Study: People started copying phrases from AI chatbots

Researchers from the Max Planck Institute for Human Development have reported signs of the impact of chatbots on users’ speech. The paper, published in July, notes that after the appearance of ChatGPT, YouTube video creators have become noticeably more likely to use words that are typical of large language models. Among the most striking examples are such terms as “underscore,” “comprehend,” “bolster,” “swift,” “inquiry,” and “meticulous.” The analysis covers a period of one and a half years and demonstrates a correlation between the growth of chatbots’ popularity and changes in people’s vocabulary.

Wired journalists have made additional observations. According to moderators of large subreddits, it is becoming increasingly difficult to distinguish texts written by users, as some authors adopt the style of presentation inherent in AI. Similar trends have been described in the New York Times Magazine, which drew attention to British parliamentarians using language that is unusual for local political rhetoric but typical of American speeches.

Researchers have drawn similar conclusions from official announcements by Starbucks about the closure of its stores. They feature emotional expressions that resemble the structure of phrases often created by generative models. There is no direct evidence of the use of chatbots in these cases, but linguists believe that this style has become widespread since the advent of modern AI systems.