AI Chatbots Return Less Diverse Answers Than Search Engines, Study Finds

The team tested the models on 155 topics, ranging from nuclear weapons to K-Pop, using 200 prompt formulations drawn from research into real AI chatbot usage. The tests generated 70 million responses. GPT-5, which produced the most varied responses among the models tested, still returned answers 18.5 percent less varied than Google Search.

The findings suggest the issue is not limited to one chatbot or search tool. Researchers said the same pattern appeared whether users queried LLMs through Google Search, DuckDuckGo, Bing, Gemini, ChatGPT or other chatbot-accessible systems. The researchers believe this can push certain content to the edges of accepted knowledge while making other content dominant.

"People are to a large extent exposed to the same information over and over again," said first author Dustin Wright. "So AI chatbots are not just changing how we find knowledge, but also which knowledge we have access to."

Senior author Isabelle Augenstein, a professor at DIKU, said the effect could create a cycle in which popular content becomes more dominant and other material is increasingly overlooked. "We risk exposing people to fewer perspectives and a narrower range of knowledge. This could create a vicious cycle in which the most popular content becomes even more dominant, while other content is increasingly overlooked," she said. She compared the phenomenon to globalization, saying that the same products and coffee chains can now be found almost everywhere, which has advantages but has also reduced diversity.

Augenstein also said that a knowledge collapse is not happening yet. She noted that more recent models produce slightly more diverse answers than older models. The study's concern is that repeated reliance on AI answers could continue to shrink the range of information and perspectives available to users.