Designing Stakeholder Personas with LLMs

We present insights from a research project that aimed at establishing an ecosystem for collaborative sharing and analysis of cybersecurity knowledge through federated learning. Within this broader context, one key challenge is ensuring that Cyber Threat Intelligence (CTI) outputs are relevant and understandable for diverse stakeholders, including technical experts and strategic decision-makers. To address this, we developed stakeholder personas using Large Language Models (LLMs) as part of the system design process. Based on workshops and interviews, we generated ten personas. These personas served two purposes: guiding the development of the MANTRA solution’s user-centric features and enabling persona-driven customization of CTI outputs, ensuring that threat intelligence is communicated in a format tailored to specific user profiles. Our findings indicate that LLMs can significantly streamline persona creation while maintaining contextual accuracy. Furthermore, integrating personas into CTI delivery improves usability and stakeholder engagement. We conclude by discussing lessons learned, limitations, and future directions.