A new study from Ruhr University Bochum and Max Planck Institute shows generative AI search tools draw heavily from less-popular websites, diverging from Google’s top ranked domains. Here's what SEOs, content strategists and agencies should know.
Researchers compared Google Search’s organic listings with AI-generated answers from LLMs (Gemini and GPT-4o) across factual, political, and product queries. They found that most domains cited in AI answers did not appear among Google’s top 10 results — and often ranked outside the top 1,000 most-visited domains in Tranco’s popularity index. Overlap between AI and Google results was below 50 percent(!) across categories, and in product queries, it dropped under 30 percent.
Elisabeth Kirsten, the study’s lead author, noted that “AI systems vary in how much they rely on internal model knowledge versus external web information,” which helps explain why they surface more obscure domains. In short, generative engines are widening their source pools while compressing the presentation — drawing from niche or lower-traffic sites instead of the same high-authority pages dominating traditional search.
For SEO and marketing professionals, this shift underscores the rise of Answer Engine Optimization (AEO) optimizing not just for ranking, but obviously for being referenced in AI responses. Structured data, clear factual writing, timestamps, and authoritative sourcing help content become machine-readable. Smaller sites may now have more visibility opportunities if their info is precise and well-formatted.
At the same time, this broader sourcing introduces risks: gen ai answers can draw from outdated and low-quality material, raising questions about credibility and fact integrity (aka: Slop On Top). The researchers didn’t label AI search better or worse, only different.
As with most of these studies, it is best to take this with a grain of salt. This is but a snapshot in time of what was - not what is, or what will be in the future. The algos could change tomorrow with dials adjusted in an ongoing basis. Best not to bank anything on this study.



