Participating in any Zoom meeting, attending a lecture, or simply watching a YouTube video makes it evident that the influence of AI-generated language is on the rise. Terms like “prowess” and “tapestry,” commonly used by ChatGPT, are increasingly permeating our conversations, while others, including “bolster,” “unearth,” and “nuance,” have seen a decline. Researchers are now documenting significant changes in our communication patterns, viewing the linguistic impact of ChatGPT as a trend that is growing rapidly.
A study conducted by the Max Planck Institute for Human Development reveals that in the 18 months following the launch of ChatGPT, the frequency of words like “meticulous,” “delve,” “realm,” and “adept” rose by as much as 51 percent compared to the preceding three years. The researchers analyzed nearly 280,000 YouTube videos from academic sources to arrive at their conclusions, ruling out other potential factors that could explain the shift. Interestingly, these findings align with those from a previous analysis comparing 10,000 human-edited and AI-edited texts, confirming that speakers are unknowingly embracing a language shaped by AI.
Researchers identified one term that has emerged as a particular marker of this shift: “delve.” This word has become an academic symbol, a clear sign of AI’s presence in conversations. Hiromu Yakura, the lead author of the study and a postdoctoral researcher, notes, “We internalize this virtual vocabulary into daily communication.”
“‘Delve’ is only the tip of the iceberg.”
The influence of AI extends beyond vocabulary; it also impacts our delivery and tone. While current research primarily focuses on word choice, experts suggest that AI’s presence might be shaping longer, more structured speech patterns and a reduction in emotional expression. According to Levin Brinkmann, a research scientist at the Max Planck Institute and coauthor of the study, “’Delve’ is only the tip of the iceberg.”
AI’s influence is evident in various functionalities like smart replies and spellcheck. Research from Cornell University highlights that the use of smart replies fosters a greater sense of cooperation and emotional closeness among users as they tend to choose more positive language. However, if individuals suspect AI involvement in their exchanges, they perceive their partners as less collaborative and more demanding. This change in perception arises not from actual AI usage but from the suspicion of it, according to Malte Jung, an Associate Professor of Information Science at Cornell.
This duality—enhanced communication through AI alongside a growing mistrust—points to a broader decline in trust, says Mor Naaman, a professor of Information Science at Cornell Tech. He has identified three critical levels of human signals that are diminishing with the rise of AI in our conversations. The first encompasses basic humanity signals that convey authenticity, such as moments of vulnerability, which communicate “This is me, I’m human.” The second involves attention signals that demonstrate effort, indicating “I cared enough to write this myself.” The third concerns ability signals that reveal personal traits like humor and competence. For instance, saying “I’m sorry you’re upset” sounds flat, while “Hey, sorry, I freaked at dinner; I probably shouldn’t have skipped therapy this week” feels more human.
Naaman emphasizes the importance of recovering and amplifying these signals as a strategy for improving AI-mediated communication. He warns that AI’s impact extends beyond language to influence our thought processes. “In dating contexts, what does it mean to be funny in your profile or texts when AI can craft humor for you?” he questions, worrying that this could lead to a loss of agency in our expression and thoughts. “Rather than articulating our own thoughts, we simply echo whatever AI helps us articulate…we become more easily influenced.” Naaman cautions that, without these essential signals, we may come to trust only in face-to-face interactions, steering clear of even video calls.
We lose the verbal stumbles, regional idioms, and off-kilter phrases that signal vulnerability, authenticity, and personhood
The issue of trust is exacerbated by the fact that AI is subtly determining who gets to be seen as “legitimate” in conversations. Research from the University of California, Berkeley found that AI responses often contain stereotypes or distorted representations when tasked with dialects outside of Standard American English. Instances involve AI repeating prompts from users of non-Standard dialects due to a lack of understanding and exaggerating their speech significantly. A Singaporean English user remarked that the “super exaggerated Singlish” in one AI response was “slightly cringeworthy.” The findings suggest that AI not only prefers Standard American English, it simplifies other dialects in a way that can belittle their speakers.
This phenomenon perpetuates misconceptions not just about various communities but also concerning what constitutes “correct” English. Consequently, the stakes extend beyond merely preserving linguistic diversity—they encompass safeguarding the imperfections that nurture trust. As communication becomes dominated by “correct” language, we risk losing the verbal blunders, regional expressions, and quirky phrases that convey authenticity and individuality.
We find ourselves at a pivotal moment where AI’s influence on language is poised between the extremes of standardization—like templated professional emails—and genuine self-expression. Central to this dilemma are three main tensions: early backlash from users rejecting terms like “delve,” efforts to avoid sounding like AI, and the potential for AI to evolve into a more expressive tool that reflects human diversity. The most profound concern, however, revolves not around linguistic uniformity but the risk of ceding control over our thoughts and expressions.
The future trajectory is not predetermined between homogenization and hyperpersonalization; it hinges on our willingness to engage actively in that transformation. Initial feedback suggests that people may resist evident AI influence, while technology could adapt to more authentically resonate with human diversity. The critical question is not whether AI will continue to shape our communication—it certainly will—but whether we will make a conscious effort to maintain the idiosyncrasies and emotional depth that define authentic human interaction.
