Generative AI just cleared a new bar in creativity, at least for the average person. This AI creativity study compared results from more than 100,000 people with several large language models, including ChatGPT, Claude, and Gemini, and it found some models can outscore a typical human on a standardized creativity task.
But the ceiling still looks human. The study reports the most creative half of participants outperformed every AI model tested, and the top 10% widened the lead even more.
AI is getting better at clearing baseline creative tasks, while exceptional human output keeps a gap that’s hard to erase.
The test behind the claim
The researchers leaned on the Divergent Association Task, a quick prompt that asks for ten words that are as unrelated to each other as possible. Scores rise when those words are more semantically distant, and most people finish in a few minutes.
That simplicity is why the team could run such a large comparison. It also helps explain the headline result, models can be tuned to generate wide-ranging word choices on demand, which maps neatly onto what DAT rewards.
Still, DAT measures one slice of creativity, the ability to produce divergent language. It doesn’t measure taste, emotional impact, or whether an idea is the right one for a specific audience.
Where humans keep an edge
The strongest signal in the findings isn’t a single winner, it’s the spread. Some AI systems can beat the middle of the pack, but high-scoring humans separate themselves, and the separation grows at the top end.
In day-to-day terms, models excel at volume. If you need ten directions fast, it can deliver. What it can’t reliably do is the selective part, choosing the one direction worth pursuing, shaping it for constraints, and making it feel intentional instead of merely plausible.
That’s also why the result shouldn’t be read as a verdict on creative careers. The benchmark shows ideation range. It doesn’t show judgment under pressure, or the kind of originality that changes what an audience expects.
What to do with it
The team also compared people and models on creative writing style tasks, including haiku, plot summaries, and short stories, which better resembles how many people use ChatGPT. Even there, top human creators kept the advantage.
If you’re using AI at work, treat it as an ideation accelerator. Use it to generate breadth, then apply the part that still separates you, decide what fits your voice, what matches the brief, and what’s worth shipping.
Keep an eye on follow-ups that pin down exact model versions and test dates, because this kind of leaderboard can move quickly as models change.
