In an era that celebrates diversity in all its forms — of skills, gender, opinions, and cultural expressions — an alarmingly uniform pattern is emerging at an accelerating pace: the way we write.
(Fun fact: one of the most recurrent patterns in AI-generated text, previously less common among humans, is the use of the em dash. To allow myself to use it, I went back to my own writing from before the AI era and confirmed I had already used it frequently. So I can say that, in this case, the AI copied me. 🙂
Researchers at the Max Planck Institute for Human Development in Germany analyzed over 740,000 hours of academic lectures and podcast episodes transcribed from YouTube, from before and after the launch of ChatGPT. They identified a set of words the model tends to favor when editing text — terms like delve, boast, meticulous, and swift, nicknamed “GPT words” — and tracked how often these appeared in spontaneous speech over time. The result was a sharp and measurable spike in the use of these words after ChatGPT, even in unscripted speech contexts, where no one is reading AI-generated text. A sister study conducted at Florida State University found a similar pattern when analyzing science and technology podcasts: nearly three-quarters of AI-associated words increased in usage, some more than doubling in frequency, while their common synonyms remained flat.
The fact is, people aren’t just quoting AI. They’re absorbing its vocabulary as if it were their own, without even noticing. And generating more content based on that same vocabulary — which, you guessed it, AI uses to train itself. Can you imagine the long-term impact of this?
Language Homogenization: A Silent Threat to Diversity of Thought
We’ve previously explored here at Action Labs Insights the benefits of a diverse environment for innovation, in our article on Innovation and Multidisciplinarity. But here we’re talking about something even deeper than the influence on corporate innovation capacity. When everyone writes (and, more broadly, expresses themselves) in a similar way, it’s not just a matter of style and form. We must first understand that the narrative aspect of language is only one of its dimensions. The more complex uses of our expressive capacity are less about describing events and situations and more about elaborating concepts and developing lines of reasoning. And the elaboration of concepts is closely tied to the neuroplasticity of our brain — to the reinforcement of neural pathways that, as a consequence, shape our analytical ability, our mental models, and our capacity to create new connections. In this sense, studies point out that language doesn’t determine how we think, but it strongly influences what distinctions we make, what we perceive more easily, what we remember, and how we organize ideas. A fascinating example of this comes from Australian indigenous tribes that don’t use “left” and “right,” only north, south, east, and west — and as a result, develop extraordinary spatial orientation abilities. It may seem like an extreme comparison, but considering the long-term effects of a sweeping standardization in form and vocabulary, what will this do to the diversity of our worldviews?
How AI Reinforces Dichotomies and Fuels Polarization
One of the most visible linguistic examples is the parallel between the polarization of opinions and one of AI’s most persistent linguistic habits: the dichotomy. A dichotomy is a figure of speech that divides the universe of possibilities into two sides. Alive vs. Dead. True vs. False. Right vs. Wrong. A true dichotomy leaves no room for a third option. But genuine dichotomies are actually quite rare and abstract. The real world is made of nuance, and the vast space of discussion between extremes is, in fact, where real debate actually lives. The extremes are just that — extremes.
The problem arises with false dichotomies, which foreclose multiple alternatives not because they are truly unviable, but because the false dichotomy artificially limits the universe of possible answers. Body vs. Mind. Important vs. Trivial. Smart vs. Dumb. Who isn’t familiar with AI-written articles that open with “It’s not about X, it’s about Y”? This is the track along which the possibility of structured, relevant, and in-depth debate is drastically reduced. Human discourse already conditioned by social media algorithms has largely produced this effect over the last decade and a half — debate between extremes, with a deafening noise that drowns out the voices that understand dichotomies can’t capture the full complexity of reality. But now, inevitably and before long, most of the texts that will reach us will be AI-generated (we covered this in a Friday Insights on AI and Elections, here at Action Labs). Our challenge, as creatives, communicators, and knowledge producers, is to ensure that this entrenched linguistic structure, despite being the majority, does not become the dominant voice.
The Human Voice as the Last Line of Defense for Creativity
The temptation is to treat this as a linguistic curiosity — just another buzzword entering the collective lexicon, as happened with internet or business jargon in other eras. There is, however, an important structural difference: this time, the source is not a social group, a cultural event, or a community — it’s a single model, trained on a specific statistical distribution of language, simultaneously talking to hundreds of millions of people. It’s not hard to see the risk we run when everyone is exposed to a single omnipresent interlocutor — AI — writing on our behalf and using our interactions to universalize its writing style, and, consequently, to universalize the very way we conceive ideas. The impact of this is absolutely devastating.
For an Innovation Studio like Action Labs, this behavior is a direct warning signal about the work process itself. Structured creativity depends on divergent thinking: generating as many possible solutions as possible before converging, questioning even the favorite idea, resisting the first “obvious” answer. Our PoC Design process — like all established methodologies in the field — has advocated for over 10 years for extreme care in the immersion and problem-framing stages. It’s essential to deeply understand the personas and the full work universe before beginning to imagine solutions, lest we fall in love with a first version of a product too early. Part of this methodology is Appreciative Inquiry — a process that begins with reflection on ideal scenarios even before any benchmarking in the real world — to avoid being boxed into the spectrum of solutions already tried by competitors (and by ourselves). If looking at the competition too early already implies a drastic narrowing of our field of vision, imagine what happens when company executives start approaching any new problem by asking the same AI that the rest of the world is asking. It is absolutely essential, in innovation methodologies, to know the right moment to bring in AI’s analysis. And that moment is not in the early immersion phases. On the contrary, turning to it at this stage is fatal to divergent thinking — and it is our job, as creatives, to lead the resistance against it. To quote Naoki Tanaka, Global Chief Creative Officer of Dentsu Lab:
The essence of AI is to average out the entire vast history of humanity’s ideas and products. In doing so, it resets the race of creation. Imagine everyone back at the starting line. A new contest begins: who is capable of breaking free? Who can stand out? This forced reset could usher in an era of unprecedented innovation — but it will demand new sensibilities and an obsessive stubbornness for the truly new.
Appendix for Skeptics on the Impact of Language on Thought
If you’re still not convinced of the depth of change that linguistic homogenization can bring to how we think, here are some research areas to explore:
Vocabulary Defining the Personality of Objects
Research shows that grammatical gender influences the associations we make with objects. Speakers of different languages tend to describe the same things with distinct characteristics. Sources: https://www.edge.org/3rd_culture/boroditsky09/boroditsky09_index.html • https://escholarship.org/uc/item/3j37f380
Words Influencing What the Eyes See
Languages categorize colors differently. These differences alter the speed and accuracy with which people identify and remember certain shades. Source: onlinelibrary.wiley.com
Sense of Direction Shaped by Grammar Some languages use only cardinal directions instead of “left” and “right.” Their speakers develop far more precise spatial orientation because they constantly exercise that ability. Source: https://www.edge.org/3rd_culture/boroditsky09/boroditsky09_index.html
Language Giving Direction to Time The way a language describes time influences its mental representation. English speakers tend to organize it horizontally; Mandarin speakers respond more naturally to vertical representations. Source: https://onlinelibrary.wiley.com/doi/10.1111/lang.12195
Verb Conjugation Recording the Guilty and the Innocent Languages that require identifying who is responsible for an action make their speakers better at remembering who did what. Grammar directs attention and influences what gets encoded in memory. Sources: https://www.edge.org/3rd_culture/boroditsky09/boroditsky09_index.html • https://www.annualreviews.org/doi/10.1146/annurev.anthro.26.1.291
