More Typos, Fewer Em Dashes: Writers Are Building an Anti-AI Literary Counterculture

A growing number of writers are deliberately breaking rules of clean prose—using more typos, fewer em dashes, and intentional grammatical errors—to signal human authorship, according to a Wired article published July 28. Researchers at Stanford tracked a 400% increase in em dash usage across Substack and Medium between 2022 and 2026, correlating with LLM writing tool adoption. The irony, the article notes, is that every stylistic defense works only until models absorb it.

More Typos, Fewer Em Dashes: Writers Are Building an Anti-AI Literary Counterculture A growing number of authors, journalists, and essayists are deliberately breaking the rules of clean prose to signal human authorship. Em dashes — the single strongest AI tell — are being replaced with parentheses. Sentence fragments and intentional typos are becoming stylistic signals. Wired covered the phenomenon on July 28, calling it an 'anti-AI literary counterculture.' Researchers at Stanford tracked a 400% increase in em dash usage across Substack and Medium between 2022 and 2026, correlating with LLM writing tool adoption. The irony: every stylistic defense works only until models absorb it. The Movement A growing number of writers are deliberately breaking the rules of clean prose to prove they're human. More typos. Fewer em dashes. Intentional grammatical errors. It's not carelessness — it's a signal. Wired covered the phenomenon on July 28 under the headline "More Typos, Fewer Em Dashes: Writers Are Creating an Anti-AI Literary Counterculture." The piece describes a quiet revolt among authors, journalists, and essayists who have realized that LLMs produce text with telltale fingerprints — overuse of transition words like "furthermore" and "conversely," perfect grammar, and em dashes that appear with mechanical regularity. The counter-move: write like a human, flaws and all. Why It Works AI detectors flag text that is statistically "clean." When every sentence is well-formed, every transition is smooth, and every punctuation mark is correct, it triggers the probability models that power detection tools. By introducing intentional roughness — sentence fragments, comma splices, inconsistent hyphenation — writers can produce text that scores as human-written. The movement has produced a set of unofficial rules for anti-AI writing: - <<<BOLD Avoid em dashes.<<<BOLDEND They're the single strongest AI tell in prose. Use parentheses instead. - <<<BOLD Vary sentence length aggressively.<<<BOLDEND LLMs default to 15-to-25-word sentences. Throw in some 3-word hitters and some 50-word runs. - <<<BOLD Make typos deliberately.<<<BOLDEND A "teh" instead of "the" in draft text. An extra space. A missing Oxford comma that you'd normally include. - <<<BOLD Use contractions heavily.<<<BOLDEND Models tend to overuse full forms like "do not" and "cannot" in formal contexts. Write "don't" and "can't." - <<<BOLD Break the paragraph structure.<<<BOLDEND LLMs love 3-to-5-sentence paragraphs. Write a one-sentence paragraph. Write a ten-sentence block. The Irony The anti-AI literary movement is adopting strategies that are themselves becoming predictable. If enough writers start using parentheses instead of em dashes, AI models will learn to do the same. Every stylistic tell works only until the models absorb it. A novelist interviewed by Wired acknowledged the paradox. "I'm building my defense on patterns that will be obsolete in six months. But six months of being identifiable as human is worth it." The Deeper Trend This isn't just about evading detection. It's about reclaiming the texture of human writing from the homogenizing force of language models. The em dash proliferation is real — researchers at Stanford tracked a 400% increase in em dash usage across Substack and Medium posts between 2022 and 2026, correlating closely with LLM /glossary/llm writing tool adoption. The backlash is real enough that publishers are now explicitly asking for "non-generic" prose. Some literary magazines include "no AI polish" notes in their submission guidelines. The aesthetic preference is shifting from clean to characterful. Bottom Line If your writing looks like it was generated by a committee of English professors who all read the same style guide, that's the problem. The market for clean, forgettable prose is being flooded by machines. The premium is shifting to writing that sounds like a specific person — typos, quirks, and all. Get AI news in your inbox Daily digest of what matters in AI.