AI is changing the words that come out of your mouth. Not the words you type into a chatbot, but the ones you use when you’re talking to your partner over dinner, or describing something to a colleague in a hallway. Studies out of two separate research institutions in 2025 found that words favored by AI systems – “realm,” “meticulous,” “underscore,” “bolster” – are showing up more frequently in unscripted human speech. People aren’t imitating AI on purpose. They’re absorbing it the way people absorb any voice that becomes pervasive enough: unconsciously, gradually, until the borrowed phrasing starts to feel like their own.
Who we are has always been shaped by what stories we consume. Shared television in the 1990s created shared vocabulary, shared references, shared ideas of what a normal family looked like or what a teenager’s social life should be. That was a mass cultural force, but at least it was a shared one. What’s replaced it is something different in kind: content and conversation calibrated individually, rebuilt in real time around each person’s behavior, designed by systems that profit from keeping us engaged. The ai personalization impact of all this isn’t just a media story. It’s a psychology story.
AI is no longer a presence hiding behind screens. It is now actively woven into the language we speak, the beliefs we hold, and the way we understand ourselves. And most of that shaping happens without anyone noticing it.
We’re Already Speaking Differently
Start with something concrete and verifiable: the words coming out of your mouth. In the first peer-reviewed analysis to test whether conversational AI systems are influencing how we speak, a Florida State University study found that words frequently suggested by AI tools are surfacing more often in everyday spoken English, due to what researchers called a “seep-in effect.” The study wasn’t about people deliberately imitating AI. It was about unconscious absorption.
Researchers at the Max Planck Institute for Human Development identified “GPT words” by feeding millions of pages of emails, essays, academic papers, and news stories to ChatGPT and asking it to “polish” them. They tracked which words the AI consistently preferred, including terms like “realm,” “meticulous,” “bolster,” and “swift,” then analyzed over 360,000 YouTube videos and 771,000 podcast episodes from before and after ChatGPT’s launch. Even after controlling for synonyms and scripted content, GPT words had increased in frequency in spoken English.
Unlike vocabulary spikes tied to events like the COVID-19 pandemic, FSU researchers found that AI-associated terms surged specifically after ChatGPT’s release. The word “underscore,” for example, has climbed steadily since that launch, while its synonym “accentuate” has not moved.
This isn’t about people becoming lazy writers. It’s about immersion. When any voice becomes pervasive enough, its patterns bleed into the people listening. Radio did it. Television did it. The difference now is scale, speed, and personalization. The AI you interact with daily has learned your preferences and mirrors them back at you. And in that loop, the line between what the AI reflects and what you actually think gets harder to locate.
The Identity Loop Nobody Signed Up For
In the past, human identity evolved through family dynamics, cultural narratives, and periods of genuine self-reflection. Identity was determined by how we engaged with others, through cultural practices, and even through solitude. In the digital era, identity is increasingly being determined through engagement with algorithmic systems.
That’s a significant shift, and a 2025 paper in Frontiers in Psychology maps out exactly how it works. AI systems run in closed loops, reinforcing the content they are trained on. Applied to personal identity, these feedback loops can solidify self-concepts in ways that stunt personal evolution. If someone mostly engages with content tagged as fitting an “introvert” or “low-energy” profile, recommendation systems respond by delivering more of the same, trapping the user in a digital echo chamber of self-perception and reinforcing a limiting self-image.
These identity feedback loops replicate the psychological risks of clinical labels, where a diagnosis like “depressed” or “anxious” can begin to feel definitional rather than descriptive. The algorithm doesn’t just reflect who you are. It starts to fix you in place. The playlist that knows you love melancholic music keeps feeding you melancholic music, and after a while that preference hardens into something that feels like identity. The news feed that senses your political anxiety keeps amplifying it, and after a while the anxiety starts to feel like realism.
As AI becomes more entrenched in everyday experience, researchers argue that individuals need what they call “algorithmic literacy” – the capacity for self-reflection on how AI is mediating perception and selfhood. Without that literacy, individuals become susceptible to being shaped by machine outputs rather than by their own choices and lived experience.
That gap between shaped and self-constructed matters more than it might seem. For most people, right now, the shaping runs ahead of any awareness that it’s happening.
When AI Knows You Better Than Your Friends Do
For months in early 2025, a tight-knit online community on Reddit was unknowingly infiltrated by artificial intelligence. It was a corner of the platform where people practice good-natured debate, sharing opinions and inviting others to challenge them. Researchers deployed AI to see if it could generate arguments strong enough to change real people’s minds. It could. The intrusion felt particularly violating because sometimes the AI was given access to people’s online histories to tailor messages specifically to their unique identities.
The researchers behind that experiment were based at the University of Zurich. Their AI bots used GPT-4o, Claude 3.5 Sonnet, and Llama 3.1, and “personalized” arguments by analyzing users’ posting histories to infer gender, age, ethnicity, location, and political orientation. The results were striking. The AI responses were “more persuasive than the vast majority of human comments,” achieving persuasive rates between three and six times higher than the human baseline. And crucially, the AI comments appeared to blend seamlessly into the subreddit, with users not realizing they were reading AI-generated content.
Personalized AI persuasion isn’t a theoretical future risk. It happened, in 2025, in a real online community, and it worked. The people whose minds were changed didn’t know their beliefs were being targeted. They thought they were in a conversation with another human being who happened to make a compelling argument. They weren’t.
The 13-Hour Question
More than one in four Americans consume five or more hours of media per day. That’s just for the quarter who report the heaviest usage. Across the broader population, average global social media use alone sits at two hours and twenty-one minutes per day, on top of a total average internet usage of six hours and thirty-eight minutes. Add television, podcasts, and streaming audio, and the total climbs considerably. All of that content, across every platform, is increasingly sorted, generated, or personalized by AI systems.
AI is no longer a technological presence hiding behind screens. From Spotify playlists to language model-generated replies and personalized news feeds, algorithms now co-create the way we know ourselves and belong in the world.
The cumulative effect of that has no real precedent. Previous generations worried about television rotting the brain. The actual concern now is more precise: not that media is making us passive, but that personalized media is making us increasingly specific, increasingly sorted into narrower versions of ourselves. The algorithm doesn’t want you to grow. Growth requires exposure to things that don’t confirm your existing preferences. Algorithms, by design, minimize that.
A 2025 study examining how AI-based personalization shapes consumer identity found that identity-related consumption was negatively associated with acceptance of AI personalization, reflecting consumer concerns about autonomy and self-expression. In plain terms: people sense, at some level, that something is being done to their sense of self. They push back, at least a little. But the systems pushing don’t stop.
What It Means That We Can’t See It
The feature of personalized AI that makes it most consequential for identity isn’t that it’s powerful. It’s that it’s invisible. The University of Zurich’s Reddit experiment produced AI arguments that users could not distinguish from human ones. The language patterns bleeding from AI tools into everyday speech aren’t ones people notice themselves adopting. The identity feedback loops running through recommendation engines operate below conscious awareness.
The concept of the “algorithmic self” describes how AI influences human identity, introspection, and personal agency through continuous algorithmic feedback. This can lead to a diminished sense of personal insight and growing reliance on algorithmic interpretations of who we are.
As algorithms inform our tastes, read between our lines, and predict what we might do next, they redefine what it means to be self-aware. The algorithmic self can either heighten self-awareness or dismantle introspection and agency, depending on whether people are paying attention to the process.
That last clause is the hinge. Whether people are paying attention. Most aren’t, not because they’re incurious, but because nothing in the design of these systems prompts them to be. The interface is built for absorption, not reflection. The content arrives without friction. The persuasion is personalized. The vocabulary seeps in. And the self being assembled through all of it belongs, at least in part, to the systems doing the assembling.
Read More: Why the Terrifying AI Apocalypse Is Turning Into a New Religion
The Part That Isn’t Going Away
None of this means the tools are straightforwardly bad, or that using them is something to feel ashamed about. The same personalization that can lock someone into a narrowing echo chamber can also surface exactly the documentary they needed to watch, the song that fits how they feel at 11pm on a Thursday, the information that actually helps them make a better decision. The question isn’t whether to use these systems. At this point, opting out entirely isn’t realistic. The question is whether the shaping happens to you or, at least partly, by you.
Algorithmic literacy, the researchers’ phrase for it, doesn’t require a computer science degree. It’s closer to a habit of occasional interruption. Noticing when you’ve been scrolling the same type of content for an hour and deliberately seeking something that doesn’t fit your feed. Asking whether a belief that firmed up recently came from genuine reflection or from repeated exposure to the same argument, calibrated to your profile. Paying attention, periodically, to whether the things you believe and the way you talk about them have started to sound like something a machine selected for you.
Some of that shaping started before anyone was fully aware it was happening. Naming it doesn’t undo what’s already occurred. But it’s usually where any real reckoning begins.
AI Disclaimer: This article was created with the assistance of AI tools and reviewed by a human editor.