Can You Tell If an AI Wrote That Book?
An author ran an experiment to find out whether readers could distinguish AI-generated text from human writing. The results suggest the gap between the two may be narrower than many assume.
AI writing does have patterns. Writer Imogen West-Knights identifies telltale signs: negative parallelisms, excessive metaphors and similes that don't quite fit, and repetitive syntactical blocks where every noun carries an adjective.
Yet the publishing industry hasn't found these patterns easy to spot in practice. AI models train on human writing-both good and bad-which means a chatbot can produce text that resembles human prose. The problem intensifies with short passages, where AI's characteristic flatness has less room to surface.
Journalist Vauhini Vara decided to test this directly. She asked readers familiar with her work to identify which pieces she had written and which an AI had generated. Her goal was to challenge what she sees as a widespread misconception: that AI writing follows a fundamentally different pattern than human writing.
What the Experiment Reveals
The assumption that AI language is obviously artificial hasn't held up under scrutiny. When fragments are short, distinguishing features disappear. Readers struggle to identify the source.
For writers, this matters. Understanding how AI generates text-and how closely it can approximate human work-changes how you approach your own writing and assess potential tools.
Learn more about how generative AI and language models work, or explore AI resources for writers.
Your membership also unlocks:
