The Flattening

Unlock seamless experiences tailored to your unique needs.

That sentence is live right now on the website of a boutique hotel. It is also live, nearly word for word, on the website of a company that sells accounting software. Different industries, different countries, same sentence under both logos — I found them in the same afternoon, two tabs side by side.

Neither company copied the other. They copied the same machine.

Last essay, I left a thread hanging: the largest retreat into the familiar ever measured is happening at the exact moment machines began producing most of the world's language. This essay pulls that thread.

The Average

Start with what a language model actually does when nobody intervenes. It predicts the most probable next word. Probable means average. Average means the register that offends no one, surprises no one, and belongs to no one. Ask a default model to write for a Nairobi safari lodge and a Frankfurt fintech and you get the same voice wearing two costumes: polished, warm-ish, weightless.

I have a name for what this does to a brand: identity flattening. The sanding-down of everything that made a voice recognisable (its rhythms, its region, its particular way of being wrong) into competent, frictionless sameness.

Flattening isn't new. Corporate committees were flattening voices long before AI arrived. What's new is the industrial scale. One company drafting by committee produced one bland voice. Millions of companies drafting from the same default produce the same bland voice, everywhere, simultaneously. Sameness used to be a local failure. Now it's ambient.

The Boundary

Here is where it collides with the retreat.

Last essay I argued that circles open from the inside, when something at the boundary registers as safe. Here's the part I didn't say yet: what registers as safe is, overwhelmingly, the familiar. A voice gets admitted when it sounds like it could belong to someone at the table: specific, placed, human in its imperfections.

Default AI language is the exact inverse. It is native to no one. It carries no region, no history, no risk. And readers have learned, fast, to recognise its texture: the tidy triads, the uniform sentence rhythm, the warmth that never commits to anything. Edelman's researchers put the macro version plainly this year: the AI boom has everything it needs except trust.

So the sequence runs like this. Anxiety contracts the circles. The circles filter for the familiar. And at precisely that moment, the majority of commercial language becomes maximally unfamiliar: the same voice from everywhere, belonging nowhere. Every flattened sentence a brand publishes answers the reader's ambient question — is this even real? — with the wrong answer.

I've now watched this land in client work often enough to call it a pattern. A founder spends years building a voice people trust, adopts AI tooling for volume, and six months later her audience is quietly colder. Nothing she published was wrong. It just wasn't hers anymore, and her readers' nervous systems noticed before their opinions did.

Presence

Now the claim I actually want to defend, because it's the one the current discourse gets backwards: AI didn't cause the trust recession. It industrialised the sameness that causes it.

Which means the fashionable responses miss. Banning the tools doesn't restore a voice; committees flattened voices for decades unassisted. AI-detection theatre doesn't either. Readers aren't running detectors, they're running nervous systems, and a flattened human-written sentence fails the same test.

The variable that matters is whether identity survives the drafting process. A brand that treats its voice as infrastructure (documented, protected, checked against every output the way an engineer checks a load-bearing wall) can use these tools at volume and still sound like itself. A brand that accepts the defaults is renting its voice from the average.

The word for language that sounds like it could only come from one particular someone is presence. Presence is what circles admit. And presence, it turns out, can be engineered. Or surrendered one probable sentence at a time.

Next essay: if trust now lives in closed circles, someone has to carry it across the boundary. I gave you a glimpse in Essay I: teachers and scientists still get through. The full rankings embarrass almost every other institution on the list.

The machines will keep writing. Don't let the defaults decide what you sound like.

Emma M. Joseph
Founder, WshUpnASUN® | Creator, CultureLensAI
Essay III — The Broker's Voice — follows next.