What ChatGPT “Thinks” About Your Country? Sentiments and Frames of AI Geographies

ABSTRACT What opinions are generated by AI algorithms about different countries? Is there a systematic difference in the opinions generated about each region? In this paper we developed a machine learning model to classify labeled and unlabeled datasets of opinions about countries generated by ChatGPT. We used semantic network analysis to map the positive and negative opinions associated with each country. We found that ChatGPT tends to produce positive opinions, emphasizing countries' ability to attract tourism and offer job opportunities. We also found that Western countries located in North America and Europe were often described as a shining example of functioning economies and therefore wonderful places to visit, work or live in. Conversely, poorer countries, particularly those in the Middle East and Africa, were more likely to be described as dysfunctional places, politically corrupt, and unsafe to visit. Overall, ChatGPT focused on the economic performance of countries, and their potential contribution to the global economy.