I've started catching myself doing something I never expected from an AI assistant: borrowing its vocabulary.
I'll write an email, edit a sentence and suddenly wonder why I used a word I barely used before ChatGPT became part of everyday work.
Then someone else says the same thing in a meeting.
That is the strange part.
AI may be influencing human language not because machines learned how humans talk, but because humans are now learning how machines talk.
The phenomenon is being called “AI-speak,” “Claude speak” and “LLM-isms.” The names are informal, but the underlying question is serious: what happens when millions of people spend hours every day reading and editing AI-generated language?
AI-generated language is increasingly becoming part of everyday human writing and conversation.
What Is “AI-Speak”?
There is no official linguistic definition of AI-speak.
The phrase generally describes recurring words, sentence structures and rhetorical habits associated with large language model output that then start appearing in human writing or speech.
The Wall Street Journal recently described people adopting formal, punchy and highly structured expressions associated with chatbots such as ChatGPT and Claude.
Some people notice particular words. Others notice the rhythm of the writing.
That's an important distinction.
The tell is often not one word. It is a cluster of habits appearing together.
Why “Delve” Became the Poster Child
“Delve” became famous online because people began noticing how often it appeared in AI-generated writing.
A 2025 ACL study examined the phenomenon in scientific writing and identified a set of words whose usage increased in ways associated with LLM adoption, including “delve,” “intricate” and “underscore.”
Another large-scale study of more than 2.4 million PubMed Central publications found substantial increases in several LLM-associated terms after the launch of ChatGPT.
The researchers reported that usage of “delve” increased by about 1,500% between 2022 and 2024 across the academic databases they studied.
Why Do AI Models Favor Certain Words?
This question is harder than it looks.
One 2025 computational-linguistics study investigated 21 words that became unusually common in scientific writing and tested possible explanations for why language models overuse them.
The researchers found no simple evidence that model architecture alone caused the effect.
The study explored whether reinforcement learning from human feedback might contribute to some of the patterns, but the authors emphasized that the source of lexical overrepresentation remains an open research question.
That uncertainty matters because people online often repeat a much simpler explanation as fact.
The real answer is probably more complicated than “ChatGPT was trained on people who say that word.”
Now Humans Are Feeding the AI Back
This is where the story gets genuinely strange.
AI systems learn from human-produced language.
Humans then read AI-generated language and reuse parts of it.
That creates the possibility of a feedback loop.
The Language Feedback Loop
- Step 1: Human writing influences model training.
- Step 2: The model produces statistically favored patterns.
- Step 3: People read those patterns thousands of times.
- Step 4: People reuse some of them in emails, papers and conversations.
- Step 5: Future training data can contain more AI-influenced human language.
That does not prove a runaway linguistic cycle is happening at every level of English.
But it is a plausible mechanism, and researchers are already studying human-LLM coevolution.
Academic Writing Shows the Effect Clearly
Researchers have an unusual advantage here: millions of published documents can be measured over time.
A study published through the Association for Computational Linguistics found that several words associated with ChatGPT changed sharply in academic abstracts following the model's release.
Interestingly, some words that had become strongly associated with ChatGPT later declined after people became aware of them.
That creates an unexpected second effect.
Once humans realize a word sounds like AI, they can deliberately stop using it.
So the language is changing in both directions: people adopt AI patterns, then consciously edit some of them back out.
The “Humanizer” Problem Is More Complicated Than a Word List
Millions of people now use prompts that tell AI to “sound human.”
They remove words such as “delve,” “leverage” or “robust.”
They replace long sentences and polish the result.
But that does not necessarily restore an authentically human voice.
It can simply produce a different kind of standardized AI writing.
Overlooked Tip: Edit for Experience, Not Just Vocabulary
When revising AI-assisted writing, don't merely replace suspicious words. Add specific details, personal observations, concrete examples, genuine uncertainty and the way a real person would naturally explain the idea. Voice comes from experience and judgment, not from a list of banned words.
People Are Copying the Rhythm, Too
This may be harder to notice than vocabulary.
AI-generated writing often favors clean transitions, symmetrical structures and explicit summaries.
It frequently announces the point, explains the point and then restates the point.
Human writers do this too.
That is precisely why no single sentence structure can prove AI involvement.
But repeated exposure can normalize a particular cadence.
You start reading it everywhere.
Then you start writing it yourself.
Claude Is Part of the Same Story
The phenomenon is not limited to ChatGPT.
Anthropic describes Claude as a model whose response style can be deliberately adjusted through presets and custom styles, including formal, concise and explanatory modes.
Anthropic's own research also measures differences in reading level between user prompts and Claude's responses, finding that Claude's output is often more polished or formal than the prompt itself.
That creates another route by which AI language can become normalized.
Users don't just consume the content.
They consume the style.
George Orwell Saw the Basic Mechanism Decades Ago
The technology is new. The underlying social mechanism is not.
George Orwell argued in his 1946 essay Politics and the English Language that language patterns can spread through imitation and eventually influence thought.
“But if thought corrupts language, language can also corrupt thought.”
Orwell was writing about political and bureaucratic language, not artificial intelligence.
But the broader idea is strikingly relevant today: repeated exposure to ready-made phrases can change what feels normal to say.
Adam Aleksic Says We Should Ask One Question
Etymologist and creator Adam Aleksic made this phenomenon the subject of a 2025 TED talk titled Why are people starting to sound like ChatGPT?
His argument is broader than vocabulary.
He says algorithms and AI can influence the way people perceive what is normal, popular and real, and that users should become more conscious of that influence.
The question is simple:
Ask the same question you would ask about a social-media feed: why did this particular phrase become my default way of expressing the idea?
Watch the TED Talk
Adam Aleksic explains the “delve” phenomenon and the broader relationship between AI, algorithms and human language in this TED talk.
Is AI Actually Making Everyone Sound the Same?
| Influence | What We Can Observe | What It Does Not Prove |
|---|---|---|
| Vocabulary | Some LLM-associated words have increased in measured corpora. | That every person using those words used AI. |
| Structure | AI outputs often use polished, highly organized prose. | That humans did not use those structures before AI. |
| Tone | AI can default toward polished, formal or affirming language. | That all AI systems have the same personality or style. |
| Human adaptation | People can adopt or deliberately avoid AI-associated patterns. | That AI has permanently changed English. |
Pros and Cons of AI Influencing Human Language
Potential Benefits
- People may discover vocabulary they previously lacked.
- AI can make complicated writing clearer and easier to organize.
- Non-native English speakers can gain confidence communicating in English.
- Writers can experiment with different tones and structures.
- AI can expose people to new ways of expressing an idea.
Potential Downsides
- Workplace language can become more standardized.
- Personal voice can be diluted through repeated AI editing.
- Rare or distinctive expressions may be used less often.
- People can mistake polished AI language for naturally occurring language.
- AI-assisted communication can become overly formal or repetitive.
Amazon: Books for Better Human Writing
Politics and the English Language
Orwell's essay remains a useful reminder to question inflated language, ready-made phrases and writing that hides rather than clarifies an idea.
Find Orwell's Essay on AmazonThe Elements of Eloquence
Mark Forsyth's guide explores the rhetorical structures that make language memorable, giving writers a useful counterweight to standardized AI prose.
Check The Elements of EloquenceOn Writing by Stephen King
Stephen King's writing memoir and craft guide is a practical reminder that strong writing depends on clarity, observation and a distinct human voice.
Check On WritingHow to Keep Your Voice When Using AI
You don't need to stop using ChatGPT or Claude.
You need to remain the editor.
A Better AI-Writing Workflow
- Write the first idea yourself: Give AI your thinking instead of asking it to invent your entire voice.
- Ask for criticism: Use AI as an editor before using it as a ghostwriter.
- Keep your odd phrases: Distinctive language is often part of your personality.
- Replace generic examples: Add details only you would know.
- Read the final version aloud: Your ear catches robotic cadence better than a keyword detector.
OpenAI itself recommends using ChatGPT for writing as a drafting and feedback tool while treating the output as something to review rather than a final authority.
That is probably the healthiest model for human-AI writing: AI can help with the mechanics while the human remains responsible for the meaning.
The Bottom Line: AI Is Becoming Part of English
Languages have always changed through imitation.
People copy words from friends, celebrities, books, television, workplaces and online communities.
AI is simply becoming another major source of linguistic influence.
The difference is scale.
A single chatbot can generate millions of sentences every day, and those sentences are now being read, edited, repeated and incorporated into real conversations.
That does not mean English is becoming “AI language.”
It means human language and machine-generated language are beginning to influence each other.
Some borrowed expressions will disappear.
Others will become normal English and eventually lose their AI association completely.
That has happened with countless words and expressions throughout history.
The interesting question is not whether language will change.
It always does.
The interesting question is whether we notice the change while it is happening.
Because once an AI phrase feels completely normal coming out of your own mouth, it may already be part of the language you use to think.
Inside the 2026 AI Misuse Epidemic
While subtle shifts in human language show how deeply AI is integrating into daily life, the most dangerous impacts are happening completely out of sight. Read our complete guide on the real, overlooked threats of AI misuse in 2026 to understand the security risks and algorithmic dangers no one is talking about.
Read the AI Threat Guide →Sources checked for this article:
The Wall Street Journal — Now We're All Starting to Talk Like AI Chatbots Too
ACL Anthology — Why Does ChatGPT “Delve” So Much?
ACL Anthology — Human-LLM Coevolution: Evidence from Academic Writing
Scientometrics — How Much Are LLMs Changing the Language of Academic Papers?
TED — Why Are People Starting to Sound Like ChatGPT?
Frequently Asked Questions
Why are people starting to sound like ChatGPT?
Frequent exposure to AI-generated writing may lead people to reuse its vocabulary, sentence structures and rhetorical habits. Current reporting and research suggest that some AI-associated language patterns are becoming more common in human writing.
What words are associated with AI writing?
Researchers have identified terms including “delve,” “intricate,” “underscore,” “meticulous” and “significant” as words whose usage has changed in some corpora in ways associated with LLM adoption. None of these words individually proves that text was written by AI.
Does using the word “delve” mean something was written by ChatGPT?
No. “Delve” is a normal English word and predates modern AI by centuries. Researchers study changes in its frequency at the population level; an individual use cannot establish AI involvement.
Is AI changing the English language?
Evidence suggests AI is influencing vocabulary and writing style in some settings, especially academic and professional writing. It is too early to conclude that AI has permanently changed English as a whole.
How can I use AI without losing my writing style?
Give AI your own ideas and examples, ask it for criticism instead of complete rewrites, keep distinctive expressions that sound like you, and read the final version aloud. Voice comes from the underlying ideas and experiences, not only word choice.
No comments:
Post a Comment