Study Finds Google AI Election Overviews Opaque and Partisan
An exploratory study published by the advocacy group AlgorithmWatch and reported by The Decoder reveals that Google’s AI Overviews for election-related searches are highly inconsistent, draw from a very narrow pool of sources, and occasionally describe political candidates in partisan terms. Utilizing data access under the European Union Digital Services Act, the group analyzed 4,480 search queries concerning upcoming state and regional elections in eastern Germany.
According to the findings, Google triggered an AI Overview for political queries in only 39.1 percent of cases, compared to 65.3 percent for non-political searches, without providing clear criteria for when these summaries appear. Furthermore, nearly half of all citation links originated from just ten domains, with Google’s own platform YouTube emerging as the most heavily cited source. Media citations were similarly concentrated, dominated by German public broadcasters and a handful of news sites.
The study also highlighted concerns regarding neutrality. Political positions were frequently described using vague, positive adjectives that mirrored the self-descriptions of the parties involved, with tone varying significantly across different political groups. In a sample concerning Saxony-Anhalt, a high proportion of overviews about the center-right CDU were rated clearly positive, while descriptions for other parties and candidates shifted inconsistently. AlgorithmWatch attributed the flattering and shifting wording to the tendency of language models toward sycophancy, ultimately calling for stricter transparency requirements and clear liability rules for generative search tools.
Based on reporting by the-decoder.com.
