Ask most writers or designers what scares them about AI, and creativity is usually the first word out of their mouth. Not their job title, not their income, their creativity, the one thing they assumed a machine couldn’t touch. A new study out of the Université de Montréal just complicated that assumption, and it did it with a sample size too large to wave away.
Researchers led by psychology professor Karim Jerbi, working alongside AI pioneer Yoshua Bengio and teams from Concordia, the University of Toronto Mississauga, Mila, and Google DeepMind, ran the largest head-to-head creativity comparison ever attempted: over 100,000 humans against today’s leading language models. The results, published in Scientific Reports, landed on a genuinely uncomfortable finding. On a standardized measure of original thinking called the Divergent Association Task, models like GPT-4 didn’t just compete with people. They beat the average human outright. This AI creativity study is the one everyone in the content world should actually read, not just skim in a newsletter.
What the AI Creativity Study Actually Found
The Divergent Association Task, or DAT, asks a person (or a model) to name ten words that are as unrelated to each other as possible. It sounds simple, almost like a party game, but it’s a well-validated proxy for divergent thinking, the cognitive skill behind brainstorming, ideation, and the kind of lateral leaps that make an ad campaign or a blog headline actually land. Score higher, and you’re demonstrating a broader, less predictable associative network in your thinking.
When the researchers ran this test at scale, GPT-4 landed above the human average. Not by a landslide, but clearly and consistently. That’s the headline. The footnote, and it’s the part I think matters more, is this: the top 10 percent of human performers left the AI models well behind, especially once the task moved from single-word association toward richer creative work like poetry, storytelling, or anything requiring emotional or narrative coherence.
So the honest read isn’t “AI is more creative than people.” It’s “AI is now more creative than the average, unpracticed person doing a narrow creative task,” which is a very different and far less dramatic claim, even if it doesn’t make for as good a headline.
A Small Story That Proves the Point
A few months ago, a client needed a product name by 9 a.m. and I had nothing. Out of options, I threw the brief at an AI model, expecting the usual pile of clichés. What came back instead were two genuinely strange, unrelated ideas I would never have reached for on my own. I didn’t use either one directly, but one of them nudged me sideways into the name we actually shipped. That’s the study, in miniature. The AI wasn’t the creative director. It was the weird, tireless collaborator who never gets tired of throwing out options.
That’s the same pattern showing up wherever AI tools actually work well for small businesses right now, from copywriting to AI voice agents handling customer calls. The people getting real value out of these tools aren’t the ones hoping AI will replace their judgment. They’re the ones who’ve learned to steer a system that’s genuinely good at volume and lateral association, then apply the human filter that decides what’s actually worth keeping.
The Gap Between Average and Exceptional Is the Real Story
It’s worth sitting with why the top creative performers still win so clearly. A few reasons keep showing up in how researchers and working creatives talk about this:
- Exceptional human creativity draws on lived experience, memory, and emotional stakes that a model trained on text patterns simply doesn’t have.
- The richest creative work (a great personal essay, a poem that actually aches, a brand voice with real edge) depends on judgment about what to cut, not just what to generate.
- Taste, meaning knowing which of a hundred ideas is the one worth pursuing, remains stubbornly human, and arguably always will be.
- Standardized tests like the DAT measure a narrow slice of creativity: associative range. They don’t measure narrative structure, comedic timing, or the ability to read a room.
None of that should be comforting in a complacent way. It should be clarifying. The gap between “good enough” and “exceptional” content is exactly where AI creativity study data suggests human creators still have leverage, and it’s shrinking on the low end faster than most of us expected even a year ago.
What This Means for Your AI Content Creation Workflow
If you’re building content, running a brand, or managing a website right now, this study is less a warning and more a map. A few practical shifts follow directly from it.
First, stop treating AI as a finished-copy machine and start treating it as a divergence engine. Its real strength, per the data, is generating a wide, weird spread of options fast. Your job is the convergence step: picking, refining, and adding the specificity a model can’t invent because it wasn’t there.
Second, invest in the parts of the process AI is measurably worse at. That’s storytelling with emotional stakes, original reporting, interviews, and anything that requires a genuine point of view rather than a statistically likely one. This is the same argument I made in my piece on authentic AI content creation: audiences are getting sharper at smelling generic AI output, and the content that still breaks through is the content with a human fingerprint on it.
Third, remember that discoverability is changing alongside creativity. As more content gets AI-assisted, differentiation and structure matter more, not less, for how both search engines and AI answer engines surface your work.
The Human Edge That Still Matters
I don’t think this study is bad news for writers, designers, or marketers, and I say that as someone whose livelihood depends on being right about it. What it actually shows is a redistribution of effort. The unglamorous, first-draft brainstorming that used to eat up a Tuesday morning can now happen in ninety seconds. That’s not a threat to creative work. It’s a threat to creative busywork, and those are not the same thing.
The researchers behind the study made a similar point when their findings were first published: generative AI isn’t replacing creators, it’s becoming a tool in service of human creativity, one that changes how people imagine and explore ideas rather than what they ultimately decide to say. According to the ScienceDaily coverage of the research, that reframing, from replacement to amplification, is where the more useful conversation actually lives.
Where This Leaves Us
So here’s where I’ve landed, a few days into thinking about this study more than I probably should have. The average bar for “creative enough” content is going to keep climbing, because the tools generating baseline options are getting better every quarter. That means the premium on genuine originality, lived experience, and editorial taste isn’t shrinking. It’s becoming the whole game.
If you’re building an AI-assisted content or web design practice right now, this is the moment to lean into the 10 percent of the work only you can do, and let the machine handle the rest of the brainstorm. The study didn’t prove AI is more creative than people. It proved that most of us have more room to specialize in what actually makes us irreplaceable.