NEW YORK – July 5, 2026 – A new forensic linguistics test is exposing a harsh truth: most people cannot reliably distinguish human writing from machine-generated text, even as accusations of AI use rock the literary world. The findings, released by Lancaster University professor Claire Hardaker, show that average users correctly identify AI-written hotel reviews only 60% of the time—barely better than a coin flip.
The controversy comes amid a summer of high-profile scandals. In May, social media erupted when doubts surfaced about Jamir Nazir's prizewinning short story, with users claiming they could "just tell" it was AI-generated. Nazir later denied using artificial intelligence. But the incident has fueled a broader panic: publishers like Hachette recently pulled debut horror novel "Shy Girl" after unsubstantiated rumors of AI authorship spread online.
Hardaker's Bot or Not test, which asks users to spot fakes among 15 reviews, reveals the pitfalls of amateur detection. Many rely on simplistic rules—flagging cliches, em dashes, or the "rule of three" (a trio of words like "Veni, vidi, vici"). Yet these same features appear in classic human writing. "You could go back to Charles Dickens and say he had AI, because he used the em dash too," Hardaker noted. The problem, linguists say, is that large language models were trained on human texts, so they naturally mimic our patterns.
The confusion has real consequences. Novelists including Jennifer Egan and Jeanette Winterson now face questions about the future of fiction as ChatGPT and similar tools blur the line between creator and machine. For authors, the stakes are personal: a false accusation can destroy a career. For readers, the uncertainty erodes trust in every byline.
As the technology evolves, experts warn that relying on gut instincts is no longer viable. "People have learned very simplistic rubrics and now just madly apply them everywhere," Hardaker said. The takeaway? In an age of AI, even the most confident reader may be wrong more often than they think.