The Influence of Large Language Models on Country Reputation | Mobiliar Center for Resilience

The Influence of Large Language Models (LLMs) on Country Reputation

This project examines how Large Language Models (LLMs) shape a country's reputation from multiple perspectives. LLMs act simultaneously as mirrors and agents of societal narratives, reflecting and influencing public discourse about countries. By analyzing how LLMs mediate country-specific information and how individuals interact with AI-generated content, we investigate their potential to alter perceptions on a global scale. Further, we explore how these systems might be leveraged for measuring and strategically managing a country's reputation.

This project is a collaboration between Prof. Diana Ingenhoff (Organizational Communication and Public Diplomacy) and Prof. Olivier Furrer (Marketing) from the Faculty of Management, Economics, and Social Sciences.

Aim

The project aims to deepen our understanding of AI’s role in a country’s reputation by:

National Narratives: Investigating how LLMs process, generate and reinforce national narratives and country-specific stories, including the possibility that they may perpetuate biases or reshape common perceptions.
Human-AI Interaction: Exploring how people engage with LLMs and how AI-generated narratives about countries influence individual attitudes, trust, and beliefs.
Strategic Communication Management: Demonstrating how LLMs can be harnessed for reputation measurement, detecting risks and opportunities to support proactive country reputation strategies.

Connecting the Dots: Building Resilience at National and Individual Levels

Thereby, the project places a focus on resilience: Consistent with the notion of resilience as a nation’s capacity to anticipate, absorb, and recover from disruptions (e.g., harmful or misleading AI outputs, false information), this project highlights two resilience fronts:

National-Level Resilience: Supporting governments and institutions in navigating reputational shocks caused by LLM-driven misrepresentations. By understanding how AI systems generate narratives and how these narratives spread, stakeholders can adapt their strategies and uphold a robust country image.
Individual-Level Resilience: Empowering users to critically assess AI-generated content, fostering media literacy, and ensuring that citizens develop informed perspectives rather than passively absorbing potentially skewed outputs.

With this dual focus on the nation’s adaptive capacity and the public’s critical engagement with AI, we aim to provide practical insights that not only safeguard and strengthen national reputation in the evolving AI era but also build resilience at both the national and individual levels.

Methodology

The study employs a multi-method approach combining theoretical and empirical components. It begins with developing a theoretical framework to conceptualize the interplay between AI technologies and a country‘s reputation. This is followed by agent-based testing of LLMs using standardized content analysis to understand how they process and present country information. An online experiment examines user interactions with LLMs when seeking country-related information, analyzing how LLM-generated content influences country perception. Finally, automated content analysis is conducted to efficiently process large datasets and identify patterns in country representations across different LLMs, enabling tracking changes in country reputation over time and across various AI systems.