AI chatbots give narrower, more uniform information than a simple Google search, according to a University of Copenhagen study led by researcher Dustin Wright. The team tested 27 large language models across 155 topics, using 200 prompt formulations per topic to generate about 70 million individual claims.

Even the most diverse model, OpenAI’s GPT-5, provided at least 18.7 percent less varied information than Google, the researchers found. The topics ranged from nuclear weapons, marriage, racism and genocide to country-specific subjects like Marine Le Pen, the Falklands War and K-pop.

Senior author Isabelle Augenstein warned of a “knowledge collapse,” in which fewer perspectives survive online, the most popular content becomes more dominant, and minority views get overlooked. The concern is that as chatbots become the default way people seek information, the shared pool of knowledge could shrink.

The study adds to a growing body of research on how AI systems shape what people know, not just how fast they can find it.