Chatbots Limiting Access to Information, Study Warns

Researchers from the University of Copenhagen have discovered that AI chatbots provide users with significantly less varied information compared to traditional Google searches. The study, which tested 27 language models on 155 topics, generated a dataset of around 70 million individual claims produced by the models.

The results show that even OpenAI's GPT-5, which is considered one of the most advanced language models, provides at least 18.7% less varied information than Google.

This phenomenon is attributed to the way large language models function. They compress huge amounts of text they're trained on and learn the patterns that occur most frequently, filtering out information that deviates from these common patterns.

The researchers warn that repeated use of AI-generated training data could further narrow future models' knowledge, potentially leading to a 'knowledge collapse' where users are exposed to fewer perspectives and a narrower range of knowledge.