Universities, media outlets, publishers, and tech companies have started using AI-content detectors. As a result, everyone who writes for a living suddenly had to rework their texts so they wouldn't get flagged as AI (deleting long dashes, etc.). Bloomberg reports on how artificial intelligence is reshaping human language — and it's happening even to people who don't use AI at all.
How Pangram cost a writer her book deal
American writer Mia Ballard recently lost her contract with publisher Hachette after being accused of using AI. The service Pangram detected signs of machine-generated text in 78.3% of her book.
Pangram is the most popular AI detector — it scans texts for patterns characteristic of neural networks (the mirror image of the invisible watermarks Anthropic now embeds in Claude's output). It's convenient for publishers and universities: upload the manuscript, get a percentage. The service genuinely does a decent job at detection, but it offers no 100% guarantee. And crucially, it can't actually determine how a text was created. Still, its scores are already becoming grounds for very real accusations.
Ballard denies using AI to write the book. For her next work, she deliberately dropped some of her more polished turns of phrase and simplified her language to look more human. The detector still flagged signs of AI in 37% of the first chapter.
People are learning to write like non-AI
According to Bloomberg's reporting and interviews, a lot of authors are now ditching long dashes, rhetorical repetition, and ordinary words that have suddenly become ChatGPT "tells." Meanwhile, people who actually do generate text with AI are deliberately throwing in typos and clunky phrasing to compensate. Pretty funny, if you think about it. Humans are trying to write like humans, and machines are trying to write like humans trying not to write like machines 😭
Why it matters
The part that bugs me most is that we're gradually starting to shape our own language around what algorithms think human language should look like. It's a bit like the three stages of "cognitive capitulation" a Claude Code engineer once described — except here it's not a coding skill that's surrendering, it's speech itself. Authors now have to think not just about whether readers will like the text, but whether some detection software — or commenters who think an em dash is an AI bug — will accept it as human. That's just depressing.
Average text quality is going up. But everything is starting to sound the same. Look at LinkedIn, which I despise — it's a dumpster of sleepy AI news takes and generated comments. The texts are, sure, structurally cleaner and more grammatically correct now — but they've lost all individuality. And the scary part is that on one hand, people are getting used to it (humans get used to anything), while on the other, all this content — and the hours we spend interacting with neural nets — is bleeding into our own, once-unique, speech. So please, let's talk to each other more, so something genuinely unique can come out of it, even with the occasional grammar mistake!
Otherwise, you know what they say about the company you keep…
P.S. For what it's worth, according to Pangram, this very text is 100% Human Written (proof).