• Books
  • Is ChatGPT a good writer?

Column

Is ChatGPT a good writer?

Literature editor Ceri Savage discusses how AI is affecting the literary scene.

IMAGO / Wolfgang Maria Weber

In the summer of 2022, I met the founder of a Berlin publishing company that used early AI-writing tools like Rytr to write novels. They were quickly going out of business. The writing made absolutely no sense and took too long to edit. Later that year, OpenAI launched ChatGPT, gaining over one million users in less than five days, and AI writing has only grown more sophisticated since. I don’t know if that company is still in business, but if it is, it isn’t alone, because using AI to write a novel is no longer the novelty it once was.

AI is all over publishing. According to a BookBub survey, 45% of authors use it. Penguin Random House recently filed a lawsuit against OpenAI, alleging that ChatGPT mimicked and sold a popular German children’s book series, Ingo Siegner’s Coconut the Little Dragon. Last year, the AI firm Anthropic agreed to pay $1.5 billion to settle a class-action lawsuit filed by authors who claimed the company stole their work to train its AI models.

That’s not to mention the Shy Girl drama.

Last month, Hachette Book Group withdrew Mia Ballard’s horror novel after online allegations that it heavily relied on AI. Shy Girl was discontinued in the UK and its US launch date was cancelled, despite Ballard’s insistence that it was her editor who used AI, not her. Nevertheless, nothing got past the readers, who commented on Reddit and Goodreads that the text included the hallmarks of AI slop.

In the original Reddit post, ‘Shy Girl by Mia Ballard. Does anyone else think this was written by ChatGPT?’, the user points out some linguistic tells. One is that “almost every noun has an adjective, and almost every action has a simile”. Another is the sentence construction “something x, something y”, exemplified in the Shy Girl sentence: “In his hands is a cake, gleaming, its pink frosting too smooth, like plastic dipped in sugar, like something that belongs on a screen.” Another is, “This isn’t x – it’s y.” Further YouTube videos also picked up on extreme word repetition, with “sharp” appearing 159 times throughout the book, and kitsch AI metaphors that appear profound but don’t really make sense.

While it’s satisfying that readers recognised ChatGPT’s style and promoted literary justice, I can’t help but feel that Mia Ballard is the made-an-example-of sacrifice to a problem that will only get worse. As readers learn AI’s style, AI learns not to implement it. That’s the point of a machine learning model; it learns. Following a system update in 2025, Sam Altman confirmed that ChatGPT already uses fewer em dashes than it used to. It won’t be long before it reduces its other tells, especially when it’s data scraping the very platforms on which they’re outlined.

I’m not against ChatGPT because it’s a bad writer. I’m against ChatGPT because it makes you a bad writer.

We might sit here all day pointing out why AI is bad at writing. But that’s the problem: it’s not. Most AI-using authors just don’t get caught, and in a system that allows it (notice that Mia Ballard’s readers flagged the issue, not her publishers, who remained conveniently ignorant until others pointed it out), it will become harder and harder to recognise. All this is to say, I don’t think you should use AI to do your writing for you. But I’m not against ChatGPT because it’s a bad writer. I’m against ChatGPT because it makes you a bad writer.

Writing is a craft, and like any craft, it only improves through constant practice. You get better by putting words down repeatedly, just as an athlete refines movement through drills or a musician masters an instrument through scales. AI, though, trains you to give instructions rather than to write. As dependence on it grows, prompting risks becoming to writing what typing is to handwriting: a practical shortcut that sidelines the deeper ability. Over time, it can dull creativity, smoothing language into predictable patterns and draining it of spontaneity, until imagining vividly or bending the rules becomes harder. So resist the pull. What feels easy now may cost you later. Even if you believe you can spot AI-generated text today, that certainty won’t last – and chances are, it’s already less reliable than you think. After all, ChatGPT has been writing this column for me since the beginning of this paragraph. Did you notice?