AI Chatbots Serve Narrower Knowledge Than Search, Copenhagen Study Finds

A study led by the University of Copenhagen has cautioned that depending heavily on artificial intelligence chatbots for information can lead to a "knowledge collapse", because users end up seeing a much narrower set of facts and viewpoints than they would through an ordinary web search.

Chatbots such as ChatGPT, Gemini and Claude have become everyday tools for questions ranging from meal ideas and drafting difficult emails to explaining what a rise in interest rates means. These services rely on large language models to interpret questions and produce answers.

Researchers at the university's Department of Computer Science tested 27 large language models across 155 topics tied to 12 countries, generating roughly 1.7 million answers and 70 million claims.

Every chatbot returned more uniform responses than a simple Google search on all topics examined, the study found. Even the most diverse model tested, OpenAI's GPT-5, offered at least 18.7 per cent less varied information than Google.

"Every language model we tested provides users with more uniform information than a simple Google search across all the topics we looked at," said first author Dustin Wright, a former postdoctoral researcher at the department who is now an assistant professor at Aalborg University. He added that people are largely exposed to the same information repeatedly, meaning chatbots are changing not only how knowledge is found but also which knowledge is accessible.

The study described a "narrower slice of knowledge" as reduced variety in content rather than shorter answers. Chatbots tend to repeat similar claims and angles instead of presenting multiple sources, contexts or interpretations. Smaller models sometimes produced more varied answers than larger ones, and newer models performed better than older ones, but none matched the diversity of a search engine that links to many websites.

"We risk exposing people to fewer perspectives and a narrower range of knowledge," said senior author Isabelle Augenstein, a professor at the department. She warned this could create a cycle in which the most popular content becomes even more dominant while other material is increasingly overlooked.

The researchers also flagged that if future AI systems are trained partly on text generated by other AIs, which is already less varied, each generation could become progressively narrower, gradually shrinking the range of information available through chatbots. The team calls this self-reinforcing cycle "knowledge collapse".

Augenstein stressed that chatbots should not be treated as the only source of information. Practical steps for students and users include cross-checking chatbot answers against official websites, textbooks and multiple news sources, particularly for exam syllabi, government schemes and admission rules; using search engines to explore different pages and perspectives before settling on an answer; and combining AI tools with classroom learning, peer discussion and library resources to maintain breadth of understanding.

AI chatbots remain powerful aids for drafting, summarising and quick revision, but the study indicates they currently offer a narrower knowledge diet than traditional search.