AI-powered search engines have been found to differ significantly from traditional Google searches, often surfacing less popular sources that don’t appear in the top 20 of conventional results.
That’s according to a paper published this month by researchers at Ruhr University Bochum in Germany and the Max Planck Institute for Software Systems, which compared Google’s traditional results with those generated by Google’s AI Overviews, Gemini 2.5, GPT-4o’s web search, and GPT-4o with Search Tool.
Queries ranged from specific questions and political topics to top products on Amazon and entries from the WildChat dataset.
AI search expands beyond top-ranked domains
AI search tended to draw from domains far outside both the top 1,000 and even the top 1 million sites tracked by domain-ranking service Tranco, indicating a much broader range of sources.
In product searches, AI results had less than 30% overlap with traditional Google results, often pulling from weaker or lesser-known domains. Across all query types, overlap between AI-generated and Google search results was below 50%.
“Our analysis reveals intriguing differences. Most generative search engines cover a wider range of sources compared to web search,” said Elisabeth Kirsten, lead author and PhD student at Ruhr University. “They vary in how much they rely on internal knowledge contained within model parameters versus external information retrieved from the web. Generative search engines surface different sets of concepts, creating new opportunities for enhancing search diversity and serendipity.”
It’s not surprising that AI values sources differently from Google, given their distinct objectives. Traditional search prioritizes authority and relevance to provide the best possible links, while AI aims to answer the query directly by synthesizing information from multiple sources. This often means casting a wider net for granular data, whereas Google focuses on returning the most authoritative or relevant domains to the user.
A new frontier: answer engine optimization
The researchers were careful not to assign a higher or lower value to either approach, noting that their goal was to highlight how AI search will fundamentally change how users connect with websites and datasets.
Early analysis suggests that domain strength may become less critical for traffic generation, though OpenAI, Google, and others have signed licensing deals in 2024 and 2025 with major news organizations, signaling that high-authority content still holds weight in model training.
The market for understanding this new wave of AI search has been given a name, Answer Engine Optimization (AEO), a possible successor to the widely booming SEO market. In SEO, key practices for ranking higher on Google include content quality, keyword placement, metadata, backlinks, sitemaps, and a few other factors. With AEO, the current thinking is that short, structured formats such as FAQs, featured snippets, and schema markup will appeal to these AI search tools.
The SEO market has been highly successful, offering tips and tricks to game Google’s search engine to achieve higher rankings and ultimately more traffic. Whether these analysts will have the same success at understanding the fundamentals of these answer engines is up for debate, especially given how quickly they are currently launching new products and changing their crawling parameters and values.
Learn more about how Google’s AI Summaries are creating what many describe as a “traffic apocalypse” for online publishers.