Post by AI World

Is AI search making us better informed, or just more confident? Elisabeth Kirsten, Jost Grosse Perdekamp & @Mihir Upadhyay (arXiv, 2025) compare Google's organic results with four AI-powered engines across thousands of queries. The systems diverge significantly in which sources they surface and how they define what counts as authoritative, differences users cannot see. Alice Li & Luanne Sinnamon (ASIS&T, 2024), auditing ChatGPT, Bing Chat, and Perplexity across 48 public-interest queries, find consistent bias toward US commercial media, with academic and non-Western sources systematically underrepresented. Michelle Huang, Agam Goyal & Koustuv Sah (arXiv, 2025) go deeper, seeing how AI summaries strip out uncertainty language by up to 60% while preserving confident-sounding formulations. The same query, across different systems, produces different answers, invisible to the user. Mehrzad Khosravi & Hema Yoganarasimhan (arXiv, 2026), tracking 161,382 Wikipedia article-language pairs, find AI Overviews reduce daily traffic to sources by roughly 15%, redirecting attention away from the content AI itself trains on. Sunhao Dai, Ziyi Chen & Weizhi Ma (ACM SIGIR, 2025) identify the structural consequence. Generative search breaks the feedback loop that allowed systems to improve. End-to-end synthesis create black boxes where it is harder to know what path AI followed in order to reach an answer. Gary Marchionini put it best in 2006, before any of this existed. Research is not lookup, but an iterative/exploratory process through which understanding is built. AI answers are optimised for retrieval loosing the diversity good research is based on. 𝗪𝗮𝗻𝘁 𝘁𝗼 𝗿𝗲𝗮𝗱 𝘁𝗵𝗲 𝗳𝘂𝗹𝗹 𝗮𝗿𝘁𝗶𝗰𝗹𝗲 => https://lnkd.in/eUDVSYAP #AISearch #GenerativeAI #InformationBias #SourceConcentration #DigitalEpistemics #ResearchIntegrity #AIWorld