AI-powered search engines rely on “less popular” sources, researchers find

Recent research has shed light on the contrasting nature of search results generated by AI-powered engines compared to traditional search methods. Since the controversial introduction of Google’s AI Overviews last year, it has become evident that these advanced search systems often prioritize lesser-known websites, diverging significantly from the familiar ranking of links used for years. In the pre-print study titled "Characterizing Web Search in The Age of Generative AI," researchers from Ruhr University in Bochum and the Max Planck Institute for Software Systems conducted a thorough analysis comparing traditional search results from Google to its AI Overviews and the Gemini-2.5-Flash. They also explored the web search capabilities of GPT-4o, which employs a search tool only when the AI determines that external information is necessary. The study examined various queries sourced from several platforms, including specific questions from the WildChat dataset, political topics from AllSides, and the top 100 most-searched products on Amazon. Findings revealed that the sources referenced by generative search tools were predominantly from less popular domains compared to those found in the top 10 results of a standard Google search, according to the domain tracker Tranco. Notably, the AI search engines were more inclined to cite websites that were not only less popular but also often fell outside the Top 1,000 and even the Top 1,000,000 domains tracked by Tranco. In particular, Gemini’s search results frequently pointed to obscure domains, with the median source consistently ranking outside the Top 1,000. Alarmingly, 53% of the sources referenced by Google’s AI Overviews did not appear in the Top 10 results of organic Google searches for the same queries, while 40% of those sources failed to make it into the Top 100. This study highlights a significant shift in the landscape of online search, raising important questions about the reliability and relevance of information accessed through AI-driven platforms.