Everything is starting to look the same.

In the glow of the cursor, a future unfolds —
a vision of sameness, of stories retold.
It's not just your hero — it's every hero out there,
a seamless, robust, rainbow-gradient affair.
Three feature cards. A sparkle. A grid.
Everything you need. (And everything they did.)

A bot wrote that. A bot wrote every poem like that.

You've been noticing it for a while. In the writing. On the websites. In the little SaaS tool your team started trialling last week that looks, somehow, exactly like the one you stopped using six weeks before. The output has a look now. You can pick it out of a lineup.

Here's a short field guide.

The writing tells

The em-dash — everywhere. The unprompted tricolon: three nouns, three verbs, three clauses, always three. The construction that will not die: it's not X — it's Y. The words: seamless, robust, leverage, unlock, delve. The hero-headline formulas: "Everything you need to run your [noun]." "Built for teams who [verb]." The sentence that arrives balanced and level, as if quietly addressing a small but attentive TED audience.

Yes — the poem.

The made-things tells

The centred hero. The rainbow-purple gradient going diagonally from top-left to bottom-right, which has somehow become a kind of LinkedIn for homepages. Three feature cards, always three, each with one of the same four icons at the top. A sparkle. There is always a sparkle. Pricing laid out in a grid with the middle tier popped up a few pixels, labelled "Most popular" by nobody. An FAQ accordion nobody asked for. A testimonial carousel featuring four photographs of people who might be actors.

Inside the product it continues. The dashboard card with the little sparkline. The empty-state illustration in the same pastel style. Buttons with rounded corners and the suggestion of a shadow. Every one of them. If your site has a sparkle icon, we should probably talk.

Why this is happening

It isn't a bug. It's the mean rendered at speed. Without a steer, you get the average of the training data, and the average is the slop. The models are doing exactly what they were asked to do — make me a website, write me a post, design me a button — and the answer, in the absence of further instruction, is whatever the middle of the distribution looks like.

So: a centred hero, a rainbow gradient, three feature cards. Of course.

A note on our own bias

We build this kind of thing for a living. Read the rest with that in mind.

That said — spotting the slop is the easy part. Everyone can now. The harder part is what goes in its place: a voice that sounds like you, a layout that looks like you, a product interior that doesn't feel like walking into somebody else's office by mistake.

We've spent a long time working out how to get visuals out of the template — what to change, which order to change it in, where the model will fight you and where it quietly gives way. There is a method to it. It just isn't a blog post. It's taste, applied in a lot of small places, usually on the second and third pass, by people who've done it enough times to know which small places matter. The shape of the work that does it is here.

Close

Spotting slop is free.

Not being slop is the work.

P.S.

Reread this. Count the em-dashes — fifteen, in a piece this short. The tricolons: three clauses, three nouns, three beats. The single-sentence paragraph, landed for effect.

Like that one.

It's not that we don't see it — it's that we're writing on the same machines as everyone else. The delve. The unlock. The seamless, robust prose that arrives balanced, level, and quietly addressed to a small but attentive TED audience.

The difference is the next pass.