Join any Zoom call, walk into any lecture hall, or watch any YouTube video, and listen carefully. Past the content and inside the linguistic patterns, you'll find the creeping uniformity of AI voice. Words like "prowess" and "tapestry," which are favored by ChatGPT, are creeping into our vocabulary, while words like "bolster," "unearth," and "nuance," words less favored by ChatGPT, have declined in use. Researchers are already documenting shifts in the way we speak and communicate as a result of ChatGPT — and they see this linguistic influence accelerating into something much larger.
In the 18 months after ChatGPT was released, speakers used words like "meticulous," "delve," "realm," and "adept" up to 51 percent more frequently than in the three years prior, according to researchers at the Max Planck Institute for Human Development, who analyzed close to 280,000 YouTube videos from academic channels. The researchers ruled out other possible change points before ChatGPT's release and confirmed these words align with those the model favors, as established in an earlier study comparing 10,000 human- and AI-edited texts. The speakers don't realize their language is changing. That's exactly the point.
One word, in particular, stood out to researchers as a kind of linguistic watermark. "Delve" has become an academic shibboleth, a neon sign in the middle of every conversation flashing ChatGPT was here. "We internalize this virtual vocabulary into daily communication," says Hiromu Yakura, the study's lead author and a postdoctoral researcher at the Max Planck Institute of Human Development.
But it's not just that we're adopting AI language — it's about how we're starting to sound. Even though current studies mostly focus on vocabulary, researchers suspect that AI influence is starting to show up in tone, too — in the form of longer, more structured speech and muted emotional expression. As Levin Brinkmann, a research scientist at the Max Planck Institute of Human Development and a coauthor of the study, puts it, "'Delve' is only the tip of the iceberg."
AI shows up most obviously in functions like smart replies, autocorrect, and spellcheck. Research out of Cornell looks at our use of smart replies in chats, finding that use of smart replies increases overall cooperation and feelings of closeness between participants, since users end up selecting more positive emotional language. But if people believed their partner was using AI in the interaction, they rated their partner as less collaborative and more demanding. Crucially, it wasn't actual AI usage that turned them off — it was the suspicion of it. We form perceptions based on language cues, and it's really the language properties that drive those impressions, says Malte Jung, Associate Professor of Information Science at Cornell University and a co-author of the study.
[...] We're approaching a splitting point, where AI's impacts on how we speak and write move between the poles of standardization, like templating professional emails or formal presentations, and authentic expression in personal and emotional spaces. Between those poles, there are three core tensions at play. Early backlash signals, like academics avoiding "delve" and people actively trying not to sound like AI, suggests we may self-regulate against homogenization. AI systems themselves will likely become more expressive and personalized over time, potentially reducing the current AI voice problem. And the deepest risk of all, as Naaman pointed to, is not linguistic uniformity but losing conscious control over our own thinking and expression.
The future isn't predetermined between homogenization and hyperpersonalization: it depends on whether we'll be conscious participants in that change. We're seeing early signs that people will push back when AI influence becomes too obvious, while technology may evolve to better mirror human diversity rather than flatten it. This isn't a question about whether AI will continue shaping how we speak — because it will — but whether we'll actively choose to preserve space for the verbal quirks and emotional messiness that make communication recognizably, irreplaceably human.
See also: Blade Runners of LinkedIn Are Hunting for Replicants – One Em Dash at a Time
« Why Are the Physical Constants of the Universe So Perfect for Life? | Scientists Unlock the Light-Bending Secrets of Squid Skin »
Related Stories
People think the em dash is a dead giveaway you used AI – are they right?
ChatGPT is rapidly changing how we write, how we work – and maybe even how we think. So it makes sense that it stirs up strong emotions and triggers an instinct to figure out what's real and what's not.
But on LinkedIn, the hunt for AI-generated content has gone full Voight-Kampff. According to some, there's now a surefire way to spot ChatGPT use: the em dash.
Yes, the punctuation mark officially defined by the width of one "em." A favorite of James Joyce, Stephen King, and Emily Dickinson. A piece of punctuation that's been around since at least the 1830s. So why is it suddenly suspicious? Is it really an AI tell or punctuation paranoia?
Rebecca Harper, Head of Content Marketing at auditing compliance platform ISMS.online, doesn't think so: "I find the idea that it's some kind of AI tell ridiculous. If we start policing good grammar out of fear of AI, we're only making human writing worse!"
She's right. The em dash isn't some fringe punctuation mark. Sure, it's used less often than its siblings – the en dash and the humble hyphen – and it's more common in the US than the UK. But that doesn't make it automatically suspicious.
Robert Andrews, a Senior Editor, explains that this is a difference in style rather than a smoking gun: "It's not just a marker of AI, but of US English and AP Style. It's quite alien to UK journalism training and style, at least my own, albeit long ago. But increasingly encountered in AP Style environments - (or –, or —) unsurprising that this would flow into LLMs."
[...] Still, because it's slightly less common in some circles, people have latched onto it as a tell. Chris McNabb, Chief Technology Officer at eGroup Communications, makes this case: "I think it's a strong indicator, especially when you see it being used often by one person. Typically most people aren't going to long press the dash key to even use the en dash BUT AI such as ChatGPT uses it by default in a lot of cases. So yes when you do see an em dash particularly more than one in a message it's a pretty safe bet for a majority of posts."
(Score: 1, Funny) by Anonymous Coward on Monday June 30, @02:14PM
3..2..1.. go
https://www.linkedin.com/posts/drjuliaschneider_chatgpt-linkedin-aibullshitbingo-activity-7161342272343539713-F4Ih [linkedin.com]
(Score: 3, Interesting) by OrugTor on Monday June 30, @02:17PM (9 children)
We never had conscious control over verbal expression. Our core expression was formed in childhood when we had little conscious control over any mental development. With adolescence came expansion and refinement as we chose reading matter for the content rather than the prose although we might find an author's writing particularly appealing. ChatGPT's influence on us is no more pernicious than that of our favorite author or podcaster.
There's a work colleague of mine who is an early adopter of bizspeak. I can't believe she consciously chooses to spout verbal abominations such as "the ask".
(Score: 3, Interesting) by gznork26 on Monday June 30, @03:22PM (6 children)
Consequentially, I'm curious about the effects of AI-written and delivered entertainment and instruction for children. After all, reading aloud to kids (or anyone for that matter) is hard work for some and a total blocker for others, so there will be definite appeal to the idea of letting an AI do it. Further, there is a subset of human writers whose work is easy to read aloud. That's a result of the author's attention to the sound of their written words, useful inflection, the cadences in them, and the necessity of breathing while doing it. AI has none of these limitations, and that is reflected in what it produces.
Khipu were Turing complete.
(Score: 2) by VLM on Monday June 30, @04:35PM (5 children)
By analogy with the effect of the calculator, I expect lots of old people avoiding the issue via "what if its not around anymore and you need to calculate a square root by hand or multiply using logarithm tables?" and kids developing habits of extreme laziness; a large fraction of the adult population will now whip out a calculator by default when faced with "whats 12 * 3?" and people who can do that in their head are rare indeed.
So I would expect lots of kids, when asked "do you want to marry me?" or "what should I have for dinner tonight" will ask chatgpt rather than think about it. You already see it impacting search engine use. I still get better results using a search engine but I'm sure AI-generated webspam will eliminate that eventually, and sometimes people don't feel they have time or its not worth the effort.
Also this is still early days, they're in the "build the addiction for free/cheap" phase. Wait until they get addicted and its Cable TV enshitification time. Every user prompt about AWS will result in ten video ads for Azure and GCE before the (honestly probably incorrect) answer is provided and "business class" services like programming will have CAD software style infinitely high prices.
(Score: 3, Interesting) by aafcac on Monday June 30, @05:28PM (4 children)
I've seen that, I remember seeing somebody working through some calculus that didn't have the basic arithmetic to do multiplication without a calculator.
I do think this is likely to be far, far worse. There's already a bunch of algorithms that encourage time wasting on places like YouTube where the ability to trap viewers for longer than necessary helps boost your profile and the number of people that see the content without having to be a subscriber. With AI being used to then take those unnecessarily long videos and summarize it, I wouldn't be surprised if we get to a point where AI is on both sides ultimately fighting over which side is more important and making the average length of a a video an hour so one side gets more views and the other side can use the longer videos as a selling point as to why you need to pay an AI to summarize it for you.
(Score: 3, Insightful) by VLM on Monday June 30, @05:48PM (3 children)
I don't disagree with your post, but I'll embrace and extend your idea above with the concept that once they start aggressively advertising on LLM platforms or aggressively nickel and dime billing users via subscriptions and limits and caps, there will be an INTENSE financial motivation to turn 10 second or 1 minute questions into 1 hour long clickfests.
"Whats 12 * 3?"
Now a days you'd get 36 as the answer, or more likely something wrong like 15 or 9, but once there's a financial motivation to not answer questions immediately, you'll get an hour long back and forth discussion surrounded by endless advertisements and paying all kinds of use fees similar to "would you like to see how to solve that on an abacus?" "would you like me to explain how to solve that in hexadecimal or octal?" "How do you want me to round that, to the closest ten or closest tenth?" "Do you mean in a mathematical sense or philosophical sense of how many angels dance on the head of a pin" "Hold on and watch this viral internet meme video while I 'think' about the correct answer" "First, would you like to see an advertisement for Texas Instrument's newest graphing calculator now with AI features and Facebook integration?" "I will demonstrate how to solve this using the long division box-method by dividing by the reciprocal instead of multiplying; would you like a short half hour lecture on the algebraic concepts explaining that?" "Are you sure you want to multiply, I could add for you instead" "Are you trying to solve a problem for your income tax return? Because Turbotax is now running a sale of 5% off their AI accelerated product ... let me find you a nice video commercial to watch about that topic" "here's a helpful tiktok video I found of a school teacher telling kids to 'do their own damn homework' that is trending on insta right now"
Ironically, this might destroy the meme of using AI to do kids homework; if the AI insists on taking two hours to show high added sugar breakfast cereal advertisements while kids can do their own homework in only one hour, kids will not use AI to do homework anymore.
(Score: 1) by khallow on Monday June 30, @06:49PM
(Score: 2) by PiMuNu on Tuesday July 01, @05:38AM (1 child)
That's already the case on many crapsites. Try searching for some common thing like conditional syntax in bash scripting (I can never remember) and you will get a whole load of junk before the actual answer.
what is conditional syntax
what is bash
different boolean logics
how to write an if statement in bash
(Score: 3, Insightful) by VLM on Tuesday July 01, @04:34PM
AI generated webslop food recipe sites are worse.
A PB&J sandwich includes the entire script of a Food Network miniseries about nonsense before it begins the ingredients. All wrapped in heavy layers of advertisements.
(Score: 0) by Anonymous Coward on Monday June 30, @05:15PM (1 child)
Interesting, after finishing a rather long book series, I chose to reread Tolkien specifically to change my working vocabulary.
(Score: 1) by khallow on Monday June 30, @07:17PM
(Score: 1) by RMS06510 on Monday June 30, @03:40PM (4 children)
so-called "AI" has had no effect on the error rate of autocorrect, which for me still hovers around 75%
(Score: 0) by Anonymous Coward on Monday June 30, @04:14PM
Haven't turned off autocorrect yet? Poor little.
(Score: 3, Insightful) by cmdrklarg on Monday June 30, @08:56PM (2 children)
AutoIncorrect gets disabled the first chance I get. I know what I'm typing dammit. I'm fine with it suggesting the correct word or spelling, but DO NOT change what I'm typing.
The world is full of kings and queens who blind your eyes and steal your dreams.
(Score: -1, Troll) by Anonymous Coward on Monday June 30, @10:38PM (1 child)
Is it Dunning-Kruger or dementia? Sorry, it just had to be asked.
(Score: 2) by cmdrklarg on Thursday July 03, @04:04PM
Is the AC a simple-minded asshole or a just a total fuckwad? Well, to be honest it did not need to be asked.
The world is full of kings and queens who blind your eyes and steal your dreams.
(Score: 3, Interesting) by VLM on Monday June 30, @04:37PM (1 child)
Oh I'm in a hurry I will just ask chatgpt for a simple inspirational quote to drop in here nobody will notice. I'd never write "my" entire thingie using AI but I just need one quote or idiom or analogy. Times a million people and suddenly for no apparent reason the word "tapestry" is trending.
(Score: 2) by krishnoid on Monday June 30, @05:08PM
+1, Inspirational.
(Score: 1) by khallow on Monday June 30, @04:44PM (2 children)
(Score: 0) by Anonymous Coward on Monday June 30, @05:16PM (1 child)
Came here to say something similar. 30 years ago I co-authored an 800 page engineering reference book. Just did a search on the text, no hits on "delve" at all.
Then I searched an engineering biography I helped proofread - one use of "delve" in 400 pages.
Next, an ebook version of "The Power Broker" by Robert Caro (1200 page biography, 1975) - 4 uses of "delve" in the whole book.
More generally, Google Ngram searches books (not the web) for word counts, currently only up to 2022. There was a spike in delve around 2015, but returned to prior level by 2022. Guess we'll have to check back in a few years to see this latest "AI" spike?
https://books.google.com/ngrams/graph?content=delve&year_start=1950&year_end=2022&corpus=en-2019&smoothing=3&case_insensitive=true [google.com]
(Score: 2, Funny) by khallow on Monday June 30, @07:13PM
(Score: 2) by krishnoid on Monday June 30, @05:17PM
+1 randomly selected mod, for examples of a sentence that uses all the ChatGPT-preferred words in a sentence. Kudos in a reply, for an adjacent sentence (or even the same one) using the less-favored words.
(Score: 2) by Deep Blue on Monday June 30, @05:24PM (3 children)
You lost me there.
(Score: 5, Informative) by VLM on Monday June 30, @05:58PM (2 children)
Zoom is extremely corporate so most of the "speech" heard over it will be along the lines of "We can circle back to the low hanging fruit because at the end of the day my reskilling assistant can reach out to the other department's augmented workforce and their talent cloud, to proactively leverage our digital twin process to accelerate the agile transformation while we polywork our quantum advantage via our adaptive workflows"
(Score: 3, Informative) by krishnoid on Monday June 30, @06:07PM
Next week, on "Utopia" [youtu.be].
(Score: 2, Interesting) by anubi on Tuesday July 01, @06:46AM
New AI technology gives even the most budget-conscious entity the telephone presence of a major corporate enterprise.
https://jollyrogertelephone.com [jollyrogertelephone.com]
"Prove all things; hold fast that which is good." [KJV: I Thessalonians 5:21]
(Score: 3, Funny) by Rich on Tuesday July 01, @01:00AM (1 child)
"delve" is what delvers do.
(Score: 2) by Rich on Tuesday July 01, @10:38AM
... and I first read that moderation as (Score: 3, Furry) ...
(Score: 3, Informative) by stormwyrm on Tuesday July 01, @03:20AM
It's so funny that Brandon Sanderson's young adult space opera book series has enemies called "Delvers". The Delvers are ancient, extra-dimensional beings that normally inhabit the Nowhere, a hyperspace that enables superluminal travel among other technologies, and make injudicious use of such technology perilous, and they are highly destructive when they emerge into normal space.
The first book in the series came out in 2018, a few years before ChatGPT was released to the public.
Numquam ponenda est pluralitas sine necessitate.
(Score: 2) by YeaWhatevs on Tuesday July 01, @04:43PM
ChatGPT has poor word economy.