- Subject
- Nutrition
- Filed
- Source
- Clin Lai and Roger Beaty
- Loans
- 806
Human Teams Generated More Diverse Ideas Than ChatGPT in a 208-Person Study
In a 208-person experiment, pairs of people produced more varied ideas than people working with ChatGPT, especially as collaboration continued.

ChatGPT helped people produce longer lists of possible uses for everyday objects, but working with another person led to ideas that were more creative and varied in a recent experiment. The finding points to a trade-off: generative AI can quickly supply suggestions, while human collaboration may help teams explore a wider range of possibilities over time.
The experiment, conducted by cognitive scientists studying creativity, assigned 208 participants at random to complete an alternate uses task either with ChatGPT or with another person. Participants were asked to think of different ways to use common objects, a method researchers use to study divergent thinking: the ability to generate ideas that branch out from familiar uses.
Researchers assessed both the number of ideas and their originality. People working with ChatGPT wrote longer responses than those working in pairs. But the human pairs produced more creative ideas, and their suggestions tended to be more distinct from one another. That difference matters: a long list can look productive while covering less conceptual ground than a shorter, more varied set.
More suggestions don’t necessarily mean more variety
Large language models can generate many suggestions quickly, which helps explain their appeal for tasks such as drafting and summarizing. But volume alone says little about whether the ideas are fresh or meaningfully different. In the experiment, some suggestions from ChatGPT were inventive in isolation. One example was using a book as a sound dampener. Yet the same idea appeared in multiple separate sessions, making the chatbot’s pool of suggestions less diverse.
That pattern fits with research on AI-generated writing and images. One study found that answers from 12 chatbots to thousands of factual questions were often much longer than human-written answers; the answer itself accounted for about 42% of the text, with the rest largely made up of repetition, background detail, or material that strayed from the question. Other studies have found that AI-assisted stories may be more creative than an individual’s earlier work, while stories produced with AI across different users tend to resemble one another.
The researchers connect this convergence to how large language models produce text. Trained on large collections of writing, these systems generate language by predicting likely next words. Models developed through similar processes can produce suggestions and writing with similar diction, sentence patterns, and rhythm. A suggestion can seem original when considered alone, yet become familiar when many people receive similar suggestions from AI tools.
Human collaboration can widen the search
The gap between people and ChatGPT also grew as the collaboration continued. Researchers used semantic distance—a way of estimating how conceptually far apart ideas are—to examine how suggestions changed over time. Interactions between people led to greater semantic distance than exchanges with a chatbot, suggesting that human partners moved into less closely related areas of thought as they continued brainstorming.
One possible explanation is that people bring individual experiences and unexpected associations into a conversation. A partner can respond to an idea in ways shaped by a personal history, then shift direction as the discussion unfolds. A chatbot can offer many plausible responses, but its suggestions may pull a conversation toward recurring patterns. If those patterns build up across a long project, they could narrow the set of directions a team considers.
That time dimension is relevant to creative work that stretches beyond a single brainstorming session. Research cited by the scientists estimates that developing a new product takes an average of 27 months from the first idea to a finished product; more innovative products can take as long as 54 months. The experiment doesn’t show what would happen in every workplace or over a project lasting years. It does suggest that the way teams collaborate may shape the range of ideas they keep exploring.
Ways to use AI while keeping human input
Persona prompts are one proposed way to encourage a chatbot to offer a wider mix of suggestions. A user might ask it to respond from the perspective of a teacher or a business founder, for example. Research has found that these prompts can increase the variety of ChatGPT’s ideas, but that variety still fell short of what human writers generated. Other work has also raised doubts about how well chatbots can imitate human personalities.
For people using AI during creative work, the practical implication is to keep human conversation in the process. A colleague or friend may offer a slower, less polished response, but personal experiences can lead the discussion somewhere that a chatbot’s repeated patterns might miss. AI can still help generate starting points; the experiment suggests that teams looking for a broad range of ideas may benefit from exchanging views with people as well.
The study measured responses to a particular brainstorming task, so it can’t establish that human teams will always be more creative than AI-assisted ones. Nor does it show that every chatbot interaction limits originality. But it offers a useful distinction: getting more words quickly is different from exploring more varied ideas. For longer creative projects, preserving room for human perspectives may help keep the search open.