Successful policy briefs – with and without ChatGPT

Targeted communication in the political arena supports evidence-based decision-making. This requires researchers and science communicators to present their findings in an understandable way and relate them in the current context. When the relevance of the topic becomes clear, the chances increase that policymakers will take scientific findings into account in their decisional processes. Policy briefs are a key tool for communicating complex content and key messages and highlighting options for action. In the workshop “Successful Policy Briefs – with and without ChatGPT” at Forum Wissenschaftskommunikation 2025 on December 3-4 in Stuttgart, these aspects of political communication were addressed in the context of this year's focus on “Algorithms, Platforms, and AI: Science Communication in the Digital Transformation.” The workshop was organized by Paulina Conrad (Deutsche Allianz Meeresforschung), Tome Sandevski (ENGAGEgreen), and Katharina Sielemann (Helmholtz SynCom).

In an initial practical exercise, participants were able to use existing policy briefs to identify best practices for successful policy briefs. A successful policy brief summarizes the political problem precisely, provides a comprehensible classification of the topic, and places it in a relevant (social, economic, or) political context. In an input on science-policy engagement, reasons for dialogue between science and politics were summarized, including increasing the social impact of one's own research and the gain of new perspectives and ideas.

AI-supported language models can assist in the creation of policy briefs by, for example, opening up new perspectives, sharpening the focus (e.g., by tailoring it to the target audience), and providing suggestions for wording and recommendations/options for action or a synthesis of provided sources/documents. They are suitable for initial drafts and text modules, but can also be used for linguistic revision or to provide initial suggestions on the political relevance of the topic. In a second practical exercise, workshop participants were able to create their own policy briefs and reflect on the benefits of ChatGPT and other Large Language Models. When prompting, a clear instruction should be crafted—including purpose, format, language, style, target audience, desired length, and structure. For greater control over the content created, reliable sources and links should be specified, to which the AI should refer exclusively. Nevertheless, plausibility checks (“sense checking”), source verification, and technical validation remain indispensable! Likewise, AI-generated texts should be checked for plagiarism. Collegial exchange and critical proofreading by technically skilled colleagues are central elements of quality assurance.

At the end of the workshop, the needs of political audiences were examined and ways in which policy briefs can reach practitioners were presented. To consolidate the results, the discussed points were then summarized in a joint guide for successful policy briefs, and made available to the participants.