The AI writing loop is eating itself

TL;DR: In the AI era, people are reworking how they write to dodge accusations that a bot did it for them. Meanwhile, some AI companies are playing both sides of the debate: selling products to help people revise their words and others to help flag AI writing. The whole arrangement is buckling under backlash as the wrongly flagged revolt against AI detectors that still can’t reliably tell humans from machines.

What happened: Big Tech companies and startups are hyping tools to help circumvent AI detection, and according to the New York Times, they’re creating massive headaches in schools. Here are some of the methods:

• Humanizers: Tools like Grammarly and Undetectable AI can swap in synonyms and restructure sentences à la Joey from Friends with a thesaurus switching “big hearts” to “full-sized aortic pumps.” (Signed, Baby Kangaroo Tribbiani.)
• Typomaxxing: Some users are deliberately salting their writing with typos and prompting chatbots to dumb it down. Detectors are apparently less likely to pick up on AI writing when a chatbot is prompted to write like “a college freshman who is a li’l dumb.”
• Autotypers: To beat Google Docs’s version history—often a professor’s last line of defense to see if someone typed something out—some are turning to tools that simulate real typing patterns when copying and pasting AI-generated text by varying keystroke speeds, pauses, and even staging error corrections.

Double agent: The same companies selling detection are also often selling the cheating. Per the NYT, Grammarly tells students to use AI “responsibly” while running TikTok ads urging them to use its app to “choose edits that feel true to you.” Last week its parent company went further, acquiring AI detection startup GPTZero, even as its head of education called the race between detection and evasion “ultimately, a dead end.”

Crying foul: The accusations are already drawing blood. In March, book publisher Hachette pulled a horror novel after widespread allegations that the author heavily used AI to write the book. (The author denied it and said an acquaintance she hired used AI to edit the self-published version.) In the same month, a NYT column about relationships sparked controversy on X after many said the writing looked like AI—some even plugged the essay into various AI detectors, which are notoriously unreliable. (The author said she used chatbots for conceptualizing and editing the piece but did not copy and paste anything directly.)

Bottom line: AI has made good writing a liability—the cleaner your prose, the likelier it is you’ll be accused of using a bot. And some companies are cashing in either way. —LC