Could the next great novel be written by AI (and would you even be able to tell)?
Get the Culture newsletter
Daily culture — film, music, books, the trends and ideas worth your attention. Free.
- Claire Hardaker, a forensic linguist at the University of Lancaster, runs the 'Bot or Not' test where users identify AI-generated hotel reviews — and most people get it right only about 60% of the time, often leaning on tells like em dashes and the 'rule of three' that are equally characteristic of human writers from Dickens to Caesar.
- Hachette withdrew debut horror novel 'Shy Girl' after online rumours of AI use (denied by the author), and Steven Rosenbaum's 'The Future of Truth' was found to contain hallucinated quotations — cases illustrating how suspicion has overtaken verification across publishing.
- Pangram, which claims false positive rates of about 1 in 10,000, was fooled on the first attempt by this article's author; Hardaker says commercial detectors routinely flag neurodivergent writers as AI and can be defeated by 'humanizer' apps, calling herself 'extremely sceptical' of their court-room reliability.
- Researchers traced the surge of 'delve,' 'showcase,' 'boast,' 'underscore,' 'garner,' 'align,' 'surpass' and 'intricate' partly to underpaid reinforcement-learning workers who treated them as proxies for quality during model training — and the frequency of 'delve' in academic abstracts actually dropped after it was publicly singled out in 2024.
- Studies of thousands of unscripted conversations show 'delve' and 'boast' spiked after ChatGPT's release, evidence that AI vocabulary is bleeding into everyday human speech; separately, LLMs flatten non-Anglo-American English toward a U.S./U.K. standard in a process researchers term 'cultural ghosting.'
- Novelist Gary Shteyngart, who teaches at Columbia, said his graduate students wrote angry letters after a peer proposed using AI in a workshop piece — calling the technology an 'assault' on the implicit bargain that literary fiction is a 'Vulcan mind meld' with another human consciousness.
Why it matters: Distinguishing AI from human authorship is breaking down in both directions: detection tools are unreliable (Hardaker's 60% accuracy rate, easily fooled Pangram), and studies confirm AI vocabulary is measurably reshaping everyday speech — with 'delve' and 'boast' spiking in unscripted conversations post-ChatGPT. For publishers, literary-prize judges, and newsroom copy desks already fielding reader complaints about AI text, the technical fixes don't work and the suspicion economy is here to stay.
