How AI detectors like Pangram are changing human writing: authors drop em dashes, "not X, but Y" phrasing, and words like "delve", or add typos, to look human

Writers are now deliberately introducing stylistic imperfections to pass AI detection.

Writers are altering their prose to avoid being flagged as AI-generated by detection tools. Sally McGraw, a ghostwriter, was falsely accused of using AI for a book proposal sample, which she believes cost her the job. This experience, alongside a flood of AI submissions overwhelming publications like Clarkesworld magazine, highlights the challenges editors face. AI detection tools, while improving, are not foolproof and can produce false positives. For example, Pangram was found to be accurate in a study but CEO Max Spero admits it can err, especially with short texts. Professor Chris Callison-Burch notes that AI detectors analyze text predictability, and previously published human text can be misidentified. Common human writing patterns, such as using em dashes, specific phrases like "not X, but Y," or words like "delve," can trigger false alarms. The Authors Guild warns of a "witch hunt" environment, where false accusations can harm reputations and contracts, while false negatives erode reader trust.

The reliance on imperfect AI detection tools is forcing human writers to modify their natural style, potentially stifling creativity and authentic expression. This creates a perverse incentive where the goal is to write *less* clearly or predictably to prove one's humanity, rather than to communicate effectively. The very tools designed to uphold authenticity are inadvertently leading to a homogenization of writing that mimics perceived human flaws rather than celebrating genuine skill.

This situation poses a significant challenge for both creators and content platforms. Publishers and editors must navigate a landscape where genuine human work can be erroneously dismissed, impacting livelihoods and the integrity of published content. For AI detection companies, the pressure is on to develop more accurate and reliable systems that can distinguish nuanced human expression from sophisticated AI output, without penalizing legitimate writers for their stylistic choices.