Dr. Magdalena Obermaier
Postdoctoral research associate at the Department of Media and Communication at LMU Munich, Munich, Germany.
Email: magdalena.obermaier@ifkw.lmu.de
Dr. Cathy Buerger
Director of research at the Dangerous Speech Project, Washington DC, USA.
Email: catherine.buerger@gmail.com
Prof. Dr. Lena Frischlich
Associate professor and vice-director of the Digital Democracy Centre, University of Southern Denmark, Odense, Denmark.
Email: lefr@sam.sdu.dk
Lea Bund
Project manager at ichbinhier e.V., Berlin, Germany.
Fay Carathanassis
Research associate at the Technical University of Munich and associated researcher at the Bavarian Research Institute for Digital Transformation, Munich, Germany.
Anne Clausen
PhD candidate at the Digital Democracy Centre, University of Southern Denmark, Odense, Denmark.
Steliyana Doseva
Research associate at the Bavarian Research Institute for Digital Transformation, Munich, Germany.
Jan Eissfeldt
External applied complexity fellow at the Santa Fe Institute, New Mexico, USA.
Dr. Joshua Garland
Center director and associate research professor at Arizona State University, Arizona, USA.
Dr. Keyan Ghazi-Zahedi
Associated faculty member at Arizona State University, Arizona, USA.
Prof. Dr.-Ing. Christian Grimme
Professor of computational social science and systems analysis at the University of Münster, Münster, Germany.
Prof. Dr. Mario Haim
Professor at the Department of Media and Communication at LMU Munich, Munich, Germany.
Dr. Alina Herderich
Postdoctoral research associate at the IDea Lab at the University of Graz, Graz, Austria.
Yuru Li
Postdoctoral research associate at the Institute of Communication Science at Friedrich Schiller University Jena, Jena, Germany.
Hannah Oetting
Research associate and PhD candidate at the Department of Communication at the University of Münster, Münster, Germany.
Prof. Dr. Cornelius Puschmann
Professor at ZeMKI at the University of Bremen, Bremen, Germany.
Prof. Dr. Eugenia Rho
Assistant professor at the Department of Computer Science at Virginia Tech, Blacksburg, Virginia, USA.
Dr. Anke Stoll
Postdoctoral research associate at the Faculty of Social and Behavioural Sciences at the University of Amsterdam, Amsterdam, the Netherlands.
Dr. Andreas Wenninger
Research coordinator and research project leader at the Bavarian Research Institute for Digital Transformation, Munich, Germany.
Prof. Dr. Marc Ziegele
Professor at the Department of Social Sciences at Heinrich Heine University, Düsseldorf, Germany.
Hate speech is a widespread phenomenon in digital environments. It has severe consequences for individuals and society, can make online discussions more hostile, and may lead people to withdraw from public debate. One way to mitigate these effects is counterspeech: users who witness hate speech respond to it, support those targeted, or help foster a more constructive tone in digital public spaces. Unlike deletions or bans, counterspeech does not directly interfere with freedom of expression, but relies on public response.
Between support and risk: AI in counterspeech
The study examines how artificial intelligence can support counterspeech efforts. Many bystanders do not intervene when they witness hate speech online – for example due to uncertainty, lack of time, fear of backlash, or because they do not know how to respond effectively. AI systems could help identify harmful content, raise awareness of hate speech, provide suggestions for counterspeech, or support users in handling reactions after speaking up.
At the same time, the study shows that AI is not a simple solution to a complex societal problem. Its use involves significant risks, including biased training data, unreliable classifications, hallucinations, privacy concerns, and the danger that AI-generated counterspeech may be perceived as impersonal, manipulative, or inauthentic. AI could also be misused, for instance to produce hate speech more efficiently or to target counterspeakers more systematically.
The publication brings together the perspectives of an international and interdisciplinary group of researchers and practitioners from communication science, media psychology, anthropology, legal studies, computer science, civil society, and the tech industry. It is based on a workshop funded by bidt and organised at LMU Munich in February 2025. The study outlines the conditions under which AI can responsibly support counterspeech, and what this means for policymakers, platforms, civil society, and future research.
Key findings in brief
AI can support people in counterspeech
AI can support different phases of the counterspeech process: detecting hate speech, deciding whether and how to respond, formulating a response, and handling the aftermath. Such tools may be particularly useful for users who feel uncertain or have little experience with counterspeech.
System quality depends on data and context
For AI to be used responsibly in this field, it needs high-quality training data, clear concepts, valid and reliable models, and regular independent audits. This is especially challenging because hate speech is highly context-dependent. Language, platform norms, cultural background, and the groups affected can all influence whether a statement is perceived as harmful, discriminatory, or dangerous.
Technical solutions are not enough
Many risks do not arise from the technology alone, but from the interaction between AI systems and social dynamics. AI-generated counterspeech may support constructive dialogue, but it may also trigger mistrust or backlash. It can help de-escalate debates, but it can also flood digital spaces with automated responses. Its use therefore always needs to be assessed within the specific communicative context.
Responsible use requires transparency and oversight
The study makes clear that AI can support counterspeech against hate speech, but it cannot replace human responsibility. What is needed are transparent procedures, independent audits, the protection of fundamental rights, and a careful assessment of potential consequences for digital public spheres, freedom of expression, and affected groups.
