Typos on Purpose: How Writers Are Fighting Back Against AI Prose

Key points

• Novelist Laura Brooke Robson changed her writing style to avoid patterns she associates with AI: em dashes and tripartite lists.
• Hachette pulled Mia Ballard's horror novel in March 2025 after online speculation it was AI-generated, a claim Ballard rejected.
• Pitchfork's editors sent a memo banning AI from review copy and pushing writers toward their own voice.
• An app called Sinceerly deliberately inserts typos into emails; when tested against Fortune 500 CEOs, only the misspelled subject lines got opened.
• Writers warn that AI could learn to mimic any counterstyle, making genuine skill the only lasting defence.

Laura Brooke Robson is a novelist who now invents phrases on a whim, drops unexpected punctuation mid-sentence, and treats a small grammatical detour as a gift to the reader. She's doing it on purpose.

Robson's new book, Love Is an Algorithm, is a love story set partly inside an AI dating app. While writing it, she became acutely aware of what AI prose looks like and started writing in the opposite direction. No em dashes, no tidy tripartite lists, sentences she says she "would not have attempted before."

"I find myself writing less mechanically," she told Wired AI. "I have this urge to make little mistakes, like winks to the reader. What if I just invented a phrase? What if I did something unexpected with my punctuation?"

She isn't alone.

Why are writers changing how they write?

Writers are reacting to a real problem: AI text generators, the software behind tools like ChatGPT and Claude, have trained on so much human writing that they produce prose with recognisable tics. Passive voice. Em dashes used like punctuation confetti. Phrases that arrive fully formed and go nowhere.

For working writers, being mistaken for AI is no longer just embarrassing. It can be career-ending.

In March 2025, publisher Hachette pulled Mia Ballard's horror novel Shy Girl after speculation spread online that it was AI-generated. Ballard denied the claim. She told Wired AI that her top priority now is making her writing "feel more real and authentic," and that she's started editing her own work because she believes an editor she hired had run the manuscript through AI-detection software. Hachette is also a plaintiff in a federal lawsuit against Google over Gemini's training data, a case we reported in July.

Steven Rosenbaum's book The Future of Truth, an examination of how AI changes reality, faced its own problem: Rosenbaum later admitted it included fabricated quotations that AI had generated.

Reputations gone. Books pulled. Contracts lost.

What does the anti-AI style actually look like?

It's mostly defined by what it avoids. No em dashes, no passive voice, no tripartite lists, none of the "not X but Y" constructions that AI reaches for automatically.

What it adds is harder to copy: specific personal anecdotes, strange similes, deliberate imperfection.

Clickhole writer Jessye McGarry published a piece titled "5 Reasons to Love Summer (NO AI)" stuffed with typos and repeated assurances that a human wrote it. One line described "beaitiful sunshine." The spelling error took effort: McGarry had to override autocorrect to leave it in.

"Whatever cosmetic counterstyle is created in response to AI, AI could mimic," McGarry said. "The only true counterstyle is good writing."

That warning matters. LLMs, the technology that powers these text generators, are trained continuously on new data. A style invented today to signal humanness could be absorbed and reproduced within months.

What does this mean for readers?

For now, the pressure is producing better writing in some corners.

Pitchfork's editorial team sent writers a memo banning AI from review copy and asking for first-person perspectives. When Playboy editor Philip Picardi opened submissions in May, he received around 1,000 pitches and judged most were AI-generated; he's since pushed for formally adventurous, personal writing. Aeon editor Richard Fisher told writers to keep the "rough edges" that AI sands away.

Ben Horwitz built an app, Sinceerly, that does the opposite of a spelling checker. It puts typos into emails and adds "sent from my iPhone" sign-offs to look human. When he A/B-tested standard emails against Sinceerly ones with Fortune 500 CEOs, only the messy ones got opened.

Writing teacher John Warner published More Than Words last April, arguing that AI might restore the value of good writing by making competent-but-generic prose worthless. That's the optimistic read. The realistic one is that the window for any particular counterstyle is short, and writers who rely on surface tricks rather than genuine specificity will find AI catching up fast.

For readers, writing that surprises you, that takes an odd turn or admits a personal detail that serves no tidy purpose, is now a small signal of humanity. It's worth noticing.

Should you worry?

Is AI detection software reliable enough to trust?

Generally, no. Detection tools produce false positives frequently, meaning they flag human writing as AI-generated. Ballard's case illustrates how damaging that can be even when the accusation is wrong.

Do I need to worry about AI-generated writing I read every day?

For most casual reading, the direct harm is low. The bigger effect is a slow flattening of style across articles and social posts. The fix is the same one it's always been: read writers who surprise you.