Heavy reliance on AI chatbots could affect how people learn, reason, and understand information, according to a new study. Researchers from the Department of Computer Science at the University of Copenhagen examined how much variety people receive from AI-generated answers.
The study looked at 27 large language models (LLMs), including some of the leading AI chatbots available today. Researchers also compared their responses with information found through Google Search. The team examined 155 topics and used 200 written prompts to gather more than 70 million responses from the AI models.
Their findings suggest that AI chatbots often provide a narrower range of information than traditional search. The best-performing chatbot produced responses with 19% less diversity than information available through Google Search.
That difference could affect how people build knowledge over time. When users repeatedly receive similar answers, they may be exposed to fewer ideas and viewpoints.
The researchers warn that this could also influence reasoning and intellectual development. In the long term, they suggest that intelligence levels could decline across generations compared with previous ones.
GPT-5 recorded the highest response diversity among the 27 models tested. However, its responses were still 18.5% less diverse than the information available through Google Search.
The researchers argue that the issue goes beyond how people find information. AI chatbots can also shape which information users encounter in the first place. When the same types of answers appear repeatedly, users may have less exposure to different perspectives and ideas. This could gradually narrow their understanding of a topic.
The findings do not suggest that AI chatbots are always harmful. Instead, the study highlights a possible downside of relying on them as the main source of information. As AI becomes more common in everyday learning and research, the researchers believe the variety of information people receive will become increasingly important.
