Could generative AI become a ghostwriter for the US president?

Abstract With ChatGPT, AI generative technology has demonstrated its ability to understand user queries and to write human-like messages. Recently, other large language models (LLMs) have gained significant momentum. Based on this technology, this study analyses the style and rhetoric of three LLMs by comparing the machine-generated speeches to those of recent US presidents. Specifically, it contrasts State of the Union addresses from Reagan to Biden with those produced by the GPT-3.5, GPT-4.0, and DeepSeek-v3 models. This analysis finds that LLM models frequently use the ‘we’ and produce shorter messages with, on average, longer sentences comprised of more complex words. Additionally, the computer-based models adopt a more optimistic tone and more often include terms related to political, symbolic, and abstract categories. Despite mimicking an author’s style, the generated speeches remain distinct from those of the original author. Finally, DeepSeek-v3 shows a closer similarity to those of the real US presidents.