For generations, the em dash—that versatile punctuation mark signaling an abrupt shift or emphatic aside—was a beloved tool for writers seeking rhythm and nuance. Now, it’s becoming a casualty in the war against AI-generated "slop." As observed in a poignant personal essay by software engineer Bassi Li, writers increasingly self-censor their natural stylistic choices, stripping em dashes from their prose purely to avoid triggering AI detection algorithms.

The Rise of the Algorithmic Tell

The core issue lies in how large language models (LLMs) like ChatGPT generate text. Trained on vast corpora, they often default to certain stylistic patterns:

  • Over-reliance on em dashes: LLMs frequently use them to structure complex sentences or inject asides, making them a statistical red flag.
  • Excessive positivity: Generated text often leans towards overly neutral or positive tones.
  • Formulaic list structures: Bullet points or numbered lists are common LLM outputs.

As educators and platforms deploy rudimentary detectors scanning for these patterns, human writers find themselves under suspicion. Bassi Li describes the anxiety: "Nowadays, I find myself avoiding em dashes because I’m afraid that my writing will be flagged as AI-generated and dismissed as slop."

The Meta-Game of Authenticity

This avoidance triggers a damaging cycle:

  1. Self-Censorship: Writers consciously abandon natural stylistic flourishes (like em dashes) that align with AI patterns.
  2. Performance of Humanity: Writers engage in a "meta-game," deliberately introducing grammatical quirks or minor errors—traditionally things to avoid—solely to signal human origin. As Li notes, this involves "choosing my words carefully—typically ensuring that I include the right amount of grammatical character and/or mistakes."
  3. Diminished Expression: The very tools meant for clarity and emphasis are sacrificed, potentially leading to flatter, less expressive human writing.

Beyond Punctuation: The Chilling Effect of LLMs

The em dash is merely the current canary in the coal mine. Li raises a critical concern: "I’m curious (and more than a bit worried) that the writing that is being produced these days is being shaped by LLMs, even if an LLM has never touched a particular piece of prose." The collective awareness of AI "slop" aesthetics means human writers constantly self-edit based on perceived reader expectations shaped by low-quality LLM output.

The implications are profound:

  • Moving Targets: If future LLM models shift towards overusing semicolons or other structures, human avoidance patterns will shift accordingly, creating an endless stylistic arms race.
  • Loss of Nuance: The pressure to avoid detection could homogenize writing styles, stripping away individuality and subtlety.
  • AI's Soft Power: LLMs exert influence not just through direct use, but by altering the creative landscape simply through their existence and the detection mechanisms they necessitate. Li laments, "If an em dash fits into one’s writing but they avoid using it out of fear, our AI overlords have won."

The Unintended Consequence

This phenomenon highlights a critical flaw in the current approach to AI detection: it penalizes human stylistic choices that statistically overlap with AI patterns. The burden falls on writers to contort their natural voice to prove their humanity, while detection tools remain blunt instruments. For developers and technologists building these tools, the challenge is clear: Can we create detection methods sophisticated enough to discern genuine human expression from machine-generated text without forcing writers to abandon the very tools that make their work unique? The future of authentic human creativity online may depend on the answer.

Source: Based on the essay "I miss using em dashes" by Bassi Li (https://bassi.li/articles/i-miss-using-em-dashes).