AI-assisted output has a smell. Not always immediately, but give it a few months and the patterns become unmistakable. That is the core observation in this post by Shiv, and it has direct implications for anyone shipping AI-generated content or UI.
The author started using LLMs to polish a math blog. At the time, the output felt genuinely better: richer vocabulary, more varied sentence structures. Three months later, those same structures were appearing across the entire internet. The improvement was real, but it was also borrowed from a shared well. When everyone draws from the same well, the water tastes the same.
On the writing side, the patterns are specific. Punchline sentences like "Symmetry becomes a trap" and "Cringe is the visible signature of moving along a gradient you chose." Consecutive short declarative sentences stacked for rhythm. The construction "X is the Y of Z." The formula "not just X, its Y." These are not random quirks. They are convergent outputs from models optimizing for a similar aesthetic target.
The UI side is equally concrete. AI-generated websites are clustering around JetBrains Mono as a font choice, identical step-and-bullet layouts, the same button shapes, the same card components, and a blinking-dot badge element. The author does not name specific tools responsible, but the visual convergence is visible enough that the pattern is now a recognizable fingerprint.
This matters for builders for a straightforward reason: if your product surface, whether that is documentation, marketing copy, onboarding flows, or generated UI, looks and reads like everything else, it signals "AI-assisted" in a way that erodes trust and distinctiveness. Your brand is being flattened into a shared default aesthetic.
The author is not arguing against AI usage. The post closes with a clear disclaimer: this is observation, not condemnation. But the observation is useful data.
What should you do with this today? Audit your AI-assisted outputs against the specific patterns listed: stacked short sentences, formulaic punchlines, the "X is the Y of Z" construction, and on the UI side, the font and component choices that have become default outputs. Build a short internal checklist. Run your generated content through it before publishing. If your team uses LLMs for writing or UI scaffolding, add a revision pass specifically targeting these convergent patterns. The goal is not to hide AI usage. The goal is to make sure your product still sounds and looks like your product.