas, visual designs, and creative writing. Their finding: a statistically significant homogenization effect associated with AI use. Their own summary is blunt — "using generative AI can indeed lead people to think alike."
It gets more specific than a general vibe, too. The effect was strongest in structured idea-generation tasks, and it didn't disappear once the AI was put away — people who'd just worked with AI carried measurably less diverse thinking into a following creative task, even without AI in the room for that second task.
Why it happens
This isn't really about any one tool being bad. It's closer to plain statistics: most generative AI tools draw on overlapping training data and similar underlying models, so independent people prompting independently still tend to converge toward the same statistical center of what those models consider "good." Individually, each output can look genuinely impressive. Collectively, across a market where everyone's reaching for the same handful of tools, the range of what gets made narrows.
The industry's own answer: taste is the scarce thing now
This is exactly the shift the AI industry itself is naming, just from the production side rather than the research side. Cat Wu, Head of Product at Anthropic, put it plainly about a different craft with the same logic: "As code becomes much cheaper to write, the thing that becomes more valuable is deciding what to write." Swap "code" for "design options" and the sentence holds up perfectly — when producing variations gets nearly free, the scarce skill stops being the producing and becomes the choosing.
Google Cloud's "Future of AI: Generative Media" report reaches the same conclusion from the enterprise side, naming creative taste, lived experience, and authentic storytelling as the durable human skills left once raw production is automated — the parts of the job that don't get cheaper just because generation does.
Using generative AI can indeed lead people to think alike. The premium doesn't disappear when production gets cheap — it just moves, from making options to choosing the right one.
What "good taste" actually means in a design process
It's worth being concrete about this, because "taste" can sound like a vague aesthetic gift rather than a real, learnable skill. In practice it's a set of specific judgment calls: which of twenty AI-generated directions actually fits this brand's personality rather than a generic "good design" average; what a brief left unsaid that only becomes obvious once you see the wrong answer; when an output is technically correct — on-brief, on-brand-guideline — but creatively flat anyway. None of that is guesswork. It's pattern-matching built from having made that call, and lived with the consequences of getting it wrong, across a lot of real projects.
What this means in practice
The homogenization finding isn't an argument against using AI in design — the speed and quality gains are real and well documented elsewhere. It's an argument for exactly one thing: a design process where AI generates the range of options fast, and a human with real, earned judgment reviews, chooses, and refines before anything ships. Not as a value-added extra layered on top of AI output, but as the specific step that answers the specific risk the research actually measured. Skip that step, and "AI-assisted" quietly becomes "AI-average" — competent, on-brief, and indistinguishable from what every other business using the same tools is also shipping.
See where the human judgment sits in our process
Every Aoibh project is drafted with AI and chosen, refined, and signed off by a human art director before it ships.
See how it works