Is AI changing how humans write, communicate, and express online?

As AI-generated writing floods the web, researchers say humans may now be subconsciously adapting to machine-like language patterns online

The internet has always changed the way people communicate. Social media shortened attention spans, texting killed formal punctuation, memes created entirely new forms of language, and platforms like Twitter (now X) trained users to compress thoughts into sharp one-liners. But over the past two years, another shift has quietly started taking shape online. Typing like a human is starting to feel strangely difficult. The internet is beginning to sound too polished.

 

LinkedIn posts increasingly read like motivational corporate templates. Emails feel unusually refined. Instagram captions sound overly structured. Even casual conversations online now carry oddly formal phrasing, excessive formatting, careful neutrality, and the infamous overuse of em dashes. Certain words like “delve,” “meticulous,” or “comprehend” have become so associated with ChatGPT that users now jokingly accuse each other of sounding like AI just for using them.

 

 

At first glance, this may simply look like the internet being flooded with AI-generated content. But researchers increasingly believe something deeper may also be happening: humans themselves are beginning to adapt to AI-generated language patterns. In other words, machines are not just learning how humans communicate anymore. Humans may also be learning how machines communicate.

AI writing has quietly become unavoidable online

Part of the reason this shift feels so noticeable is because AI-generated writing is now deeply embedded into everyday digital life. Even people who have never opened ChatGPT directly interact with AI-generated language every single day.

 

Gmail drafts replies automatically. Workplace software summarises meetings and writes emails. Social media users generate captions with AI. Students use chatbots to structure assignments, while companies increasingly rely on AI tools for customer support, outreach, and internal communication.

 

According to a 2025 Brookings survey cited by a US-based news platform Axios, 32 per cent of small businesses already use AI for customer service and outreach, while 16 per cent of individuals use large language models for communication or social media tasks. The scale of exposure matters because language patterns naturally spread through repetition and imitation.

 

Researchers are already observing measurable signs of that spread. A University of Southern California study found that writing style diversity across scientific journals, local news articles, and social media dropped sharply after ChatGPT’s release. Researchers at the Max Planck Institute for Human Development also found that words strongly associated with ChatGPT, including “delve,” “boast,” “meticulous,” and “comprehend,” began appearing more frequently in spoken language after reviewing more than 740,000 hours of content.

 

USC professor Morteza Dehghani told Axios that people gradually get used to an idealised and predictable form of language, eventually beginning to write more like large language models because that style increasingly feels influential or authoritative online.

 

That may be why so much online communication suddenly feels like it is being filtered through the same tone.

Internet is slowly rewarding AI-like communication

Large language models are designed to optimise for clarity, structure, politeness, readability, and safety. The result is text that is grammatically polished but emotionally flattened. Sentences transition smoothly. Arguments are neatly organised. Tone remains measured and diplomatic. Yet after reading enough AI-generated content, many users instinctively recognise the same patterns repeating over and over again.

 

GPTZero, in an article discussing how to write like a human, argued that AI-generated writing often feels robotic because of repetitive sentence structures, monotonous tone, predictable rhythm, and formulaic phrasing. The issue is not necessarily that AI writing is bad. In fact, it is often technically competent. The problem is that it increasingly sounds interchangeable.

 

Alex Mahadevan, chief AI instructor at the US-based school of journalism and research centre Poynter Institute for Media Studies, described AI writing to Axios as soulless and mediocre, even if grammatically correct. He said that there’s no art in it.

 

But the strange part is that the internet increasingly rewards this kind of communication. Polished writing often appears more professional, more competent, and more algorithm-friendly. Hyper-professional language, once mostly confined to LinkedIn posts or corporate emails, is now leaking into everyday conversations online.

 

Emily Bender, a linguist at the University of Washington, described the phenomenon as the rise of the “LinkedIn average,” referring to the bland corporate tone increasingly visible across online communication. Over time, constantly consuming this style of writing may slowly reshape how people communicate themselves.

Humans may be subconsciously adapting to AI tone

Humans naturally imitate the communication styles they are constantly exposed to. Linguists often refer to this as the exposure effect. Over time, people begin mirroring sentence rhythms, vocabulary choices, formatting styles, and emotional tone without consciously realising it.

 

As AI-generated writing becomes embedded into workplace software, online content, and everyday communication tools, those patterns slowly become part of the digital environment people inhabit daily. That means even users who are not directly copying AI output may still begin writing more like AI. That may explain why many users now feel oddly self-conscious while typing online.

 

The irony is that this shift has become so recognisable that many users are now actively trying to avoid sounding like ChatGPT altogether. Certain words, sentence structures, and punctuation styles increasingly trigger suspicion. Em dashes, in particular, have become a running internet joke because of their frequent appearance in AI-generated writing.

 

Mahadevan told Axios that he now second-guesses his own writing, worrying people might assume it was generated by AI. Others have gone even further. A Medium post by Wesley Van Peer criticised the growing trend of users intentionally inserting spelling mistakes or awkward phrasing into their writing to appear more human online.

 

That behaviour reveals something important about the current state of internet culture: people are no longer just communicating online, they are increasingly performing humanity.

Why human imperfections are starting to feel more authentic

Typos, lowercase texting, abrupt phrasing, internet slang, and emotionally messy writing now often feel more human than perfectly structured communication. Casual Reddit comments seem more trustworthy than polished AI-generated responses precisely because they contain unpredictability, emotional roughness, and personal specificity.

 

Mahadevan described missing good-bad writing, the kind of flawed, uneven, but deeply human expression that AI struggles to replicate. GPTZero similarly argued that strong human writing often contains unpredictability and variation because real thinking has bumps, pauses, and uneven moments.

 

That idea cuts to the heart of the broader anxiety surrounding AI-generated communication. Writing is not simply a method of delivering information. For many people, writing itself is part of thinking.

 

As per Axios, Emily Bender warned that constantly outsourcing expression to AI may carry broader societal consequences because people learn how to articulate ideas through the struggle of writing itself. She added that each time people choose not to do that, they are losing out, both individually and societally.

 

The anxiety has become strong enough that some users now run their own writing through AI detectors before publishing it online, worried that sounding too polished could damage credibility. In some corners of the internet, people are even deliberately introducing imperfections into their writing simply to appear authentic. Ironically, sounding human online is no longer happening naturally. Many people are now consciously trying to simulate it.

Workplace is already changing the way people communicate

The shift becomes even clearer in workplaces increasingly shaped by AI-generated communication. A BBC report documenting the rise of AI humanising jobs described how writers increasingly spend their time editing AI-generated content to make it sound more natural, emotional, and believable.

 

One copywriter interviewed by the BBC described repeatedly removing awkwardly enthusiastic phrases, overly formal transitions, and formulaic wording from AI-generated articles. Another worker who spent months cleaning up chatbot-generated content said the repetitive editing process eventually made him feel like the robot. The contradiction is difficult to ignore. Companies increasingly rely on AI to generate content quickly, only to then hire humans to inject humanity back into the same content afterward.

 

At the same time, workers increasingly feel pressure to communicate in polished, AI-friendly ways themselves because that style appears more professional and efficient. Typing casually or imperfectly online can now sometimes feel less acceptable than sounding polished and algorithmically clean.

The internet’s growing identity crisis

This has created a bizarre online environment where people simultaneously distrust AI-generated writing while also adapting to its norms. The result is an internet going through a subtle identity crisis.

 

Authenticity is becoming harder to detect. Human personality risks being flattened into algorithmically safe communication patterns. Spontaneity increasingly competes with optimisation. At the same time, users are developing new ways to signal humanness through imperfection, emotional irregularity, and unpredictability.

 

Importantly, this does not necessarily mean AI is destroying writing or eliminating creativity altogether. Many writers, artists, students, and professionals genuinely find AI tools useful for brainstorming, translation, research, or overcoming creative blocks. The issue is less about AI-generated text existing online and more about how constant exposure to machine-generated language may gradually reshape what people perceive as normal communication.

 

The internet once taught people to type faster, shorter, and more casually. AI may now be teaching people to sound cleaner, safer, and more polished.

 

The strange irony is that after years of training machines to communicate more like humans, humans themselves are increasingly editing their language to avoid sounding like machines.