The tells of AI-written text and AI-built websites

The polish that LLMs add to writing was impressive for about a month. Then, as developer Shiv argues in a post called "Various LLM Smells," the same patterns started showing up everywhere, and now they read as a signature rather than a style. He catalogs the tells. In prose, there is the dense run of punchlines, where every other sentence tries to land as a quotable line like "Symmetry becomes a trap." There is the staccato of consecutive short sentences used for weight. There is the "X is the Y of Z" formula, and the "not just X, it's Y" move that frames an ordinary thing as secretly more.

The design half is just as recognizable. Shiv points to AI-built sites that reach for JetBrains Mono, lay out identical numbered steps and bullet blocks, reuse the same button and card components, and sprinkle little blinking-dot badges. None of these is wrong on its own. The problem is the sameness: once you have seen the pattern a few times, you cannot unsee it, and content meant to look sharp instead looks generated. His post is here. The useful takeaway is not a rule against any one device but an ear for the whole cluster appearing together.

Why it matters

If you ship writing or interfaces with AI help, these are the tells your readers are starting to clock, so the edit that strips the most obvious ones keeps your work from being filed under generic.