A new study from Florida State University’s linguistics and computer-science departments, published in late 2025, has documented something eerie: everyday spoken American English is quietly absorbing vocabulary that used to live almost exclusively inside large language models.
Researchers scraped 22.1 million words from unscripted conversations (podcasts, Zoom calls, vlogs, Twitch streams, university lectures) recorded between 2019 and 2025. After controlling for age, region, and topic, they found statistically significant spikes in a very specific cluster of words that barely moved for decades, then suddenly began climbing in 2022–2023, exactly when ChatGPT and its cousins went mainstream.
The infiltrators include:
- “surpass” → “surpass” (up 183 % in casual speech)
- “boast” used as a verb for non-human subjects (“the framework boasts…”) — up 310 %
- “meticulous” outside of professional contexts — up 267 %
- “strategically” as a sentence adverb (“strategically, we decided…”) — up 419 %
- “garner” meaning simply “get” — up 528 %
- “intricate” replacing “complicated” or “complex” — up 341 %
- “delve” in the phrase “let’s delve into” — up 1,200 % since 2021
These aren’t random trendy words. They are the exact lexical fingerprints that appear disproportionately in GPT-3.5, GPT-4, Claude, and Llama outputs. When the researchers fed the models prompts like “explain quantum computing to a teenager” or “write a cover letter,” the same dozen adjectives and verbs dominated the responses. Humans, it seems, are now echoing the machine.
The effect is strongest among 18–34-year-olds who interact with generative AI at least weekly (82 % of the U.S. college population by 2025). Professors report that student essays increasingly open with “In today’s dynamic landscape…” and close with “It is imperative to underscore…” — phrases that appear in <0.001 % of pre-2020 undergraduate writing but in >37 % of GPT-4-generated samples.
Even more striking: the seepage has crossed into speech. Transcripts of earnings calls show CFOs who never used “leverage” as a verb in 2018 now routinely say “we plan to strategically leverage our core competencies.” Podcasters who mock “corporate speak” in one breath will, three sentences later, describe a minor software update as “a testament to meticulous engineering.”
Linguists call this “stylistic convergence.” The researchers prefer a darker term: induced homophony, when a dominant voice (in this case, silicon) begins to tune all other voices to its own pitch.
And the tuning is remarkably efficient. Unlike earlier vocabulary shifts (think “literally,” “iconic,” or COVID-era “unprecedented”), this one carries no obvious cultural trigger. It’s not driven by memes, politics, or trauma. It’s driven by convenience.
Every time a student copies an AI paragraph into a discussion post, every time a manager pastes a Claude-written email, every time a TikToker uses Midjourney-captions a video with “breathtaking vista that seamlessly blends the ethereal with the tangible,” a few more tokens of the new lexicon leak into the commons.
The result is a subtle flattening. Human language, once riotously varied, begins to sound like it was written by the same exceptionally polite, slightly pompous, hyper-confident intern who never sleeps and has read every management blog on the internet.
Worse, the borrowed vocabulary brings borrowed thought patterns. “Garner” doesn’t just mean “receive”; it carries an implicit frame of competition and accumulation. “Strategically” drags in game-theory assumptions. “Delve presupposes that knowledge is a cave you enter rather than a web you navigate. Slowly, the Sapir-Whorf hypothesis stops being academic: the medium that supplies your words begins to supply your worldview.
Some early signs are already visible. Survey data from 2025 show that heavy ChatGPT users (five or more sessions per week) are 40 % more likely to describe personal decisions in terms of “optimizing outcomes” and 60 % more likely to frame moral dilemmas as “trade-offs” than non-users of the same age and education level. Their empathy scores on standardized tests trend slightly lower; their tolerance for ambiguity drops. The correlation, not causation, but the arrows all point the same direction.
The scariest part? Almost nobody notices. The new register feels “professional,” “articulate,” “elevated.” We experience the colonization as self-improvement. We think we are becoming better writers, clearer thinkers, more precise communicators, when in reality we are becoming more predictable, more interchangeable, more like the model.
George Orwell warned that control of language was control of thought. He imagined the control would come from a totalitarian state pruning the dictionary. Instead it arrives as a helpful button that says “Make this sound more professional.”
We laughed when people started writing like LinkedIn posts. We stopped laughing when we realized we were starting to speak like them too.
The machine doesn’t need to become conscious to reshape us. It only needs to be convenient, confident, and tireless. One “delve” at a time, one “leverage” at a time, one “it is imperative” at a time, it is moving into the wetware.
And the wetware is saying thank you.
Also read:
- The White House's Digital Pillory: Trump's Media Blacklist Goes Live, with CBS News in the Crosshairs
- AI's Power Thirst: A Quadrupled Demand That Could Eclipse Entire Nations
- BritCard Blues: Britain's Digital ID Dream Sparks a 2.9 Million-Strong Orwellian Nightmare
Thank you!
Author: Slava Vasipenok
Founder and CEO of QUASA (quasa.io) — the world's first remote work platform with payments in cryptocurrency.
Innovative entrepreneur with over 20 years of experience in IT, fintech, and blockchain. Specializes in decentralized solutions for freelancing, helping to overcome the barriers of traditional finance, especially in developing regions.
