What Is the 'Knowledge Collapse'? New Studies Warn AI Chatbots Are Narrowing Human Information

AI chatbots and large language models are altering not just how people find information, but also which knowledge they have access to, according to a massive new research study warning of an impending "knowledge collapse." The findings, reported by outlets including Inc., The Guardian, and TechRadar, detail how automated synthesis tools supply a much narrower range of perspectives compared to traditional web searches.

Why are AI chatbots leading to a 'knowledge collapse'?

AI chatbots give users more similar information than traditional search engines across all studied topics, meaning people are repeatedly exposed to the same data points, according to Dustin Wright, a researcher at Aalborg University and lead author of a University of Copenhagen study. Wright stated that AI is actively altering which knowledge users can access.

The study from the University of Copenhagen's Department of Computer Science (DIKU) and Aalborg University tested 27 AI chatbots and large language models across 155 topics spanning 12 countries, generating over 1.7 million answers. The research concluded that every AI chatbot provided more uniform responses than a simple Google search. Depending on the reporting outlet, OpenAI's GPT-5 gave between 18.5% and 18.7% fewer varied information responses compared to Google Search.

Isabelle Augenstein, a professor and senior author at DIKU, compared the phenomenon to globalization. Augenstein stated that while modern standardization has advantages, it reduces diversity. She warned that society risks exposing people to fewer perspectives, creating a vicious cycle where popular content becomes dominant while other information is overlooked.

How does AI tool use impact human cognition?

Heavy AI use correlates with reduced critical-thinking performance and cognitive under-engagement, according to separate studies highlighted in recent analyses. A June 2025 MIT Media Lab study titled "Your Brain on ChatGPT" found that students using ChatGPT to write essays could not recall key content afterward, with over 80% failing to accurately quote their own work. Brain scans during the study revealed reduced alpha and beta connectivity, with lingering cognitive suppression even after students switched back to writing without AI.

Additional data points cited in the research include a 666-participant study showing a strong negative correlation between heavy AI reliance and critical-thinking performance, alongside a 2023 systematic review identifying "metacognitive laziness."

What are the broader impacts on publishers and health information?

The shift toward AI-generated summaries has severely impacted organic traffic for web publishers. A January 2026 Guardian investigation into Google's AI Overviews health advice found potentially inaccurate interpretations for liver function tests and pancreatic cancer dietary advice, alongside concerning guidance on eating disorders and psychosis. Representatives from medical organizations, including Pancreatic Cancer UK, the British Liver Trust, Mind, and The Eve Appeal, voiced concerns over the data.

Meanwhile, publisher traffic has dropped sharply following the integration of AI Overviews:

• DMG Media desktop click-through rates dropped from 25.23% to 2.79%, marking an 89% collapse, while mobile dropped 87%.
• CNN search referrals decreased by 27% to 38% year-over-year.
• Business Insider traffic fell by 55%, leading to a 21% staff cut.
• HubSpot search referrals dropped by 70% to 80%.
• Chegg traffic fell by 49%, prompting the company to file an antitrust lawsuit.

Data from Similarweb and SparkToro indicated that 60% of searches now end without any click, up from 56% to 69% between May 2024 and May 2025, while Google processes between 9.1 billion and 13.6 billion searches daily.

When will a total knowledge collapse occur?

A total knowledge collapse is not happening yet, according to Professor Isabelle Augenstein, who noted that more recent AI models produce slightly more diverse answers than older iterations. Autonomous generative AI studies from January 2026, authored by Hintze, Proschinger Åström, and Schossau, also demonstrated that iteratively linking text-to-image and image-to-text systems caused diversity to collapse into stock imagery when isolated from new data.

This is educational information, not legal, financial, tax, or investment advice.