A new study from the University of Copenhagen warns that AI chatbots give users a narrower slice of knowledge than search engines, a trend researchers call ‘knowledge collapse’. The team tested 27 large language models across 155 topics using 200 prompt formulations each, generating about 70 million claims.

Every model tested returned more uniform information than a Google search. Even GPT-5, the most diverse model, provided at least 18.7% less varied information than Google.

First author Dustin Wright, now at Aalborg University, said people are ‘to a large extent exposed to the same information over and over again’. Senior author Isabelle Augenstein warned of a vicious cycle in which popular content dominates while other content is overlooked.

A separate MIT working paper from February 2026, updated in May, models how agentic AI substituting for human learning could tip society into a ‘knowledge-collapse’ steady state despite high-quality personalized advice.

Context: as chatbots replace search for everyday questions, the findings suggest society may be quietly losing the diversity of information it once browsed freely.