One of the biggest criticisms of writing with AI is that it “sounds” like AI. People want their writing to sound human—like them. They get irritated or feel duped when they read something that sounds like it was created with AI.
A lot of this anxiety centers on voice. It’s one of the major factors we use to distinguish human writing from AI writing. That’s why there’s so much talk about how to preserve voice when using AI.
But what is “voice,” exactly? What do we want that AI isn’t already delivering? After all, AI was trained on an enormous amount of human writing. It knows more about what humans sound like than any individual person could. So why do we read AI-generated writing and think, “Ugh, no”?
AI is sounding more and more like humans all the time. Writers often argue that AI will never replace them because it will never be able to bring the unique perspective that a person can. But what happens when readers can no longer tell the difference?
What do we mean by “voice”?
Voice is the combination of tone, rhythm, phrasing, and perspective that gives writing a distinctive feel.
Frankly, it’s a bit of a squishy concept. One of those things that you know when you see it (or read it). But when you try to pin it down, you quickly end up in a sea of adjectives like conversational, authentic, warm, authoritative, witty, quirky, or confident.
For now, there’s a gap between what we consider human voice and the “voice” that AI either has or imitates. Differences people have highlighted include:
| Human voice | AI voice |
| Sounds like a particular person | Sounds like a capable version of many people |
| Has preferences and biases shaped by lived experience | Simulates preferences based on patterns |
| Has a natural rhythm shaped by speech patterns | Often has a polished, manufactured cadence |
| Uses detail because it mattered to the writer | Uses detail because it improves plausibility |
| Can sound idiosyncratic, inconsistent, and awkward | Often sounds optimized and generalized |
| Shows clear opinions and preferences | Defaults to smoothness and neutrality |
| Notices some things and ignores others | Provides exhaustive coverage |
| Feels like it comes from somewhere | Feels like it was assembled for a purpose |
| Is expressive | Performs expression |
AI is good at mimicking certain style markers, like sentence structure, level of formality, and rhetorical patterns. But it still struggles with the deeper patterns that make a voice feel individual, like what people notice, what they care about, what they’ve lived through, and the things they can’t stop thinking about.
Two ways of thinking about voice
When talking about voice, lot depends on whether we define it based on:
- the reader’s experience of the writing
or - the source of the writing
Let’s look more closely at both.
1. Voice comes from reader experience
If voice is what the reader perceives in a piece of writing, AI can certainly have a voice. If you read text written by AI, it sounds like something. And we can call that something voice.
Still, most people treat human voice as inherently better, with AI being a pale imitation.
But that may not be true in all cases. For most everyday reading, people don’t actually want a deep, unique human voice. They just want the text to be:
- clear
- easy to follow
- familiar
- straightforward
- understandable
In many contexts, readers care less about originality or individuality than writers assume. That can feel threatening, especially to those of us who put high value on authenticity and human perspective—and who make a living through their writing.
2. Voice comes from the source
What if voice depends less on the writing itself and more on where the writing came from? We could argue that AI can never truly have a voice, because it isn’t speaking from experience or identity but generating patterns from existing language.
However, very soon AI will get so good at writing in a “human” voice that we won’t be able to tell the difference.
“I’ll always be able to tell the difference,” you may think.
Research suggests otherwise. For example, a 2024 University of Pittsburgh study found that people often couldn’t distinguish AI-generated poems from poems by well-known poets like Emily Dickinson and Shakespeare. And many actually preferred the AI-written poems—until they found out they were created by AI.
The authors concluded:
“Non-expert poetry readers expect to like human-authored poems more than they like AI-generated poems. But in fact, they find the AI-generated poems easier to interpret; they can more easily understand images, themes, and emotions in the AI-generated poetry than they can in the more complex poetry of human poets. They therefore prefer these poems, and misinterpret their own preference as evidence of human authorship.”
This study led me to an unnerving question:
- Do we actually prefer human voice?
- Or do we prefer the idea of human voice?
If we’re biased toward human voice, even if we objectively can’t tell the difference, that’s not about the writing itself. At least part of that reaction is about fear. Fear of being erased. If we’re not bringing anything unique to the table, why would anyone need us?
Fundamentally, we’re not having a conversation about voice. We’re having a conversation about what it means to be human.
Beyond voice
If we don’t move beyond a defensive, knee-jerk reaction about voice in AI writing, we won’t be able to explain why human voice matters in the first place.
No one has all the answers. But it’s essential to think about the questions. Here are a few more to chew on:
- If readers can’t consistently tell the difference between AI and human voice, does the distinction still matter?
- If you only value “human voice” once you know a human wrote it, what exactly are you responding to?
- Does it matter who wrote something if the writing itself works?
- How much of what we call “voice” is really just familiarity with certain styles and patterns?
- Will “human-written” eventually function more like a luxury label (think “handmade”)?
- If audiences can’t reliably distinguish between AI and human voice, what exactly are clients paying for?
- How much of the defense of human writing is economic self-preservation?
- If the value of human voice depends on readers knowing a person wrote the text, where does the value actually reside?
I could see “human writing” eventually becoming less a category of quality and more a category of provenance. Like handmade pottery, live music, and handwritten notes. People value these not because the result is objectively superior, but because they know there was a person on the other end.
We care about effort, connection, and authenticity—even when we can’t reliably detect them. Writing may no longer reveal authorship, but people still care deeply about authorship anyway.