Research led by a Korean search company argues that as AI-generated pages encroach into search results, they undermine the stability of search and ranking pipelines and weaken systems – such as Retrieval-Augmented Generation (RAG) – that rely on those rankings to decide what information is surfaced and trusted, thereby increasing the risk that misleading or inaccurate material will be treated as authoritative.

The term coined for this syndrome by the researchers is Retrieval Collapse, as distinct from the known threat of model collapse (where AI trained on its own output becomes progressively worse).

In a Retrieval Collapse scenario, AI-generated content progressively dominates search engine results, to the extent that even when answers remain superficially accurate, the underlying evidence base will have become divorced from original human sources. Nonetheless, this ‘rootless’ data seems poised to achieve a high place in search results*.