Who Does Google Cite for Health? Auditing the Source Ecosystem of AI Overviews in Taiwan

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Google’s AI Overviews (AIO) now synthesize answers atop the search results page for a large fraction of health queries, displacing the ranked list of links that long mediated public access to health information. This represents a new venue for users to get information; however, almost all scrutiny of AIO concerns English-language, U.S.-localized search. We present the first systematic audit, to our knowledge, of the AIO source ecosystem in a non-English, non-Western setting: Traditional-Chinese (zh-TW) health search in Taiwan. Drawing on the 625 most-searched real-world Google Taiwan health queries of 2025, we characterize which sources Google elevates across the 417 overviews they triggered, and how those sources relate. Data collection was repeated over four runs to control for AIO’s instability. Each overview presents a dense citation slate (median 10 URLs, 8 hosts), often with an image (59%) but rarely a query-specific medical warning (16%). Citations are overwhelmingly Taiwan-based (90%) and dominated by healthcare providers (40%) and commercial sites (24%); academic, Wikipedia, and social media sources are each marginal (2%). The pool is long-tailed and unconcentrated (HHI ≈ 55), yet co-citation analysis of the 975 domains reveals a single hub-dominated component anchored by hospital and government portals, with Facebook and Wikipedia among the most-cited domains. We discuss implications for source transparency, AI-generated content curation, and cross-lingual algorithmic accountability.