Examining the Evolution and Influence of Inauthentic State Messaging Campaigns

Do inauthentic messaging campaigns enable autocrats to deflect criticism and inflate perceptions of their support? ICT penetration has seen propagandists deploy a variety of messaging strategies to alter citizen perceptions of state responsibility and levels of public outrage following episodes of state repression. In addition, recent advances in large-language models (LLMs) have raised concerns over the spread of cheap means of generating and widely spreading sophisticated propaganda. To assess the influence of inauthentic campaigns carried out in autocracies we are currently preparing to deploy an online survey experiment to study the influence of state messaging efforts among Tanzanian citizens. After securing a training dataset of tweets posted by paid pro-regime actors, we create a classification of propaganda strategies that have been recently deployed by the governing party (CCM). Using this data, we train a machine-learning model to generate responses to tweets critical of the Tanzanian government. Our survey experiment intends to measure the following quantities of interest: How do the common propaganda strategies of authoritarian states alter the ability of citizens to identify pro-government propaganda, assess societal support for the regime, and accurately recall details of governmental repression? Moreover, we hope to determine whether these effects differ if each inauthentic strategy is written by human propagandists or modeled by an open source LLM. The study has implications for researchers and policymakers interested in understanding and combating the continued use of contemporary technologies to implement inauthentic campaigns by autocrats.