Researchers at Florida State University have reported early signs that artificial intelligence may already be influencing the way people speak. Their study examined more than 22 million words taken from unscripted science and technology podcasts and found that certain expressions favored by large language models are appearing more often in casual conversation.
The analysis compared transcripts from before and after the release of ChatGPT in late 2022. The team noted a rise in terms that AI systems frequently use, such as “align,” “strategically,” “surpass,” and “boast.” At the same time, other possible alternatives like “accentuate” or “probe” showed little change, pointing to a selective shift rather than a broad expansion of vocabulary.
Why the findings matter
Language has always shifted alongside technology, but the current change appears unusually concentrated and rapid. The study showed that nearly three-quarters of words associated with AI grew in frequency, with some, including “surpass,” more than doubling in usage. The increase was especially strong in spontaneous speech, which suggests the influence is not limited to text generated directly by machines.
The researchers describe this as a problem of model alignment. In their view, linguistic habits built into AI outputs may be drifting into human speech, even when those patterns do not reflect natural preferences.
How the study was carried out
Because authorship in written text can be uncertain when AI tools are involved, the team chose to focus on spoken language. They compiled 1,326 podcast episodes from shows such as Lex Fridman, EconTalk, and Radiolab, dividing them evenly between recordings made before and after 2022. Transcripts were then processed with machine learning systems that standardized grammar and marked parts of speech, allowing for comparisons across time.
The results showed that some terms like “delve” rose only slightly, while others, including “align” and “significant,” climbed more consistently. According to the researchers, this selective adoption indicates that people exposed to AI-generated text may be unconsciously echoing its lexical style in conversation.
Possible consequences
If these patterns continue, they could create a cycle where human speech begins to reflect AI usage, which in turn feeds back into the training of future systems. The concern is that this seep-in effect could spread beyond vocabulary and affect the way ideas are framed, eventually shaping broader norms of communication.
At the same time, the evidence is not conclusive. Some of the tracked words were already becoming more common before ChatGPT was released, and language often changes within specific subcultures before it spreads more widely. Because the dataset mainly reflects technology-focused speakers, the results may show an early stage of diffusion rather than a broad shift in English overall.
What comes next
The Florida State researchers plan to expand their work with larger datasets that include more diverse speech communities. They also emphasize the importance of monitoring how model biases may filter into everyday expression. Their findings will be presented at the upcoming Conference on AI, Ethics, and Society, where the work will be discussed alongside broader questions about fairness and the role of artificial intelligence in daily life.
For now, the study indicates that conversational English is already starting to reflect the vocabulary of AI systems, raising questions about how far machine-generated patterns may reach into human communication.
Notes: This post was edited/created using GenAI tools.
Read next: Language Is Changing Again, and This Time, It’s Machines We’re Imitating