Core Distinction
Voice Is the Moat
Voice Is the Moat names the one advantage no competitor with the same AI tools can buy: the way you sound after years of becoming who you are. Generative models drift toward the safest average version of any output by default. That is not a flaw. It is how the training works. Left unprotected, months of generated output quietly sand a distinct voice down toward that average. Protected on purpose, voice becomes the moat: the thing that stays yours when every tool is available to everyone.
The one advantage no competitor can buy with the same tools: a voice built over years, protected on purpose.
Updated July 2026
Voice Is the Moat
Every tool you use is for sale to everyone else. The same models, the same automations, the same stack. Whatever advantage a tool gives you lasts exactly as long as it takes a competitor to open an account.
A voice is different. The way you frame a problem, the words you refuse, the rhythm a reader recognizes in three sentences. That took years to build, and it is not for sale anywhere.
Voice Is the Moat names that asymmetry. In a market where AI equalizes everything buyable, the unbuyable thing carries the weight.
Regression to the mean
Generative models are trained on enormous averages of how everyone writes. Ask one for a draft with no direction and it returns the statistically safest version. The average.
The old statistical name for this pull is regression to the mean. The model draws every output toward the center of everything it has seen. Your voice lives somewhere off that center, and that distance is exactly what makes it yours.
So the default motion of every AI collaboration is toward the middle. Nothing malicious. Just gravity.
The slow sanding
The erosion is quiet. One generated draft reads fine. So does the next. Each one sits a few degrees closer to the average than what you would have written, and each one trains your ear to accept the drift.
Months later the work sounds like everyone else's work. Not because AI wrote it, but because nobody was holding the line while it helped.
Readers notice first. The voice they came for stops being there, and they rarely say so. They just stop coming.
Held on purpose
A voice survives AI collaboration the way anything survives: by structure, not by vigilance. That means naming what your voice is, in writing, precisely enough that a model can be held to it. What you sound like. What you refuse. The moves that are yours.
With that named, every collaboration starts from your edges instead of the model's center. The tool still moves fast. It just moves fast in your direction.
This is the practical half of the claim. The moat is real, and it holds only when someone maintains it. The maintenance is a structure, written once and applied every time, which is a far lighter discipline than proofreading your own personality back into every draft.
AI for the Business You Actually Want is the Field Guide that turns this into practice as seven Moves, including the voice brief: naming what you sound like precisely enough to hold every AI collaboration to it. The reading is the on-ramp. The Move is the point.
The B1 Field Guide · $9Frequently Asked Questions
Why does AI make everyone sound the same?
Generative models are trained on enormous averages of how everyone writes, and with no direction they return the statistically safest version of any output. The average. Everyone who uses the tool without holding a voice standard gets pulled toward the same center, which is why so much AI-assisted writing reads as one interchangeable register.
What does regression to the mean have to do with AI writing?
Regression to the mean is the pull every model output feels toward the center of its training data. A distinct voice lives off that center, and the distance is what makes it recognizable. Every undirected AI collaboration closes that distance a little. The mechanism is not a defect. It is how the technology works, which is why protecting a voice takes deliberate structure rather than better hoping.
Is voice really a business moat?
Yes, and increasingly it is the only one a small operation holds. Features can be rebuilt, tools can be bought and distribution tactics can be copied the week they work. A voice a reader recognizes in three sentences took years to build and is for sale nowhere. When AI makes everything buyable equal, the unbuyable advantage carries the business.
How do I keep my own voice when writing with AI?
Name the voice in writing, precisely enough that a model can be held to it: what you sound like, what you refuse, the moves that are yours. Bring that standard into every collaboration so the work starts from your edges instead of the model's center, and review output against the standard rather than against your mood. Structure holds the line better than vigilance does.
Can AI learn my voice from examples?
Examples help inside a single conversation, and the drift returns the moment the context resets. A written voice standard outlasts any one session because it travels with you into every new collaboration. The examples show a model what you sound like. The standard tells it what to be held to, and only one of those survives tomorrow.