Google and AI systems: how source selection and content presentation differ

—

dodsno the inside of a logistic building lots of high shelves cc2e4d05 606a 4bba b344 277dc6da37ef 2

A recent study by Ruhr University Bochum and the Max Planck Institute for Software Systems examines how selected AI search systems – Google AI Overview, Gemini 2.5 Flash with search, GPT-4o Search and GPT-4o with its search tool enabled – differ from conventional Google Search. The researchers analysed more than 4,600 queries across six topic areas. The findings show that AI systems draw on a different range of information sources and often include less well-known websites that are barely visible in conventional Google Search.

The systems also differ considerably in how they work. While GPT models generate information largely from internal knowledge and refer to external sources only selectively, Google’s AI-powered products such as AI Overview and Gemini incorporate a broader range of websites into their answers. For ambiguous or current queries, conventional web search often remains more precise.

The study also found that repeated queries on the same topic can produce substantially different results in AI systems, and that they can struggle to integrate current information reliably. The researchers therefore argue for new evaluation criteria that give greater weight to factors such as transparency, source diversity and the traceability of automated summaries.

By Ida Kandler

Source: the-decoder.de, 26 October 2025, by Jonathan Kemper