The Accusation
Somewhere along the way, clear writing, complete sentences, and a well-placed dash became suspicious. Apparently, I don’t write like a person anymore. I write like AI. Which is interesting because I’ve been writing like this since before AI was a product category. I’ve got receipts… or I’m a time traveler.
Corporate large language models didn’t just show up. They showed up sounding like they stole my schtick.
Back when my biggest influence was a mix of broadcast scripts, discussion boards, and whatever version of Emily Dickinson the girl I was into was reading… or worse, Ayn Rand.
That shift – from “clear writing” to “AI voice” – is worth paying attention to.
We Weren’t Just Online. We Were Writing It
A lot of us didn’t just grow up with the internet. We wrote it. Forums. Blogs. LiveJournal. AIM. Early social media. Then Reddit, where the comments somehow mattered more than the post.
In a lot of ways, large language models feel like early Reddit solutions. Same idea, just fewer arguments and no one telling you to “use the search bar.”
Writing wasn’t occasional. It was constant, public, and immediate. You wrote to be understood, and you wrote knowing someone would respond. That environment does something to your voice. It sharpens it.
You learn pacing without calling it pacing. You learn tone because if you get it wrong, someone tells you immediately. You learn structure because walls of text get ignored.
What comes out of that is writing that feels organized, but still human. Intentional, but not sterile.
Some of Us Were Trained to Sound Like This
Clear writing gets labeled as artificial now. Structured. Direct. A little too clean.
That’s the tell. Apparently. Let me get meta.
[Tone shift: clipped, punchy, intentionally fragmented. This is the style people now associate with AI.]
And here’s the part that matters most, right up front:
That voice didn’t come from AI. It came from training.
Everything else is just explanation. Meta, again.
[Structure callout: this is the inverted pyramid. Point first. Context after.]
You lead with the point. You don’t build toward it. You don’t save it for later. You put it first, because if it doesn’t land immediately, it doesn’t land at all.
You learn to write like this when the message has to land the first time. No second chances. No re-read. Just impact.
Broadcast will do that to you.
If a sentence doesn’t make sense on first listen, it fails. So you adjust. Shorter sentences. Clearer ideas. Less room for ambiguity. Everything has to carry its weight.
Then you layer in academic writing, where structure, organization, and clarity aren’t expectations you negotiate – they’re expectations you meet.
Then professional communication, where tone and audience matter just as much as the message itself.
Stack it all together, and what you get isn’t robotic. It’s intentional. Built for understanding. Built to land. One more time…
[Tone shift: moving back to natural flow. Longer sentences, more connective tissue, less fragmentation.]
Then the pandemic hit. And suddenly, that style wasn’t just useful. It was necessary. Daily updates. Constant pivots. Reassurance, all written, all the time.
Trying to keep people informed. Trying to keep people grounded. And yeah, there were dad jokes. There was at least one Tiger King reference that made perfect sense at the time.
Because clarity wasn’t just about efficiency anymore. It was about stability. About making sure the message actually reached people when they needed it. So now that same voice shows up. Clear. Structured. A little conversational.
And it gets labeled as artificial. Which is a strange kind of full circle. Because nothing about it changed. Only the context did.
The Part Everyone Skips
Back to normal. Large language models didn’t come out of nowhere. They were trained on the internet. And for a long stretch of that internet… we were the ones writing it.
AI didn’t invent this voice. It learned it. Some of that data is literature. Some of it is journalism.
And a surprising amount of it is probably someone explaining why the Oxford comma matters in a forum thread that got way too heated for what it was.
Why It Feels Off Now
Writing norms shifted. There was a time when clarity signaled competence. Now, in some contexts, clarity signals suspicion. Communication got faster. Shorter. More fragmented.
Texts. DMs. Slack messages. Half-finished thoughts sent between meetings. So, when something shows up that’s structured, complete, and deliberate, it stands out. And anything that stands out gets questioned.
Even punctuation has caught strays. The em dash, which has been around long before any of this, now gets flagged like it’s a tell. Which is a strange outcome for a piece of punctuation that predates both the internet and the idea that it might one day be considered suspicious.
Where This Breaks Down in Education
This shift isn’t just cultural. It’s showing up in the classroom. Students are being flagged based on tone. Faculty are relying on instinct to determine what “sounds human.”
And the problem is, those instincts were built in a different writing environment. We’re not just misidentifying AI. We’re starting to misidentify human competence.
At this rate, we’re not far from treating any well-structured document like it was drafted by a large language model. The Constitution might be next.
A Better Question
The question shouldn’t be, “Does this sound like AI?” That’s not a reliable signal anymore. A better set of questions looks like this:
- Can the student explain their thinking?
- Can they apply ideas in context?
- Can they show their process?
Those are harder to fake. More importantly, those are what we actually care about. Style is easy to mimic. Thinking is not.
What We Do About It
This is where course design starts to matter more than detection. A few practical shifts:
- Build in drafts and checkpoints so thinking is visible over time
- Pair submissions with short reflections on process and decisions
- Use applied, context-specific assignments that require interpretation
- Include low-stakes, in-class work where students show how they think in real time
These aren’t about catching students. They’re about seeing them.
Bring It Back Around
If this sounds like AI, it’s worth remembering that AI learned this voice from somewhere. The goal isn’t to write worse so we sound more human. The goal is to get better at recognizing human thinking, especially when it shows up clearly.
At some point, I started to wonder if I’m not just a person who writes clearly. I might be what we’d now call a legacy language model. LLM, if you will.
And here’s the part I can’t quite let go of… When you ask an AI to be a little less efficient, a little less structured, a little more uneven, it suddenly feels more human.
So, if sounding human now means sounding less precise, what exactly are we measuring when we evaluate writing? That’s a conversation worth unpacking on its own.
