Context
Ma and colleagues studied approximately ten thousand websites across Google conventional and generative search and reported differences in language-model predictability and semantic similarity of cited content. Additional controlled Gemini RAG and participant experiments examined LLM content polishing.
What changed
The attributable preprint reported that Google AI Overview cited sources were more linguistically predictable and semantically similar than conventional ranked results in the historical sample, and tested content refinement in a separate controlled RAG environment.
Why it matters
It adds original evidence that citation selection and answer synthesis need separate measurement from organic search rank and that source-text properties can correlate with generative citations.
Visibility OS interpretation
Visibility OS interpretation: Observed style and semantic associations are hypotheses about source selection, not established live ranking signals or proof that homogenizing content increases current AI visibility.
What remains uncertain
How much is explained by retrieval, topical relevance or source selection bias? Will the findings replicate with later Google models and other search systems? Can controlled RAG polishing outcomes transfer to production?