How AI Can Help You Write Like a Human
Plus, getting philosophical with AI!
Dear Prompt Experimenters,
I’ve seen much chatter around AI detection tools. How to use them. How to train AI to trick them. How to modify your own writing style (don’t use em dashes) to avoid getting flagged by them. And so on.
In some very specific situations, these tools are relevant. But not for most of us, most of the time.
Who cares if you use em dashes, or too many “AI-sounding words.” What matters is whether or not your writing is good. And to be good, it must feel human (at a much deeper level than punctuation patterns.) But what does that even mean?
Let’s talk about that today.
THIS POST HAS A BACKSTORY: I had a bad idea, and AI convinced me to write about something else
I wanted to connect the philosophy of Charles Taylor to AI, in a way that is accessible and useful.
Fair enough.
So I asked o3: When people criticize the inherent limitations of AI large language models, I often find myself responding “yeah, but humans are really just large language models, too, sitting on top of an ancient reptile brain and a sensory apparatus.”
Could I adapt Charles Taylor’s work in The Language Animal to more rigorously defend this argument, in a way that my readers will find interesting and compelling (and somehow useful in the real world?) Be completely honest with me. If this is a bad idea, don’t tell me it’s a good one!
AI told me it was a horrible idea, in its own polite way.
It’s an intriguing comparison—after all, both humans and today’s LLMs do “spin out” linguistic continuations from prior text. But if you run the Humboldt/Taylor checklist all the way through, the overlap looks more like a surface analogy than a deep identity. Four big fault-lines show why the equation “human ≈ LLM + sensors” is too thin.
We could dig into the Humboldt/Taylor checklist, but let’s not, since that whole idea was a dud. Instead, I asked AI to suggest a related and equally provocative idea to write this post about.
It provided five. This was my favorite:
LLMs are mirrors that exaggerate: The shock of talking to a fluent but un-embodied agent exposes how much of modern discourse is already treated as mere text swapping. The discomfort we feel isn’t just “the model lacks consciousness,” but “I sometimes talk as though I were an LLM.”
Okay, that last sentence is interesting. Let’s unpack that. What does it mean to talk (or write) like a human as opposed to an LLM? And how can we use that insight to shape how we use language, to give our words greater power and resonance and impact?
AI, CAN YOU PLEASE UNPACK THAT LAST IDEA?
Let’s talk about Charles Taylor. Not in a “philosophy for philosophy’s sake” kind of way. Let’s understand, specifically, how Charles Taylor can help us improve our own use of language, by making it more HUMAN.
1. Why Charles Taylor Can Help Us Here
Philosopher Charles Taylor has long argued that language is not just a tool for representing the world but a medium through which we articulate who we are, shape social norms, and bring latent feelings or intuitions into shared awareness. Following the tradition of Wilhelm von Humboldt, Taylor insists that language doesn’t merely describe reality—it discloses it. That is, it helps us see and feel dimensions of experience and meaning that might otherwise go unspoken or unnoticed.
This matters for how we talk and write today. In a world increasingly shaped by large language models (LLMs) and a glut of surface-level communication, Taylor’s insights help us discern what gets flattened or lost when our discourse slips into what I’ll call “text-swapping.” The point is not to resist technological tools, but to better understand the layers of meaning they often bypass—and how we might recover them.
We all experience this everyday, right? Some of the stuff that people say and write is real, and it matters. Other stuff isn’t. And our world seems ever fuller of the latter.
This next bit starts off somewhat abstract, but leads into a real diagnostic which you can apply right now.
2. What Does “Modern Discourse as Text-Swapping” Mean?
Think of any conversation or piece of writing as operating across three intertwined layers:
World-Directed Reference: At this level, language points to objective facts or shared realities. For example: “The meeting is at 3 p.m.” This kind of statement is checkable; it enables coordination. Both Humboldt and Taylor see this as essential but only a starting point.
Norm-Laden Commitment: Here, the act of speaking or writing alters what’s expected of us. Saying, “I’ll chair the meeting,” doesn’t just share information—it changes my obligations. Language becomes a force that reshapes the social landscape.
Self-Involving Articulation: At the deepest layer, language helps articulate a felt but perhaps vague experience: “I’m nervous about presenting.” This is not just disclosure—it’s also a bid for recognition, connection, and new understanding.
What I’m calling “text-swapping” happens when communication sticks to only the first layer—world-directed reference—and lets the other two wither. The result is a kind of linguistic minimalism, optimized for:
Fluency and form: Is it smooth and grammatically sound?
Topical fit: Does it look like it belongs in this genre or thread?
Low-friction circulation: Will it pass easily through inboxes, slides, and social feeds?
You’ve seen this. Corporate mission statements, clickbait headlines, many LinkedIn posts, and—let’s admit it—plenty of academic abstracts fall into this category. They’re polished, but oddly noncommittal. You’ll rarely find precise referents (“this Friday at 2 p.m. in Room 305”), clear personal stakes, or genuine emotional texture. These texts often function more like tokens in a reputational economy: drop in the right buzzwords, get the likes, avoid friction.
LLMs don’t cause this mode, but they do supercharge it. When you watch a model churn out plausible, contextually appropriate text on command, you may realize how much of your own communication already fits this mold.
3. A Quick Diagnostic Kit for Spotting Your Own “LLM Moments”
Here are some reflective questions to help you recognize when you might be defaulting to a flattened, “text-swapping” mode—and how to correct for it:
Could someone else in my field have written this sentence word-for-word?
→ Add something only you could say: a date, a local reference, a tiny anecdote.Does this sentence bind me to anything? Could someone later hold me to it?
→ Add an accountable phrase: “I’ll circulate the draft by Friday.”If the reader asked ‘Why?’ or ‘So what?’, would my message already contain the answer?
→ Include a rationale or a consequence, even in brief.Am I leaning on stock intensifiers like “excited,” “robust,” or “innovative”?
→ Replace at least one with a specific descriptor or example.Could a three-word paraphrase say the same thing with no real loss?
→ Rework until every word earns its place.Try running your next few emails or posts through this filter. It’s often humbling—and clarifying.
Can you imagine a “communicative utopia” in which everyone consistently applies those five bullet points above? Let’s build it!
4. Exercises for Reclaiming the Deeper Layers of Language
If you want to write and speak in ways that re-engage all three layers—fact, commitment, and articulation—here are some practical methods:
The 1-Fact-1-Stake Rule
Target: Reintroduce commitments (layer ②).
How: For every factual statement, add a forward-moving verb.
Example: “The numbers dipped 3% last quarter. So I propose we revise the rollout plan.”Embodied Anchors
Target: Ground your language in concrete context (layers ① and ③).
How: Include one specific detail—time, place, or sensory texture.
Example: “At 4 p.m., when the sunlight hit the whiteboard, the charts became unreadable.”Mini-Narratives
Target: Invite self-involving articulation (layer ③).
How: Replace abstractions with a short story.
Example: “Two quarters ago, we tried that shortcut. We shipped three weeks late.”Promise Audits
Target: Test the norm-binding force of your message (layer ②).
How: At the end of a meeting or email, list every implied promise. If none appear, ask: Should there be any?“Why Does This Matter?” Loop
Target: Counter hollow fluency.
How: Ask “Why does this matter?” until the answer involves a concrete risk, an affected person, or a core value.
Paid subscribers, I’m sure you probably have a sense by now of what’s coming next! I’ve taken the above, and converted it into a prompt you can use to evaluate, refine, and revise anything you’ve written.