AI has managed to do something rather unexpected: it is making people afraid of sounding like themselves. As detection tools such as Pangram increasingly influence debates over whether a piece of writing was created by a human or a chatbot, authors and students are changing their language to avoid raising suspicion. That means fewer familiar writing habits, more deliberate imperfections and, in some cases, people second-guessing perfectly ordinary sentences.
A Bloomberg report has highlighted how AI detection has moved beyond catching machine-generated content. It is now influencing the way people write, even when they claim they have never used AI.
Writers are changing their style to avoid sounding like ChatGPT
One of the most striking examples involves novelist Mia Ballard, whose book Shy Girl was withdrawn by publisher Hachette in March after it received a 78.3% AI-written score from Pangram. Ballard has denied using AI to write the novel and says her editor may have used the technology without her knowledge. The controversy took a serious personal toll, and even her later work continued to attract AI accusations.
The numbers make the situation particularly uncomfortable. Bloomberg reports that Pangram gave the first chapter of Ballard’s Poly, which she says she wrote by hand, a 100% AI-generated score. Her earlier novel Sugar received an 83% score.
Ballard is not the only writer caught in this increasingly messy debate. Authors are reportedly avoiding stylistic habits associated with AI, including rhetorical repetition, parallel sentence structures, and the rule of three. Some are even steering clear of em dashes and deliberately introducing typos to make their prose appear more human. Apparently, good grammar now comes with an identity crisis.
The consequences could extend beyond individual authors. Nigerian writer Jerry Falade has warned that avoiding supposed AI writing patterns could flatten prose rooted in other languages and cultures.
AI detection is creating a new kind of writing anxiety
Pangram says its detector analyses thousands of signals rather than looking for individual words or punctuation marks. Its latest model claims a false-positive rate of one in 10,000, although the company acknowledges that mistakes are possible. The distinction between AI-generated and AI-assisted writing also remains complicated. According to Pangram, asking a chatbot to turn bullet points into an article counts as AI-generated, while using it to rephrase existing human-written material is considered AI-assisted.
Students have faced similar scrutiny, with AI detectors used in universities since 2023. Some have reported being wrongly accused, while tools such as BotOut have emerged to flag words that might trigger suspicion.
The bigger concern is that the line between human and machine writing is becoming increasingly blurred. Linguists say people are already absorbing and avoiding chatbot language, creating a hybrid style. As AI tools and detection systems evolve, writers may find themselves spending as much time proving their work is human as actually writing it.
And this is where I find the whole debate particularly frustrating.
Writing is already difficult enough without having to second-guess every sentence, punctuation mark, or choice of expression because it might look too much like something an AI would produce. I don’t want to write a sentence differently just because a detection tool might misunderstand it. I don’t want to deliberately make my work less polished, throw in awkward phrasing, or introduce mistakes to prove that I am human.
As someone who writes for a living, I want my work to be judged on its ideas, reporting, research, and the effort behind it, not on whether an algorithm approves of my writing style. AI can certainly help people write, but we should be careful about letting machines decide what authentic human writing is supposed to look like. If the price of proving that I am human is making my writing worse, then perhaps it is time we started questioning the detectors, not the writers.
RELATED COVERAGE:
