Em dashes used to be a stylistic choice - the punctuation equivalent of a dramatic pause, used by writers who wanted to add weight to a mid-sentence break. Now they're one of the first things readers flag when they suspect AI-generated text.
The reason is specific: Claude uses em dashes constantly. Anthropic's model structures sentences with em dashes where a human writer might use a comma, parentheses, or just a period. ChatGPT does this too, though less heavily. When enough AI-generated content flooded the internet carrying the same stylistic fingerprint, readers learned to pattern-match. Now the em dash is cargo - a signal that something was probably written by a machine.
The Pattern Corruption Problem
The "AI slop tax" describes what happens when AI models colonize a human writing convention. Em dashes have been a valid grammatical tool for centuries - Emily Dickinson used them obsessively, and good editors defend them. But once AI adopted the pattern in massive volume, the association shifted. A writer who reaches for an em dash now risks having their work read as AI-generated, even when it isn't.
This is bigger than one punctuation mark. Several writing patterns have gone through the same cycle. "It's important to note" was a legitimate academic phrase before AI started using it everywhere. The same happened to "In today's fast-paced world" as an opener, to bullet lists with exactly three points, to paragraphs that end with a call to action in every section. Each started as a human convention. Each got adopted by AI models at a frequency that turned it into a red flag.
The mechanics are straightforward: AI models train on human text, so they pick up human stylistic patterns. But they don't reproduce them at human frequency - they overrepresent patterns that appear heavily in online writing like blog posts, LinkedIn articles, and newsletters, because that's what training data skews toward. The overuse makes the pattern legible as AI, and readers start treating it as a warning sign.
Who Pays
The cost falls unevenly. Writers who've always used em dashes - and there are many - now face a credibility penalty for their own stylistic preferences. They either keep using their preferred punctuation and risk being flagged, or they adapt their style around whatever AI hasn't yet polluted. That's a real constraint imposed by tool behavior they had no part in.
The practical fix is editing. If you're using AI tools to draft content, em dashes are the first thing to look for - not because they're wrong, but because they're visible. Two in a piece reads as intentional. Eight reads as Claude wrote it and nobody checked.