When Google inserts AI-generated answers at the top of search results, when Gemini begins directly answering your questions rather than returning links, search engines are undergoing a silent revolution. A new empirical study accepted by ACM SIGIR 2026 used 11,500 real user queries to systematically compare traditional Google search, AI Overview (AIO), and Gemini Flash 2.5. The conclusions are both counter-intuitive and worth pondering.
The core finding points to an unsettling reality: generative search is significantly changing how users get information. The three search engines' retrieval source overlap for the same query is extremely low — average Jaccard similarity below 0.2. This means when you use traditional Google search and AI search for the same question, you're actually seeing two completely different information worlds. More noteworthy, traditional search prioritizes authoritative content from government or educational institutions, while generative search clearly prefers retrieving Google's own content. AI Overview's circumvention of crawler restrictions exposes a structural problem: websites' robots.txt declarations have almost no binding force on AI crawling.
The deeper meaning of this research: it reminds us to re-examine the changing semantics of the word "search." When 51.5% of user queries trigger AI Overview, and generative search engines favor their own content ecosystems, traditional SEO logic is being completely rewritten.
For ordinary users, this is both convenience and risk — you get quick answers more easily, but the system filtering information for you is opaque. For content creators and website operators, when your content is directly cited by AI but no longer brings you traffic, when your robots.txt is meaningless to AI crawlers, the rules of the game have changed. This research calls for building a new revenue framework, forming a sustainable symbiotic relationship between publishers and generative search providers — this may be the most important unresolved issue in the information ecosystem of the LLM era.