My colleague Sam posted recently about LinkedIn shipping a “seems like AI slop” button. His take: it’s pointed at the wrong target, because slop predates the tools. “AI didn’t create the emptiness; it industrialized it.” I commented, offered a yes-and, and promised to stop before going full Foucault in his comment section with my own artisanal, handmade slop. This post is me breaking that promise in a more appropriate venue.
I keep watching the same scene play out at work. You might recognize it. You draft something with an AI assistant, read the output, wince, and start the laundering pass. Kill the em dashes. Ban “delve” and “elevate.” Shorten the sentences. Add a contraction. There are word lists for this now, and browser tools, and prompts that end with “write like a tired human.” (A tired human?!) An entire cottage industry has formed around one question: how do I make my AI-generated writing sound human?
I think it’s the wrong question. Not because the problem isn’t real. The problem is VERY real. But the question misdiagnoses the root cause. And the misdiagnosis is why the solutions — the laundering — don’t work.
The instinct is right.
Let’s start with what the question gets right. When you read AI-sounding text and something feels off, you’re detecting something that actually IS off. It’s human words written in human sentences, and it’s miserable to read. Readers bounce off this stuff. Your clients bounce off it. You bounce off it, which is why you winced. That wince is a data point!
The mistake is in what we decide the wince is about.
The fixes don’t fix the AI smell.
Here’s a test you can run yourself. Take a paragraph that sounds like AI wrote it and apply every fix on the list. Swap the vocabulary, vary the rhythm, break up the walls of text, purge the punctuation of the month. “We are excited to announce enhancements to our documentation process” becomes “We’re improving our docs.” Shorter. Plainer. It includes a contraction. Human!
Still dead. 😵
It’s still a sentence to nobody, about nothing anyone can act on. It will pass any AI detector you point at it, human or automated, and it will still connect with no one. That’s because the tells were never the problem. They were symptoms. The disease is that the text was written to no one, no clear audience.
What readers are actually smelling
Writing doesn’t stumble onto an audience. When a piece of writing connects with you, when it feels like it’s reaching toward you specifically, that’s because the writer crafted it to reach people like you and to move you somewhere: toward a decision, an action, a changed mind, a feeling.
That reaching quality is what you’re naming when you say a text “sounds human.” You don’t have an AI detector in your head. What you have is a deep desire to connect and feel seen and a distaste for mass-produced “slop.” Slop is older than the large language models, too. “Word vomit” was a flag for slop. So was every “tl;dr” ever typed. LinkedIn shipped a button for a reflex you’ve had forever. Text written to a specific audience satisfies your delight in connecting with another human, even when a model helped produce it. Text written to no one in particular sets off your slop detector, even when a person typed or spoke every word.
This is also why genuinely good writing gets away with breaking every rule on the style lists underpinning “anti-AI” tools. Fragments. Strange syntax. Repetition, on purpose. (See: Richard Brautigan! Emily Dickinson! Gertrude Stein!) Nobody runs writing they love through a checklist, because the question, “Does this sound human?” only comes up when the answer to, “Is this thing talking to me?” is already no.
Here’s the part that’s hard to say.
If AI exposed this gap in your writing, the gap was there before AI. I’ve missed the mark plenty of times with and without AI in the mix.
Most of us were never taught to write to anyone. We were taught to write about things. School essays addressed to no one in particular, graded on conformity to genre conventions – 5 paragraphs, thesis statement, stick the landing in the conclusion. Status updates addressed to everyone on the CYA mailing list. Documentation created mostly to mitigate a liability risk. I’m not diagnosing from a clean record of offenses: I have sent the 45-minute meeting recording instead of the three sentences somebody could act on. AI didn’t invent writing without an audience. It industrialized it. Sam picked the right word, alright.
This is not about being a bad writer. It’s about being an inexperienced one, and inexperienced writers are preoccupied with syntax and word choice at the cost of connecting with their audience. That’s exactly why, “How do I make it sound human?” is the question you reach for. It’s a syntax and word choice question. Experience in writing is mostly accumulated audience judgment: years of watching what lands and what dies with your readers, until “who is this for” becomes a reflex instead of a step. Nobody is born with that judgment. Everyone who has it built it through practice. This means you can build it too, starting with your next draft.
The question to ask instead
My recommendation is to retire “How do I make this sound human?” It leads to laundering, and laundering produces text that’s harder to connect with and exactly as easy for folks to ignore.
The replacement question: Who is this for, and what do I want to happen when they read it?
Answer it before you draft, or use it to re-aim a draft you already have, and most of the style problems resolve themselves, because now every sentence has a job and a judge. Skip the audience aiming, and no amount of humanizing will help, because there’s no human on the other end of the writing to reach.
One last thing. You may have noticed this post has been talking to you the whole time. That wasn’t an accident. That’s the entire trick.
Lunch is scheduled, Sam. 🌭