A collaborative study by researchers at the University of California, Riverside, warns that the increasing reliance on artificial intelligence for internet research is filtering out the human element from online discourse. Presented at the ACM Web Science Conference in Braunschweig, Germany, the research indicates that while AI offers efficiency, it fails to replicate the rich, multidimensional reasoning found in human-written content. The study highlights a growing divergence between the cold logic of large language models and the ethos and pathos that characterize human argumentation. Led by doctoral student Md Taukir Azam Chowdhury, the team compared responses from major AI systems like ChatGPT and Gemini against traditional web searches via Google and Bing. Using hundreds of subjective and controversial questions, the researchers analyzed the types of justifications used to support various positions. They categorized reasoning based on Aristotle's rhetorical triangle: logos (logic and factual consistency), ethos (authority and credibility), and pathos (emotion and shared experience). The findings reveal that AI models overwhelmingly rely on logos, prioritizing factual consistency and neutral explanations. In contrast, human-authored web pages blend logic with emotional appeals, personal anecdotes, moral concerns, and storytelling. For instance, when searching for a margarita recipe, an AI might provide a standard formula, whereas a human culinary site would offer diverse variations, historical context, and personal narratives that add depth and meaning to the information. Similarly, in discussions about public policy or social issues, human content often draws on lived experiences and emotional resonance, while AI responses remain statistically probable but emotionally flat. Researchers suggest this limitation stems partly from the safety and alignment protocols built into AI models, which are designed to minimize harmful or controversial output by steering responses toward neutral, fact-based language. Consequently, AI systems tend to produce a homogenized version of knowledge that lacks the nuance, diversity, and personal authority found in the broader internet. Vagelis Hristidis, a co-author and computer scientist at UCR, expressed concern that as users substitute AI for traditional search, the web may gradually lose the human nature that has shaped it over the last 25 years. Kevin Esterling, a professor of public policy and political science, noted that while humans are wired to attribute cognition to any entity producing language, machines simply predict word sequences without true understanding of audience or emotion. Esterling emphasized that unlike human conversation, which involves anticipating reactions and engaging in two-way interaction, AI operates as a one-way prediction engine. The study warns that this shift could have significant societal implications. By relying on AI for information regarding politics, health, and ethics, the public may lose exposure to the messy but vital diversity of human reasoning that fosters empathy and mutual understanding. The researchers argue that while AI provides a distilled version of knowledge, it strips away the emotional and moral context necessary for a fully informed public discourse. The full paper, titled Comparing the Subjective Opinions and Justifications of LLMs and Web Search Engines, was published in the Proceedings of the 18th ACM Web Science Conference 2026, with contributions from the UCR Computer Science and Engineering Department.