Self-preferencing is when an AI answer surface disproportionately cites its owner's own properties as the source for its answers, rather than independent third-party pages. In AI search it describes an engine routing its citations back into surfaces the platform controls — a search-engine's own results pages, its business listings, its video platform — so a large share of the "sources" in an answer belong to the same company that generated it.
The clearest measured example as of mid-2026 is Google AI Mode: SE Ranking's 2026 study of about 1.3 million citations found Google.com was the single most-cited domain at roughly 17.42%, tripling from about 5.7% a year earlier, with Google-controlled properties near 20% of all sources (attributed and hedged as single-vendor, directional data).
Self-preferencing is distinct from a walled garden, which is about blocking outside agents, and from a first-party citation, which is a brand citing its own domain. Here it is the platform preferring itself. For brands it raises effective visibility concentration: where an engine cites itself heavily, the external-citation pool is smaller, so the surfaces the platform hands back to itself — like a claimed business profile — become the practical place to compete.