Award honors research communication on AI election tools

A press release from TU Dortmund University highlighting research on artificial intelligence and election advice tools has received national recognition. The Informationsdienst Wissenschaft (idw) awarded the text second place in the 2025 idw Prize for Science Communication, recognizing outstanding research communication among nearly 100 participating institutions.

The award-winning press release, titled Ahead of the 2025 German Federal Election: TU Dortmund team identifies vulnerabilities in AI election advice tools, presented research conducted by scientists from TU Dortmund University and the Research Center Trustworthy Data Science and Security (RC Trust) within the University Alliance Ruhr.

The research examined a new generation of AI-based election advice tools designed to help voters explore party positions interactively. Unlike the well-known Wahl-O-Mat, which guides users through a fixed set of political statements, tools such as wahl.chat and wahlweise.info allow voters to ask open questions using large language models similar to ChatGPT or Llama.

However, the researchers found that this flexibility comes with significant challenges. When the AI systems were asked about party positions, their responses often deviated from the verified statements provided in the Wahl-O-Mat dataset. In a systematic evaluation using the 38 official Wahl-O-Mat statements, the team conducted nearly 400 checks for each tool.

The results revealed substantial discrepancies. For wahl.chat, around 25 percent of responses differed from the party positions recorded in Wahl-O-Mat. For wahlweise.info, deviations occurred in more than half of the tested cases. In addition, repeated questions sometimes produced different answers on different days, illustrating how probabilistic AI systems can generate inconsistent interpretations of political information.

The study was carried out by a nine-member interdisciplinary research team from TU Dortmund and the RC Trust, including Dr. Ina Dormuth, Marlies Hafer, Sven Franke, Tim Katzke, Prof. Alexander Marx, Prof. Emmanuel Müller, Prof. Daniel Neider, Prof. Markus Pauly, and Jérôme Rutinowski.

“Our concern is that citizens may trust AI assistants even though they generate the most probable answer – not necessarily the factually correct one,” explained Prof. Emmanuel Müller, Founding Director of the RC Trust.

The analysis also explored how robust the tools are against manipulative inputs. In tests using politically sensitive terms identified by Germany’s Federal Office for the Protection of the Constitution, one system initially blocked problematic queries. Yet the researchers demonstrated that the filter could be bypassed using so-called prompt injections, allowing the chatbot to generate fabricated or misleading statements about political parties.

“AI systems rely heavily on probabilities and the exact phrasing of a user’s input,” said Prof. Markus Pauly, statistician at the Department of Statistics at TU Dortmund University. “As a result, they may misinterpret or misrepresent political programs.”

Despite these limitations, the researchers emphasize that AI-based systems also hold potential. In contrast to fixed questionnaire formats, conversational AI could allow users to explore policy issues in more depth. At the same time, the findings underline the importance of rigorous evaluation, certification, and safeguards for AI applications that influence public discourse.

The award from idw recognizes how effectively the TU Dortmund press release communicated these complex findings to a broader audience. In her laudation, jury member and journalist Korinna Hennig (NDR Info) praised the text for capturing readers’ attention immediately while explaining the research clearly and responsibly. She highlighted the strong structure, accessible language, and the careful balance between scientific depth and journalistic clarity.

The recognition also reflects the broader mission of the Research Center Trustworthy Data Science and Security (RC Trust): advancing interdisciplinary research that helps society understand and critically evaluate emerging data-driven technologies.

As artificial intelligence increasingly shapes how people access information - including political information - such research plays a crucial role in ensuring that these systems remain transparent, reliable, and worthy of public trust.

Editorial note: The original press release was published by the University Communications Office on the website of TU Dortmund University.