Chatbots Are Changing How People Talk

Woman wearing headset on video call — Photo: fizkes / Shutterstock

People are starting to sound like their chatbots, and researchers can now prove it with numbers.

Quick Take

A new study found a real jump in ChatGPT-style words in everyday speech after the tool launched.
Scientists say humans naturally copy the speech patterns of whoever, or whatever, they talk to most.
Writing experts warn that asking AI to “clean up” your words often erases your personal voice.
Speech researchers say hesitations, accents, and even stutters carry identity that polished AI text strips away.

The Evidence That We Are Already Copying The Machines

A recent research preprint tracked word usage before and after ChatGPT’s release and found something striking. Certain words tied closely to ChatGPT’s writing style suddenly spiked in how often people used them out loud. The researchers call this the first hard proof that humans are picking up speech habits straight from a chatbot. That is not a guess or a hunch. It is a measurable shift in how ordinary people talk.

This matters because language change used to take generations. A slang word might spread through a school, then a city, then maybe a country, over years. Now a chatbot used by hundreds of millions of people can nudge word choices within months. Once you notice it, you start hearing it everywhere: friends saying “delve,” “boast,” or “underscore” in casual conversation, words that used to sound stiff outside a boardroom.

Why We Copy The Voices Around Us

Linguists have a name for this old, well-documented habit: entrainment. It means two people talking to each other slowly start sounding alike, matching pace, tone, and word choice without even noticing. This has always happened between humans. Married couples pick up each other’s phrases. Coworkers start finishing sentences the same way. The new twist is that one of the “speakers” in millions of daily conversations is now a machine.

Researchers studying how people talk to computers found the same pattern holds true there too. People adjust their speech in real time based on what they think a computer needs to understand them, speaking slower, simpler, and more direct. That habit does not stay locked inside the conversation with the machine. It bleeds into how we talk to our spouse, our boss, and our kids at the dinner table.

What Gets Lost When Speech Gets “Polished”

A guide for scientists and writers put it bluntly: telling an AI tool to “improve” or “clarify” your writing often flattens your voice into something generic. The fix, the guide argues, is to deliberately put back your own hedges, your own hesitations, your own way of saying things. That advice matters because AI-written text tends to be suspiciously smooth. One writing analysis found that machine text often uses the same tidy sentence rhythms over and over, while real human writing keeps small rough edges that are not textbook perfect.

Those rough edges are not mistakes. They are fingerprints. A pause before answering a hard question, a stumble over a word, a sentence that trails off and restarts, these are the things that make a voice sound like a person and not a script. Commentary in The Guardian warns that as people lean on AI more, they may start copying its style directly, even though AI struggles to recreate the messy, spontaneous rhythm of real spoken language.

Accents, Stutters, And The Cost Of Sounding “Normal”

The stakes go beyond sounding a little too formal at brunch. Research on speech recognition systems shows that tools trained mostly on smooth, standard speech often fail people with accents or stutters entirely. When technology only rewards one “normal” way of talking, it quietly punishes everyone who does not sound that way. That is a real problem for anyone who values regional dialects, immigrant accents, or just plain human variety over robotic sameness.

Efforts to preserve regional voices treat accents and speech patterns as a kind of heritage worth protecting, not a flaw to be corrected. That framing feels right. A grandmother’s Southern drawl, a Boston accent, a halting speech pattern from a stroke survivor relearning to talk, none of that is broken. It is identity. The danger is not that AI exists. The danger is treating efficiency as the only virtue and letting genuine human texture get sanded down in the process.

Nobody is claiming every filler word or awkward pause makes a person more trustworthy in every setting. Context still matters, and plenty of professional situations reward clear, tight language. But the evidence so far points one direction: speech is already drifting toward machine patterns, and keeping our own voice takes a conscious choice, not an accident.