Study: Students Without AI Produce More Diverse Ideas Than ChatGPT Essays
A new study found that college students writing without AI tools generate significantly more diverse ideas than essays produced by large language models, raising questions about how widespread AI use could narrow the range of perspectives in classrooms and creative work.
Researchers analyzed 2,200 college admissions essays, comparing real student submissions from 2018 to 2022 with essays generated by GPT-4 using identical prompts. While individual GPT-4 essays often appeared highly creative, a different pattern emerged when researchers examined groups of essays together.
How Researchers Measured Diversity
The team created a metric called "diversity growth rate" to track how much each new essay contributed to the overall pool of ideas, rather than assessing creativity in isolation.
Human-written essays consistently introduced fresh experiences, perspectives, and new combinations of ideas. As more essays accumulated, the collective pool expanded. GPT-4 essays behaved differently: while individual pieces scored well on creativity measures, new AI-generated essays added little novelty to the group.
Across three studies, human-written essays increased collective diversity between two and eight times more than GPT-4 essays. The gap widened as more essays were added.
Researchers described this pattern as a "homogenising effect"-AI-generated writing converging around similar themes and expression patterns rather than continuously introducing new ones.
Individual Creativity Differs From Collective Diversity
The researchers stressed that their findings do not prove AI cannot be creative. Earlier studies showed GPT models sometimes match or exceed humans on creativity tests.
When assessed individually, GPT-4 essays often matched human essays on semantic diversity-a measure of how many different concepts and ideas connect within a piece of writing. Some GPT-4 essays even exceeded human work on this metric.
But creativity involves more than how impressive a single essay appears. It also involves whether many different people bring different perspectives to a discussion. Large numbers of GPT-4 essays resembled one another more than large numbers of human-written essays.
Attempts to Reduce the Homogenising Effect Failed
Researchers tested whether adjustments could reduce the pattern. They instructed GPT-4 to maximize creativity, changed model settings to encourage novel language and less repetition, and tested chain-of-thought prompting-a technique that asks AI to reason through tasks step by step.
These interventions improved individual essay creativity. In some cases, modified GPT-4 outputs surpassed human essays on diversity scores.
Yet the broader pattern persisted. Even after prompt changes, parameter adjustments, and chain-of-thought prompting, GPT-4 essays contributed fewer new ideas to the collective pool than human-written essays. Notably, the newer GPT-4 model showed an even stronger tendency toward homogenisation than an earlier version.
The Risk of Algorithmic Monoculture
The main concern is not that students will become worse writers by using AI. Instead, researchers warn that widespread reliance on the same AI systems could gradually reduce the diversity of perspectives in classrooms, universities, and creative environments.
This connects to "algorithmic monoculture"-a situation where large numbers of people rely on the same technology and produce increasingly similar outputs.
"If organisations, educational institutions, or creative industries rely too much on a certain AI model, the collective pool of ideas may become more uniform over time," the researchers wrote.
The authors called for AI literacy policies that encourage originality when students use AI tools. They also recommended further research into how AI affects creativity in academic writing, journalism, literature, and social media.
While AI can support individual creativity, human writers still contribute far more diversity of thought when viewed collectively-a difference that may become increasingly important as AI tools become routine in education. For writers, understanding these limitations matters when deciding how and when to use these tools in your work.
Learn more about AI for Writers and how to use these tools effectively while maintaining originality.
Your membership also unlocks:
