AI content and the sameness problem

Scroll long enough and you start seeing the seams. The same reassuring cadence. Tidy three-part structures, over and over. A faintly upbeat conclusion that resolves nothing, attached to a thousand different logos in a hundred different industries. The internet is filling up with writing that is competent, grammatical, frictionless, and eerily interchangeable.

No mystery about why. A huge share of it now comes out of the same handful of models, and a model is, at its core, a machine for predicting the most probable next word. Probable is another word for average. Ask the average of all writing to write, and you get writing that sounds like the average of all writing. Every draft is pulled gently toward the center of the distribution, like water finding level.

Let’s be clear about our own position first, because this isn’t a hands-clean lament. We use these tools daily. They’re in our build pipeline, our research process, our first drafts of things nobody should hand-craft. The productivity is real and we’re not giving it back. What we’re describing isn’t a tool problem. It’s a defaults problem, and defaults are a choice.

When a whole category converges

Here’s the mechanism worth understanding. When one company publishes generic content, that company sounds generic. Annoying, survivable. When every company generates from the same models with the same prompts pointed at the same keywords, something different happens: the entire category converges. Ten competing firms publish functionally identical thought leadership in the same quarter. The individual pieces aren’t bad. Collectively they’re indistinguishable, and indistinguishable is the one thing marketing exists to prevent. An arms race where every weapon fires the same bullet isn’t an arms race. It’s a very expensive way to stand still.

Regression to the mean used to take a whole industry years of imitation. Now it ships by default.

But flip the frame and the picture gets interesting. Sameness at scale is a market condition, and market conditions are openings for whoever notices them first. When the baseline of all content converges on smooth and average, the bar for standing out drops. Distinctiveness used to require beating talented competitors. Now it mostly requires sounding like a specific someone in a sea of prose that sounds like no one, and that’s a much softer target than it was five years ago.

So what can’t regress to the mean?

Specificity. The mean is general by definition, so the general is exactly what the machines produce best and what everyone now gets for free. What no model can generate is the detail only you possess because only you were there.

• The client who kept a failed prototype on his desk as a warning.
• What your support inbox actually sounded like the week something broke.
• The reason you turned down the project everyone said you were crazy to turn down, and the second reason, the one you don’t put in decks.

Lived detail is nonreplicable by definition, which makes it the closest thing to a moat that content can have, because a moat is just an asset your competitors can’t copy at any price, and your competitors cannot copy having been in your rooms.

Taste survives too. Tools don’t have any, as we’ve written before. A model will hand you fifty options; deciding which one is true to you is a judgment the machine can’t render, because “true to you” isn’t in the training data.

The division of labor

None of this means artisanal-everything, hand-typed and proud of it. That’s nostalgia cosplay, and it loses on volume. The workable posture is a division of labor: let the models do what averages do well, which is structure, summary, the eighty percent that was never going to differentiate you anyway. Then spend the hours you saved where the average can’t follow. Opinions with costs attached. Details with dates on them. Sentences only your company could type, because they’re made of things only your company knows.

The uncomfortable test: take your latest post, strip the logo, swap in a competitor’s name. If it still reads fine, the models aren’t the problem. The post never contained you to begin with, and the machines merely made that cheaper to discover.

Everyone now has access to infinite average. It’s the new free, the new zero. Nobody can buy their way to distinctive anymore, which was always the honest arrangement anyway.

You can only be someone. Fortunately, that job just got a lot less crowded.