AI Writing Cues Shifting, The Economist Report Finds

A new report from The Economist offers an updated analysis of the telltale signs of AI-generated writing, challenging widely held beliefs about its characteristics.

The study compared text produced by AI models, including ChatGPT and Google Gemini, against human-written articles from news outlets and literary excerpts. The analysis spanned over 55,000 sentences and 1.2 million words.

One significant finding challenges the notion that overuse of the em dash (—) is a definitive marker of AI writing. The Economist found that only one tested model, Claude, used em dashes more frequently than humans. Instead, the report suggests that a lack of punctuation and the prevalence of long sentences, with "and" being a frequently used conjunction, may be better indicators of AI-generated content.

However, the report does confirm some previous observations. AI-generated text is often characterized by wordiness, a tendency to use rarer words and scientific jargon, and a preference for nominalizations (nouns derived from verbs). AI also frequently employs rhetorical devices like "it's not X, it's Y" and the rule of threes, which can lead to monotonous sentence structures and uniform paragraphs.

As AI models rapidly evolve, identifying them becomes increasingly challenging. According to The Economist's report, verbose yet punctuation-light writing is currently the most likely indicator of AI-generated content.