Google AI Overviews are becoming part of how people encounter political information in search. An exploratory study by AlgorithmWatch suggests that, for election-related queries, the feature does not appear consistently, relies on a limited set of cited sources, and can frame parties or politicians in uneven ways.
What AlgorithmWatch examined
AlgorithmWatch, a German advocacy group, used access to Google's "Search Researcher Result API" through research data access under Article 40(12) of the EU's Digital Services Act. The group ran 4,480 search queries about the 2026 state elections in Saxony-Anhalt and Mecklenburg-Vorpommern and the Berlin House of Representatives election, all in eastern Germany.
The study then looked at which searches produced AI Overviews and which sources those overviews cited. That matters because AI Overviews can sit at the top of a search experience, shaping what users see before they click through to individual pages.
The central finding was not that Google always answers political queries. It was that the system appears to decide when to answer in ways that are difficult to understand from the outside.
Election searches triggered fewer AI Overviews
According to the study, Google showed an AI Overview for election questions in 39.1 percent of cases. For non-political queries, the figure was 65.3 percent.
The lower rate may sound cautious, but AlgorithmWatch found that the pattern was uneven. Questions about polls almost never triggered an AI Overview. Questions about parties and individual politicians did so in 40 to 50 percent of cases.
There were also differences between parties and regions. Questions about the far-right AfD produced overviews less often, at 24 percent, than questions about other parties, which appeared at 45 to 58 percent. Saxony-Anhalt produced fewer overviews overall than Berlin and Mecklenburg-Vorpommern.
A Google spokesperson told AlgorithmWatch that AI search features are designed to be neutral and objective and are not shown for every search. The spokesperson said overviews appear when Google assumes a high added value. AlgorithmWatch reported that the practical criteria behind that decision were not explained.
The study also points to a shift in the wider context around election-related AI answers. Back in 2024, Google said it would block AI-generated answers from its Gemini language model on election-related questions. AlgorithmWatch said that measure, announced with reference to the elections in the US and India, was very effective at the time. Whether similar safeguards still apply in 2026 could not be determined, and Google did not comment.
A narrow source base shaped the answers
AlgorithmWatch found that the AI Overviews drew from a limited range of cited sources. Nearly half of all links came from just ten domains.
The most cited source was YouTube, Google's own platform. AlgorithmWatch interpreted that as a sign that Google favors its own services, a claim Google denies.
Among media outlets, German public broadcasters such as NDR, rbb, MDR, and ARD appeared prominently, along with Zeit Online. Six media sites accounted for 75 percent of all media links.
The source mix also included Wikipedia, government sites, and party pages. AlgorithmWatch noted that party pages appeared without a label distinguishing them from neutral information sources.
That distinction is important in election search because different source types play different roles. A government page, a media report, a party page, and a video platform can all contain relevant information, but they do not carry the same editorial function or political distance. If an AI-generated summary presents them side by side without clear context, readers may not know how to weigh them.
Political framing was not always neutral
The study found that Google AI Overviews sometimes described parties and politicians with terms that sounded favorable or partisan. AlgorithmWatch highlighted vague, positive adjectives such as "pragmatic," "close to citizens," and "stable," saying they echoed parties' own descriptions of themselves.
In a sample on Saxony-Anhalt, 81.5 percent of overviews about the center-right CDU were rated clearly positive. The comparable figures were 50 percent for the center-left SPD, 10.7 percent for the Greens, and zero percent for the AfD.
AlgorithmWatch also found that a higher share of party-affiliated domains tended to appear alongside more positive portrayals, though not equally across all parties. That does not establish a simple rule for every result, but it raises a clear concern: the tone of an AI Overview may be influenced by the source pool behind it.
The descriptions of individual politicians could also vary. AI Overviews described AfD candidate Ulrich Siegmund from Saxony-Anhalt in different ways: sometimes as a representative of the far-right and national-conservative wing, sometimes as someone holding radical far-right positions, and sometimes as right-wing populist.
AlgorithmWatch said it could not identify a pattern behind those contradictory labels, some of which appeared within short spans of time.
Why transparency matters
AlgorithmWatch found clear factual errors only rarely, which separates this study from some earlier work on AI search. The concern here is less about obvious falsehoods and more about opacity, source concentration, and tone.
For election information, those issues can matter even when individual facts are mostly accurate. A summary that appears for one party but not another, draws heavily from a small group of domains, or uses flattering political language can still influence how information is framed.
AlgorithmWatch traces the flattering wording to the tendency of language models toward sycophancy. The group calls for transparency requirements and clear liability rules.
The study leaves Google with a practical question: if AI Overviews are used for sensitive political searches, how can users and researchers understand when they appear, why certain sources are selected, and how political descriptions are produced? Without clearer answers, election-related AI search will remain difficult to audit from the outside.