Human Teams Produced More Diverse Ideas Than ChatGPT in a 208-Person Study

ChatGPT can help someone produce a long list of ideas quickly, but a 208-person experiment suggests that speed and volume don’t guarantee creative range. Participants working with another person generated more creative and varied ideas than those collaborating with ChatGPT, and the difference grew as human conversations continued. The result offers a useful distinction for teams: AI can contribute suggestions, while people may help a group explore more varied ground.

The study, published in Computers in Human Behavior, compared people doing a common creativity exercise with a human partner against people working with ChatGPT. Researchers randomly assigned 208 participants to one of those approaches. The task asked participants to come up with different uses for everyday objects, a method known as the alternate uses task. Researchers assessed both the number of suggestions and their originality.

More words didn’t mean more original ideas

People working with ChatGPT wrote longer responses than people taking turns with a human partner. Yet the human pairs came up with more creative ideas. The findings draw attention to a limit of counting suggestions: a lengthy response can contain repetition or extra explanation without offering more distinct possibilities.

That distinction fits a broader pattern in research on language models. They can generate many suggestions quickly, but high output doesn’t by itself show that the suggestions are original or useful. In one study of 12 chatbots answering thousands of factual questions, responses were often much longer than human-written answers; the answer itself made up about 42% of the text. Much of the rest was repetition, background information, or detail that strayed from the question. The researchers’ analysis is about factual answers, rather than brainstorming, but it illustrates why response length is an imperfect measure of value.

In the creativity experiment, an individual suggestion from ChatGPT could be inventive. For instance, the chatbot suggested using a book as a sound dampener. But researchers saw the same idea appear in separate sessions. That repetition made the pool of AI-assisted suggestions less varied than the ideas developed by human pairs.

AI can make separate projects sound alike

This finding echoes research on creative work produced with AI. One study found that AI-assisted stories could be more creative than an individual writer’s earlier work, while stories produced across users were more similar to one another. That study points to a difference between helping one person improve an output and expanding the range of ideas across a group.

There are similar concerns in image generation: repeated use of AI tools can yield generic-looking pictures, described by researchers as “visual elevator music.” These findings don’t mean every AI suggestion is dull. They suggest that suggestions from many users can begin to resemble one another.

A likely reason lies in how large language models produce text. Trained on large collections of writing, they generate likely sequences of words based on patterns in that material. Different systems, including ChatGPT, Claude, and Gemini, can therefore share recognizable habits in wording and structure, as well as overlap in the suggestions they make. Research comparing these systems describes similarities in their cadence, diction, and syntax.

That doesn’t make AI-assisted work automatically unoriginal. It does mean an idea that feels unusual in one conversation may be familiar across many conversations. When many people use similar tools, shared patterns can narrow the variety of language and ideas circulating in their work.

Human collaboration gained ground over time

The researchers also found that the advantage of working with another person became stronger as collaboration progressed. They measured how distinct concepts were using “semantic distance”: greater distance suggests ideas have moved farther apart in meaning. As people interacted, their ideas covered more semantic distance than ideas produced through chatbot collaboration.

This matters because many creative efforts take time. Research on new product development has estimated an average of 27 months between an initial idea and a finished product, with more innovative products taking as long as 54 months. That work provides context for why a tool’s performance in a short brainstorming session may not tell the whole story. If suggestions keep converging during longer projects, the range of options a team considers could narrow.

The study doesn’t establish that human collaboration is better for every task, or that its results apply to every team. It tested a particular exercise with 208 participants. Still, it offers evidence that the process of working together can matter: people bring individual experiences and ways of thinking that can lead a discussion in less predictable directions.

Using AI without narrowing the idea pool

One proposed way to broaden chatbot responses is “persona prompting,” which asks a system to respond from a specified perspective. Researchers found that this approach increased the diversity of ChatGPT’s ideas, but the range still fell short of ideas from human writers. Related work also suggests that personas don’t reliably reproduce the breadth of human perspectives; other research has found chatbots to be poor simulators of human personality.

For a creative project, a practical takeaway is to include people in the idea-making process, especially when a team wants a wide range of approaches. AI may help generate material quickly, but its suggestions can repeat familiar patterns. A conversation with another person can bring in personal experiences and unexpected connections that are harder to capture through a prompt.

Creative work can be as small as a message to a friend or as involved as a product concept. As teams add AI to their workflows, the study’s findings suggest that human input remains valuable for expanding the ideas under consideration. Other research has also linked work done with people to a stronger sense of meaning. The evidence is still developing, but for teams weighing how to brainstorm, a useful starting point is simple: exchange ideas with people, then use AI as one more source of suggestions.