This Is The Reason Why You Sound More Like AI

I was replying to an email two days ago—nothing important, just a work thing—and I paused mid-draft. It wasn’t that I didn’t know what to say; I’d just caught myself doing the thing: Add one sentence of context. Start a new paragraph. Follow with bullet points.

That’s the pattern I use when I’m working with AI. Start with a short lead, then break it out. It’s clean, scannable, and structured. And there I was, doing it in a regular email to a real person, completely on autopilot. I sent it anyway because it wasn’t wrong, and it was still me writing.

However, I couldn’t unnotice it after that. I started catching it in my Teams replies. In the way I’d naturally reach for a numbered list instead of just talking through something.

I’d turned into this oddly instructive, descriptive communicator, where I used to narrate.

In fact, we’re all starting to sound like the AI we’re trying to humanize.

In this post, I won’t give you the final answer on AI and authenticity. I don’t have that, and honestly, I’d be suspicious of anyone who says they do 🤨

What I want to do is name what’s happening by looking at the research, the incentives, and where this is going, and think through some approaches that might actually help you hold on to your voice in a world that’s increasingly optimized against it.

Can You Even Write Anything Properly Without AI?

What unsettled me more than the email was the realization I’d been handing AI my voice for long enough that I wasn’t totally sure anymore which parts of my style were actually mine, and which parts I’d just absorbed from working inside that loop every day.

I’ve given AI instructions on how I write—my style, tone, structure—but those were formed before AI. Through school, through writing a lot, through figuring it out the long way.

Now those instructions are frozen in time while AI uses them to produce more content, and I’m not sure my voice is actually growing from that. It’s just being replicated.

We spend a lot of energy asking AI to sound more human when the reverse current is already running. It’s quieter, and in some ways more interesting.

If you’ve had that low-key background anxiety about whether you can even write anything “properly” without AI anymore, you’re not being dramatic.

There’s something real there.

Why Your Brain Treats AI Like a Social Partner (And What That Costs You)

Humans are wired to be social. Like, evolutionarily, can’t-turn-it-off social. Our brains are constantly scanning for social cues: Is this entity responding to me? Does it understand me? Is it safe?

Since language is one of the most powerful social cues we have, when something speaks to us fluently and responsively, our brain tends to treat it as a social partner—whether we consciously register that or not.

A YouGov poll found that 46% of Americans believe people should be polite to AI chatbots by saying “please” and “thank you.” Not because the chatbot needs the courtesy, but because the social wiring kicks in automatically.

We’re not really choosing to be polite. We’re just being social creatures, and social creatures say please.

Tip: The real issue isn’t politeness. It’s that adding all those “please” and “thank yous” in your prompt is adding more noise that the AI model has to parse through in order to produce a succinct response. Be direct and save yourself some tokens and time by skipping the niceties.

That gets even more pronounced when you introduce voice mode. When you switch from typing a mechanical prompt to speaking to an AI with all the rhythm, inflection, and pause of actual conversation, the effect intensifies.

Evolutionary research suggests our brains are primed to interpret human-like appearance or behavior as having social motivations. Voice and rhythm trigger that interpretation in ways that typed text alone doesn’t quite manage. The interface feels like a dialogue, so your brain treats it like one.

Once you’re in social mode, you adjust accordingly. You soften. You become more conversational. You start fitting yourself to what the exchange seems to call for.

The more human the interaction feels, the more human the communication patterns you bring to it.

The problem is that those patterns don’t clock out when the conversation ends. They stick.

How Accepting AI Suggestions Rewires Your Sense of Your Own Voice

There’s a second layer here that’s more subtle than the social wiring thing, and I think it’s the more insidious one. It’s about what happens cognitively the moment you accept an AI suggestion as your own output.

When you prompt AI to edit or polish your writing, you’re making a series of small decisions:

keep this word
cut that phrase
accept this sentence

Those acceptance moments don’t feel like consuming external content. They feel like choosing. And choice tends to get encoded in memory differently than passive consumption.

In cognitive psychology, this is known as implicit learning, referring to the recurring phrasing and word choices that get unconsciously stored in memory.

You don’t notice you’ve filed away “delve” or “it’s worth noting” as vocabulary that feels right. You just notice, weeks later, that you’re using them. The content bypassed the part of your brain that would have flagged it as external and settled in alongside things you actually think and believe.

Accepting a suggestion isn’t the same as reading someone else’s writing and moving on. It’s closer to authoring it. Your brain logs it as a decision you made, which means it gets reinforced the next time you reach for that word or structure.

What gets lost in that process is what linguists describe as the meanders, interruptions, and leaps of logic that signal genuine human thinking. AI-generated language optimizes away from those. It produces something smoother, more organized, more predictable. It reads cleaner, but also not quite you.

The more you accept those suggestions, the more your internal model of what “good writing” sounds like quietly recalibrates toward the AI’s version, without you signing off on the trade.

The Prompting Feedback Loop That’s Quietly Changing How You Communicate

Prompting is its own skill, and most people who use AI regularly have gotten decent at it. You get specific, use structured language, and over-explain your intent.

You develop what amounts to a second dialect—prompting language—that’s precise, neutral, and stripped of the ambiguity that regular human communication relies on.

The part that doesn’t get a spotlight much is that prompting language starts bleeding into how you communicate elsewhere.

FSU researchers studying 22 million words from unscripted podcasts coined the term “lexical seepage” as the process by which AI-inflected language, unlike slang spread by subcultures, originates with an algorithm and seeps into human communication without the speaker realizing it’s happening.

It’s not about copying AI output. It’s about exposure creating residue. You read enough AI-polished prose, edit enough AI-generated drafts, and interact enough through AI-mediated interfaces that the patterns start feeling natural.

The unsettling part? It happens while you’re actively trying to make the AI sound less like AI.

The loop closes when AI gets trained on content it helped create. It learns from our AI-inflected writing. We keep polishing our output with AI. The gap between human and AI language narrows not because AI got more human, but because humans drifted toward the machine.

Once that gap closes, there’s no clean way to tell where the drift started.

Research Confirms It: AI Is Changing the Words Humans Actually Use

Researchers at Germany’s Max Planck Institute decided to measure whether this was happening at scale. They first fed ChatGPT millions of pages of emails, essays, and articles using prompts like “polish this text” or “improve its clarity.”

They identified words ChatGPT consistently added during these editing sessions, like “delve,” “realm,” “meticulous,” and “boast.” They coined the term “GPT words” for them to act as a linguistic fingerprint of AI influence.

Then they turned to real human speech. They analyzed over 360,000 YouTube videos and 771,000 podcast episodes, tracking GPT word usage before and after ChatGPT’s release.

In the 18 months following ChatGPT’s launch, words like “meticulous,” “delve,” “realm,” and “adept” increased in human speech by between 35% and 51%.

The synonyms for those words? They barely moved.

It wasn’t a general vocabulary trend but specifically the words AI favors, showing up more in human mouths and keyboards.

And “delve” was appearing in videos that seemed to be unscripted, meaning people hadn’t just absorbed these patterns into their writing but also absorbed them into their speech.

Note: That’s not people copying AI output. That’s people who’ve internalized AI language patterns as their own without any awareness that the migration happened.

Linguists describe this as the first time in history that machines are directly dictating human communication style.

Not through cultural trends or media influence, but through direct, repeated interaction between people and a language system that consistently favors certain patterns 😬

AI Language Flattening: Why It Hits Some Voices Harder Than Others

This drift doesn’t stay at the individual level. Scale it across millions of people accepting AI suggestions daily, and something larger starts happening to language itself.

Regional slang. Inside jokes between communities. The casual irreverence of certain cultural voices. The particular rhythm of how someone from one part of the world structures a sentence differently from someone across the globe.

All of that is under quiet, consistent pressure.

Researchers at USC found that AI systems often align with what they call “WHELM” perspectives (Western, high-income, educated, liberal, and male) because those are the communication styles most common in the English-language data these models were trained on.

It’s not that the AI is consciously promoting that worldview. It’s that the training data skews hard in that direction, and the model reflects what it learned.

A Cornell University study found that AI writing suggestions caused Indian English speakers’ writing to converge significantly more toward American writing styles than the reverse.

Note: The homogenization has a direction. It’s not everyone meeting somewhere in the middle; it’s everyone drifting toward one particular center of gravity.

That matters a lot for anyone whose natural voice doesn’t fit that mold.

If you grew up writing and speaking in a way that reflects a different cultural context, using AI to “polish” your communication means repeatedly accepting suggestions that pull your expression toward WHELM defaults.

Each acceptance is small. The cumulative effect is a quiet erosion, optional on the surface, but consistent underneath.

The irony is that you’re probably using AI to communicate better. And in many ways, you are.

But the version of “better” being reinforced might not actually be yours. That’s the flattening effect, and it doesn’t just flatten language. It flattens identity.

How Platform Algorithms Compound the Pressure on Your Writing Voice

Platforms don’t show your content to everyone who follows you. They show it based on how well your content fits what the algorithm has decided is “engaging.”

That means your writing, along with its structure, format, and hooks, is quietly shaped by what gets rewarded, not what feels natural to you.

If you’ve spent any time trying to grow on LinkedIn or Instagram, you’ve felt this. The posts that perform tend to follow recognizable patterns:

punchy opening line
short paragraphs
a hook that creates just enough curiosity or friction
some kind of invitation at the end

Not because that’s how people naturally communicate, but because that’s what the engagement mechanics favor.

The result is algorithmic complicity, where you compromise your natural voice because visibility is the implicit incentive, not because someone told you to.

You become a slightly better-performing version of yourself, optimized for the platform’s definition of engaging. Whether or not that version actually sounds like you becomes a secondary concern.

I noticed this in my own LinkedIn drafts. There was a period where I was doing structured research on the most engaging formats by prompting AI to analyze what performed well, then building posts around those patterns.

The posts were cleaner. They did better. But looking back at some of them now, they also feel a bit, dare I say, generic? Like they could’ve been written by anyone talking about the topic. That specific me-ness was edited out in favor of something the algorithm was more likely to amplify.

Note: The longer you operate in those environments, the more performing-for-the-algorithm starts feeling like how you write.

SEO Is Old News, AEO Is the New Pressure on Human Writing

For years, writers and content creators learned SEO by structuring content around keywords and search intent signals so Google would rank it favorably. That already created pressure toward certain patterns:

lots of headers
clear direct sentences
lists over narrative

Now there’s AEO (Answer Engine Optimization), and it pushes that pressure further 🙈

Where SEO was about ranking a page for human clicks, AEO is about becoming the source that AI-powered tools like ChatGPT, Perplexity, and Google AI Overviews cite when someone asks a question.

That means structuring content for machine parsing, writing in language that mirrors how queries get phrased, and presenting with clear authority signals.

The key distinction:

SEO gets a search engine to rank your page
AEO gets an AI to extract and cite your answer directly

The audience is increasingly a machine before it’s a human, and the writing that a machine finds trustworthy tends to be structured, hedged, and stripped of personality by design.

More: Who Are You Actually Writing For in the AI Era?

At the individual level, this shows up as what I think of as the “human disclaimer” default. It’s the creeping tendency to over-explain, preface everything with qualifications, and hedge every claim.

For example:

“It’s important to note that…”
“This may vary depending on context…”
“While there are multiple perspectives…”

These are AI-adjacent patterns since they represent the language of a system trained to avoid overconfidence.

When humans start adopting them, they stop sounding like people thinking out loud and start sounding like cautious institutional voices.

When AI Speech Patterns Migrate Into Unscripted Conversation

The Max Planck data wasn’t from emails or polished LinkedIn posts. It was from YouTube videos and podcasts, a lot of them unscripted. Over 360,000 videos and 771,000 episodes. And the surge in GPT words showed up there, too.

Now, pause and digest that. Because “delve” appearing in unscripted videos means people aren’t just using AI-inflected language when they write. They’ve absorbed it into how they talk.

It moved from the text editor into the voice, into the moment before you even know what words you’re about to say.

Watch enough content from creators in the same space, and you’ll notice a sameness in cadence and vocabulary that didn’t used to be there.

The crisp, unpredictable sentence. The genuine hesitation. Or the off-beat observation that goes nowhere but feels distinctly human. These are getting rarer.

They’re getting rarer in content that was never scripted, meaning the seepage isn’t something you can catch before you hit publish. It’s already in how you think when you open your mouth.

What gets lost isn’t just vocabulary variety but the signal that tells a listener a real person had to actually think to produce this.

How to Protect Your Authentic Voice Without Ditching AI

The feedback loop is real, the platform incentives are structural, and “just be authentic” doesn’t do much against something that works through implicit learning.

But there are approaches worth trying, and I’ve been working through some of them myself.

1. Notice the Drift Before You Try to Fix It

Read something you wrote a month ago and ask not whether it’s well-written, but whether it sounds like you.

Do you recognize your specific way of thinking in it?

Or does it read like something anyone using similar prompts could have produced?

That gap, when you feel it, is useful. It tells you something real about where your baseline has moved.

2. Write Your First Draft Before Bringing AI In

When I draft posts on my own first, even messily, just getting my actual thoughts on the page, and then use AI to refine in a second pass, the result feels more like mine.

When I let AI start from a blank page, I’m inheriting its patterns before I’ve even established my own.

Tip 🎯
There’s a real difference between using AI to refine your expression and using AI to generate it. The first pass is where your voice either gets set or gets skipped.

3. Leave the Imperfections That Make It You

Jazz musicians talk about the expressive value of the bent note, a deliberately off-pitch moment that makes music feel alive.

Writing has an equivalent: the weird specific observation, the unpredicted tangent, the honest “I don’t actually know” that breaks the smooth surface.

These feel like weaknesses if you’re optimizing for clarity. They’re actually the things that make your voice yours.

The roughness is the signal. Don’t edit it all away.

4. Question What “Professional” Has Come to Mean

A lot of AI-inflected language has taken over the idea of professional communication— “meticulous,” “leveraging,” “adept at navigating”—and it sounds authoritative because there’s nothing personal or risky in any of it.

Communication that anyone could have written is less memorable and less convincing than communication from a recognizable person with a specific perspective.

Your particular voice with its quirks, rhythms, and slightly odd word choices isn’t the thing you need to polish away. It’s the credibility signal.

My Own Honest Experience With This (Since We’re Here)

I’ve been intentionally writing more outside of AI lately through journaling, drafts I don’t finish, and things that aren’t going anywhere. Not because I’m anti-AI, but because I started to realize I was losing confidence in my ability to start something from nothing.

I used to write short stories. Somewhere along the way, I started needing a scaffold (an AI-generated first draft to react to) before I could get going.

The blank page started feeling harder than it used to. Not impossible, just harder in a way it hadn’t been before.

Part of what I’ve realized is that my writing instructions to AI, things like my style, voice, and other instructions about how I communicate, were all formed before AI became part of my daily workflow.

I developed them through school, through years of writing, reading, and finding what felt like me.

Now, those instructions are kind of frozen there, while AI uses them to produce more content. My voice is being replicated but not evolving. That feels like a specific kind of loss, one I’m not sure how to name cleanly, but one I can feel.

The journaling helps. Writing something with no SEO implications, no audience, no engagement metric, just the honest rambling version of how I actually think, acts like a reset.

It’s not a solution, but it’s a way of keeping the door open. Of reminding myself that I still have a way of thinking that exists separately from the loop.

If You’re Talking Like AI, Are You Starting to Think Like It Too?

If you’re talking like AI by reaching for its patterns, its vocabulary, its hedged and structured way of making a point, are you also starting to think like it?

That’s the question I keep coming back to, and I don’t have a clean answer for it.

Language shapes thought. The words available to you change what you can express, and what you can express changes what you can conceptualize.

If the vocabulary you reach for gets increasingly optimized for machine-clarity over human-complexity, something happens to the messy, uncertain, associative thinking that produces original ideas.

It gets harder to access. Not gone, but quieter.

At a cultural level, linguists have raised real concerns about AI homogenizing regional and international variations of English, making it less likely that distinctive voices and expressions survive.

The individual version of that loss is just as significant. It encompasses the gradual smoothing of your particular way of seeing things into something more universally legible, and therefore less distinctly you.

Why AI-Polished Communication Quietly Erodes Trust

We pick up on signals of effort in communication that come through as the sentence that took work, the honesty that cost something, the specific detail that only comes from actually being there.

AI-polished prose is missing those signals.

When your human communication starts mimicking AI-polished prose, it starts missing them, too.

People notice, even if they can’t articulate what feels off.

The tools are useful, the uses are real, and I use them every day. But it’s important to ask yourself: Why am I saying this? Is this my voice, or a pattern I absorbed?

You don’t need to answer it every time.

Just don’t stop asking it.

It’s a Wrap

The irony hiding inside the “humanizing” prompt, that reflex to ask AI to sound more like a person, is that the very act of training AI to speak human is quietly helping train humans to speak AI.

That loop is already running. The vocabulary and speech patterns are moving.

The platform incentives, the SEO-to-AEO shift, the accumulated weight of AI-mediated communication—all of it is pushing in the same direction. And it’ll keep getting louder.

The goal isn’t to resist AI or put a no-AI sign on your writing. It’s to stay awake to what’s happening.

To know when you’re using the tool and when the tool might be shaping you.

Your voice took years to develop. It has texture, quirks, and rhythms that belong specifically to you, formed through real effort and real time.

The goal is not to keep AI out, but to keep you in 💁‍♀️

AEO AI Digital Marketing SEO