TY - RPRT A1 - Oswald, Lisa A1 - Munzert, Simon T1 - Exposure to untrustworthy news media then and now: Declining news media quality over 7 years N2 - In a rapidly evolving digital media landscape, understanding how exposure to untrustworthy news changes over time is essential for evaluating its potential effects on public attitudes and behavior. However, there is limited evidence on how demand for untrustworthy news develops across longer time frames. Linking web data with surveys, we compare exposure to untrustworthy news sources across demographic and political groups and over a 7-year time frame for two samples of German adults (N = 1,212 in 2017 and N = 436 in 2024). Visits to untrustworthy news sources make up less than 1% of media diets and are associated with low satisfaction with democracy and a preference for a far-right party. Propensity score matching reveals stability in untrustworthy news exposure, yet a notable decline in the average quality of news diets over 7 years. These results suggest that while untrustworthy news exposure remains limited and stable, a broader erosion in news quality may pose a growing challenge for informed democratic engagement. Y1 - 2025 U6 - https://doi.org/10.31235/osf.io/5ndjx_v1 PB - OSF ER - TY - RPRT A1 - Hartmann, David A1 - Oueslati, Amin A1 - Staufer, Dimitri A1 - Pohlmann, Lena A1 - Munzert, Simon A1 - Heuer, Hendrik T1 - Lost in Moderation: How Commercial Content Moderation APIs Over- and Under-Moderate Group-Targeted Hate Speech and Linguistic Variations N2 - Commercial content moderation APIs are marketed as scalable solutions to combat online hate speech. However, the reliance on these APIs risks both silencing legitimate speech, called over-moderation, and failing to protect online platforms from harmful speech, known as under-moderation. To assess such risks, this paper introduces a framework for auditing black-box NLP systems. Using the framework, we systematically evaluate five widely used commercial content moderation APIs. Analyzing five million queries based on four datasets, we find that APIs frequently rely on group identity terms, such as ``black'', to predict hate speech. While OpenAI's and Amazon's services perform slightly better, all providers under-moderate implicit hate speech, which uses codified messages, especially against LGBTQIA+ individuals. Simultaneously, they over-moderate counter-speech, reclaimed slurs and content related to Black, LGBTQIA+, Jewish, and Muslim people. We recommend that API providers offer better guidance on API implementation and threshold setting and more transparency on their APIs' limitations. Warning: This paper contains offensive and hateful terms and concepts. We have chosen to reproduce these terms for reasons of transparency. Y1 - 2025 U6 - https://doi.org/10.48550/arXiv.2503.01623 N1 - This is the author's version of the paper accepted at CHI Conference on Human Factors in Computing Systems (CHI '25), April 26-May 1, 2025, Yokohama, Japan. The version of record is available at: Hartmann, D., Oueslati, A., Staufer, D., Pohlmann, L., Munzert, S., & Heuer, H. (2025). Lost in Moderation: How Commercial Content Moderation APIs Over- and Under-Moderate Group-Targeted Hate Speech and Linguistic Variations. Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, CHI ’25, 1–26. https://doi.org/10.1145/3706598.3713998 PB - arXiv ER - TY - CHAP A1 - Hartmann, David A1 - Oueslati, Amin A1 - Staufer, Dimitri A1 - Pohlmann, Lena A1 - Munzert, Simon A1 - Heuer, Hendrik T1 - Lost in Moderation: How Commercial Content Moderation APIs Over- and Under-Moderate Group-Targeted Hate Speech and Linguistic Variations T2 - Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems N2 - Commercial content moderation APIs are marketed as scalable solutions to combat online hate speech. However, the reliance on these APIs risks both silencing legitimate speech, called over-moderation, and failing to protect online platforms from harmful speech, known as under-moderation. To assess such risks, this paper introduces a framework for auditing black-box NLP systems. Using the framework, we systematically evaluate five widely used commercial content moderation APIs. Analyzing five million queries based on four datasets, we find that APIs frequently rely on group identity terms, such as “black”, to predict hate speech. While OpenAI’s and Amazon’s services perform slightly better, all providers under-moderate implicit hate speech, which uses codified messages, especially against LGBTQIA+ individuals. Simultaneously, they over-moderate counter-speech, reclaimed slurs and content related to Black, LGBTQIA+, Jewish, and Muslim people. We recommend that API providers offer better guidance on API implementation and threshold setting and more transparency on their APIs’ limitations. Warning: This paper contains offensive and hateful terms and concepts. We have chosen to reproduce these terms for reasons of transparency. Y1 - 2025 U6 - https://doi.org/10.1145/3706598.3713998 N1 - The author's version is available at: Hartmann, D., Oueslati, A., Staufer, D., Pohlmann, L., Munzert, S., & Heuer, H. (2025). Lost in Moderation: How Commercial Content Moderation APIs Over- and Under-Moderate Group-Targeted Hate Speech and Linguistic Variations. arXiv. https://doi.org/10.48550/arXiv.2503.01623 SP - 1 EP - 26 PB - ACM CY - New York, NY, USA ER - TY - JOUR A1 - Romanello, Marina A1 - Walawender, Maria A1 - Hsu, Shih-Che A1 - Moskeland, Annalyse A1 - Palmeiro-Silva, Yasna A1 - Scamman, Daniel A1 - Smallcombe, James W A1 - Abdullah, Sabah A1 - Ades, Melanie A1 - Al-Maruf, Abdullah A1 - Ameli, Nadia A1 - Angelova, Denitsa A1 - Ayeb-Karlsson, Sonja A1 - Ballester, Joan A1 - Basagaña, Xavier A1 - Bechara, Hannah A1 - Beggs, Paul J A1 - Cai, Wenjia A1 - Campbell-Lendrum, Diarmid A1 - Charnley, Gina E C A1 - Courtenay, Orin A1 - Cross, Troy J A1 - Dalin, Carole A1 - Dasandi, Niheer A1 - Dasgupta, Shouro A1 - Davies, Michael A1 - Eckelman, Matthew A1 - Freyberg, Chris A1 - Garcia Corral, Paulina A1 - Gasparyan, Olga A1 - Giguere, Joseph A1 - Gordon-Strachan, Georgiana A1 - Gumy, Sophie A1 - Gunther, Samuel H A1 - Hamilton, Ian A1 - Hang, Yun A1 - Hänninen, Risto A1 - Hartinger, Stella A1 - He, Kehan A1 - Heidecke, Julian A1 - Hess, Jeremy J A1 - Jankin, Slava A1 - Jay, Ollie A1 - Pantera, Dafni Kalatzi A1 - Kelman, Ilan A1 - Kennard, Harry A1 - Kiesewetter, Gregor A1 - Kinney, Patrick A1 - Kniveton, Dominic A1 - Koubi, Vally A1 - Kouznetsov, Rostislav A1 - Lampard, Pete A1 - Lee, Jason K W A1 - Lemke, Bruno A1 - Li, Bo A1 - Linke, Andrew A1 - Liu, Yang A1 - Liu, Zhao A1 - Lowe, Rachel A1 - Ma, Siqi A1 - Mabhaudhi, Tafadzwanashe A1 - Maia, Carla A1 - Markandya, Anil A1 - Martin, Greta A1 - Martinez-Urtaza, Jaime A1 - Maslin, Mark A1 - McAllister, Lucy A1 - McMichael, Celia A1 - Mi, Zhifu A1 - Milner, James A1 - Minor, Kelton A1 - Minx, Jan A1 - Mohajeri, Nahid A1 - Momen, Natalie C A1 - Moradi-Lakeh, Maziar A1 - Morrisey, Karyn A1 - Munzert, Simon A1 - Murray, Kris A A1 - Obradovich, Nick A1 - Orgen, Papa A1 - Otto, Matthias A1 - Owfi, Fereidoon A1 - Pearman, Olivia L A1 - Pega, Frank A1 - Pershing, Andrew J A1 - Pinho-Gomes, Ana-Catarina A1 - Ponmattam, Jamie A1 - Rabbaniha, Mahnaz A1 - Repke, Tim A1 - Roa, Jorge A1 - Robinson, Elizabeth A1 - Rocklöv, Joacim A1 - Rojas-Rueda, David A1 - Ruiz-Cabrejos, Jorge A1 - Rusticucci, Matilde A1 - Salas, Renee N A1 - San José Plana, Adrià A1 - Semenza, Jan C A1 - Sherman, Jodi D A1 - Shumake-Guillemot, Joy A1 - Singh, Pratik A1 - Sjödin, Henrik A1 - Smith, Matthew R A1 - Sofiev, Mikhail A1 - Sorensen, Cecilia A1 - Springmann, Marco A1 - Stowell, Jennifer D A1 - Tabatabaei, Meisam A1 - Tartarini, Federico A1 - Taylor, Jonathon A1 - Tonne, Cathryn A1 - Treskova, Marina A1 - Trinanes, Joaquin A A1 - Uppstu, Andreas A1 - Valdes-Ortega, Nicolas A1 - Wagner, Fabian A1 - Watts, Nick A1 - Whitcombe, Hannah A1 - Wood, Richard A1 - Yang, Pu A1 - Zhang, Ying A1 - Zhang, Shaohui A1 - Zhang, Chi A1 - Zhang, Shihui A1 - Zhu, Qiao A1 - Gong, Peng A1 - Montgomery, Hugh A1 - Costello, Anthony T1 - The 2025 report of the Lancet Countdown on health and climate change: climate change action offers a lifeline JF - The Lancet Y1 - 2025 U6 - https://doi.org/10.1016/S0140-6736(25)01919-1 VL - 406 IS - 10521 SP - 2804 EP - 2857 PB - Elsevier BV ER -