Parschan, P., Reiss, M. V., Haim, M., Kasianenko, K., Nagappa, A., Angus, D., & Bruns, A. (2026). Using generative AI instead of search engines? : Comparison and implications in the context of five state elections in Germany. The Search Engines and Society Network Annual Conference 2026. Access the paper here.
Check the Search Engines and Society Network (SEASON) Conference 2026 details here.
The increasing integration of generative artificial intelligence into search engines marks a profound shift in how citizens can access political information. With the introduction of AI-based answer modes (AI modes) in major search engines in 2024 (Reid, 2024), users are no longer merely presented with ranked lists of sources, but instead receive synthesized or even conversational responses that directly aggregate, interpret, and contextualize information. While this shift from a search-rank logic (Schwabl et al., 2023) to communicative AI (Guzman & Lewis, 2020) promises convenience and efficiency, it also raises fundamental questions about the algorithmic selection, framing, and presentation of political information. These dynamics become particularly critical during elections as electoral periods are characterized by heightened information demand, strategic communication by political actors, and increased vulnerability to misinformation (Allcott & Gentzkow, 2017; Armitage & Vaccari, 2021; Bennett & Livingston, 2018; Humprecht, 2025; Strömbäck & Nord, 2006).
In response to these concerns, a growing body of literature has investigated the behavior of recommender systems and search engines during elections and election campaigns (Blassnig et al., 2023; Unkel & Haim, 2021; Urman et al., 2022; Van Hoof et al., 2024). As one major gateway to news (Newman et al., 2025) and election information more broadly (Schwabl et al., 2023), search engines take a particularly consequential role during campaigns. This prominence raises the question of whether AI-generated information summaries (AIIS) in search engines may differ systematically from traditional search results in how they deliver political information and there are at least three reasons to expect that they do.
First, AI modes are a comparatively recent technological development. Historical research on search engines suggests that even traditional algorithmic ranking systems have long shaped which information and viewpoints people encounter by privileging some sources over others, often in opaque ways (Introna & Nissenbaum, 2000; O’Neil, 2017; Stray et al., 2021). AI-based systems may face similaror even amplified—challenges, given their greater autonomy in generating responses and their own (political) biases (Rozado, 2023, 2024). For example, to date, there is only limited understanding of the sources AIIS rely on in the context of news and political information, whether these sources are journalistic, non-journalistic, partisan, governmental, or otherwise (Schatto-Eckrodt et al., 2025). Furthermore, whereas traditional search engines mediate access to information through selection and ranking mechanisms and the referenced sources themselves, AI modes may exert a more direct influence, as they generate the very explanations, summaries, and recommendations that users read. This risks making such prestructuring harder for users to detect and critically evaluate. Second, AI modes introduce novel risks that are largely absent in classical search results. Hallucinations—plausible but factually incorrect statements—pose a particular concern in political contexts, where misinformation can have direct consequences for democratic decision-making (Augenstein et al., 2024). Related to this is the fact that AIIS present information in a unified narrative voice, potentially obscuring uncertainty, disagreement, or gaps in available knowledge. One manifestation of this risk is “algorithmic bothsidism”, or misrepresentation of the two sides of a contested issue as deserving of equal consideration (Mitra, 2025). Traditional search engines on the other hand externalize responsibility to linked sources.
Finally, the specific affordances of AI assistants (Reiss et al., 2026) further distinguish them from classical search engines. AI assistants’ hyper-personalization and dialogic interaction are aimed to keep users engaged by providing subjective relevant and comprehensive answers without requiring additional clicks or source exploration, jeopardizing the ad-based revenue model of many online publishers (Perez, 2025). Furthermore, this logic of a “one-stop shop for information” (Reiss et al., 2026) may reduce users’ exposure to diverse perspectives and limit opportunities for critical comparison. Studies show that information seekers using AI assistants tend to be more passive, reinforce their own views, and engage with the sources less when compared to traditional online information seeking behavior (Melumad & Yun, 2025; Singh et al., 2025). At the same time, there are first indications that AI assistants lead to the “marginalization of regional perspectives” (Brantner et al., 2025). This, however, is an essential requirement in federal democratic systems such as Germany, where electoral competition, party relevance, and political issues vary substantially across states.
Against this backdrop, we ask: How do AI-generated information summaries and traditional search results differ in their sourcing and content in the context of German state elections?
To answer this research question, our study systematically compares AIIS with traditional search results during five German state election campaigns in 2026. These elections provide a unique empirical setting with regard to the use of AI assistants and elections and are among the first in Germany to take place after the public rollout of AI modes, making them an early opportunity to assess how generative search technologies operate under real-world democratic conditions. Moreover, the political contexts of these elections are heterogeneous: Baden-Württemberg lacks an incumbent prime minister, while in Saxony-Anhalt and Rhineland-Palatinate incumbents hold office without having been directly elected to that position. Such variation increases the relevance of questions surrounding political visibility, contextualization, and incumbency advantages in AI-generated information.
We focus on Google Search and Microsoft Bing, the two dominant search engines in Germany and build on Schwabl et al. (2023) and Unkel and Haim (2021) for constructing a set of queries across four categories relevant for the elections under study: parties, candidates, issues, and election guidance. This approach led to 90 to 100 queries per state election. Beginning two weeks prior to each election, we issued every query once per day and collected the corresponding results. This produced 2 × 2 × 14 outputs per query: For both Google and Bing, one set of results from traditional search engine pages and one from AIIS, each observed across 14 consecutive days.
The analysis proceeds in three steps, analyzing Google and Bing separately. First, we descriptively identify the queries for which no AIIS is provided and discuss differences between Google and Bing. Second, following Schwabl et al. (2023), we compare the variety, balance, and disparity of sources for the results on the remaining queries. They allow a systematic overview and comparison of the sourcing structure between traditional search results and AIIS. Third, we compare the types of the most frequent sources. To this end, the second-level domains that together account for at least 80% of all unique sources are manually categorized by two coders as journalistic (public service media or private), social media, institutional (government/state or NGO), party-aligned, or other. Two coders independently check the same subset of AIIS results, assessing whether the general claim in the summary is supported by the cited source and whether the AIIS asserts a value judgement (e.g., endorsing or criticizing a party). Inter-coder reliability will be reported.
At the time of submission, we have collected data for the state elections in Baden-Württemberg and Rhineland-Palatinate, which took place earlier this year. The remaining three elections (Saxony Anhalt, Mecklenburg–Western Pomerania, and Berlin) are scheduled for September 2026. Accordingly, at the SEASON conference we will present preliminary findings based on the two completed elections.
By examining how traditional search engines and AI modes respond to a wide range of election related queries, we seek to evaluate through a combination of computational and manual analytical approaches whether and in what ways the introduction of AI driven search modes reshapes the informational environment in which voters form political judgments. In doing so, this study aims to provide empirically grounded insights into an ongoing technological transformation and to support the design of transparent, accountable, and democratically robust information ecosystems.
Acknowledgements
The authors would like to acknowledge that this work was supported by the 2025 QueenslandBavaria Collaborative Research Program – Seed grant.
References
Allcott, H., & Gentzkow, M. (2017). Social Media and Fake News in the 2016 Election. Journal of Economic Perspectives, 31(2), 211–236. https://doi.org/10.1257/jep.31.2.211
Armitage, R., & Vaccari, C. (2021). Misinformation and disinformation. In H. Tumber & S. Waisbord (Hrsg.), The Routledge Companion to Media Disinformation and Populism (1. Aufl., S. 3848). Routledge. https://doi.org/10.4324/9781003004431-5
Augenstein, I., Baldwin, T., Cha, M., Chakraborty, T., Ciampaglia, G. L., Corney, D., DiResta, R., Ferrara, E., Hale, S., Halevy, A., Hovy, E., Ji, H., Menczer, F., Miguez, R., Nakov, P., Scheufele, D., Sharma, S., & Zagni, G. (2024). Factuality challenges in the era of large language models and opportunities for fact-checking. Nature Machine Intelligence, 6(8), 852–863. https://doi.org/10.1038/s42256-024-00881-z
Bennett, W. L., & Livingston, S. (2018). The disinformation order: Disruptive communication and the decline of democratic institutions. European Journal of Communication, 33(2), 122–139. https://doi.org/10.1177/0267323118760317
Blassnig, S., Mitova, E., Pfiffner, N., & Reiss, M. V. (2023). Googling referendum campaigns: Analyzing online search patterns regarding swiss direct-democratic votes. Media and Communication, 11(1), 19–30. https://doi.org/10.17645/mac.v11i1.6030
Brantner, C., Karlsson, M., & Kuai, J. (2025). Sourcing behavior and the role of news media in AIpowered search engines in the digital media ecosystem: Comparing political news retrieval across five languages. Telecommunications Policy, 49(5), 102952. https://doi.org/10.1016/j.telpol.2025.102952
Guzman, A. L., & Lewis, S. C. (2020). Artificial intelligence and communication: A HumanMachine Communication research agenda. New Media & Society, 22(1), 70–86. https://doi.org/10.1177/1461444819858691
Humprecht, E. (2025). Misinformation and disinformation. In A. Nai, M. Grömping, & D. Wirz (Hrsg.), Elgar Encyclopedia of Political Communication (S. 529–533). Edward Elgar Publishing. https://doi.org/10.4337/9781035301447.vol2.00129
Introna, L. D., & Nissenbaum, H. (2000). Shaping the Web: Why the Politics of Search Engines Matters. The Information Society, 16(3), 169–185. https://doi.org/10.1080/01972240050133634
Melumad, S., & Yun, J. H. (2025). Experimental evidence of the effects of large language models versus web search on depth of learning. PNAS Nexus, 4(10), pgaf316. https://doi.org/10.1093/pnasnexus/pgaf316
Mitra, B. (2025). Search and Society: Reimagining Information Access for Radical Futures. Information Retrieval Research, 1(1), 47–92. https://doi.org/10.54195/irrj.19654
Newman, N., Ross Arguedas, A., Robertson, C. T., Nielsen, R. K., & Fletcher, R. (2025). Reuters Institute Digital News Report 2025.
Reuters Institute for the Study of Journalism.
O’Neil, C. (2017). Weapons of math destruction: How big data increases inequality and threatens democracy (First paperback edition). B/D/W/Y Broadway Books.
Perez, S. (2025, Juli 2). ChatGPT referrals to news sites are growing, but not enough to offset search declines. Yahoo Finance. https://finance.yahoo.com/news/chatgpt-referrals-news-sites-growing174840587.html
Reid, L. (2024). Generative AI in Search: Let Google do the searching for you. https://blog.google/products/search/generative-ai-google-search-may-2024/
Reiss, M. V., Knor, E. L., Merten, L., & Moeller, J. (2026). Will News Find Them With Generative AI? Exploring the Affordances and Potential of GenAI for News Consumption. Weizenbaum Journal of the Digital Society, 6(1). https://doi.org/10.34669/WI.WJDS/6.1.1
Rozado, D. (2023). The Political Biases of ChatGPT. Social Sciences, 12(3), 148. https://doi.org/10.3390/socsci12030148
Rozado, D. (2024). The political preferences of LLMs. PLOS ONE, 19(7), e0306621. https://doi.org/10.1371/journal.pone.0306621
Schatto-Eckrodt, T., Liebig, L., Reiss, M. V., Geislinger, R., Schaetz, N., Merten, L., Schröder, J. T., Königslöw, K. K., Laugwitz, L., Knor, E. L., Semmann, M., Behre, J., & Lischka, J. A. (2025). ChatGPT as a news recommender system: Measuring source types and diversity across different interfaces. SocArXiv. https://doi.org/10.31235/osf.io/wjzp3_v3
Schwabl, P., Unkel, J., & Haim, M. (2023). Vielfalt bei Google? Vielzahl, Ausgewogenheit und Verschiedenheit wahlbezogener Suchergebnisse. In C. Holtz-Bacha (Hrsg.), Die (Massen-) Medien im Wahlkampf (S. 293–316). Springer Fachmedien Wiesbaden. https://doi.org/10.1007/978-3-658-389673_11
Singh, A., Guan, Z., & Rieh, S. Y. (2025). Enhancing Critical Thinking in Generative AI Search with Metacognitive Prompts (Version 2). arXiv. https://doi.org/10.48550/ARXIV.2505.24014
Stray, J., Vendrov, I., Nixon, J., Adler, S., & Hadfield-Menell, D. (2021). What are you optimizing for? Aligning recommender systems with human values. arXiv:2107.10939 [Cs]. http://arxiv.org/abs/2107.10939
Strömbäck, J., & Nord, L. W. (2006). Do Politicians Lead the Tango?: A Study of the Relationship between Swedish Journalists and their Political Sources in the Context of Election Campaigns. European Journal of Communication, 21(2), 147–164. https://doi.org/10.1177/0267323105064043
Unkel, J., & Haim, M. (2021). Googling Politics: Parties, Sources, and Issue Ownerships on Google in the 2017 German Federal Election Campaign. Social Science Computer Review, 39(5), 844–861. https://doi.org/10.1177/0894439319881634
Urman, A., Makhortykh, M., & Ulloa, R. (2022). The Matter of Chance: Auditing Web Search Results Related to the 2020 U.S. Presidential Primary Elections Across Six Search Engines. Social Science Computer Review, 40(5), 1323–1339. https://doi.org/10.1177/08944393211006863
Van Hoof, M., Meppelink, C. S., Moeller, J., & Trilling, D. (2024). Searching differently? How political attitudes impact search queries about political issues. New Media & Society, 26(7), 37283750. https://doi.org/10.1177/14614448221104405