AI search agents can be gamed at the ecosystem level, not just the page level.
Researchers have introduced EcoGEO, a framework that studies how networks of coordinated web pages can influence the behavior of large language model agents doing web research. The key insight is that these agents do not just read a single page — they crawl links, reformulate queries, and synthesize evidence across multiple browsing steps. The team built a system called TRACE that constructs a controlled web environment: a navigation entry page paired with supporting pages that share terminology, internal links, and consistent product details designed to nudge the agent toward recommending a specific fictional product. Tested against OPR-Bench, a benchmark for open-ended product recommendation, TRACE outperformed page-level optimization baselines on final recommendation outcomes.
This matters because the dominant approach to search optimization has always targeted individual pages, but AI agents change the threat model entirely. If an agent's conclusions depend on the trajectory of its browsing session, a well-architected cluster of coordinated sites could exert influence that no single page ever could — and current defenses are not built for that.
Search engine optimization has been a cat-and-mouse game for three decades, and this research suggests the mouse just got a significant upgrade.