The Economist: ‘How to Spot AI Writing’

The Economist published an article explaining how to identify AI-generated writing by comparing it to human writing. To conduct this study, The Economist compared the writing of four large language models (LLMs) - OpenAI's ChatGPT, Anthropic's Claude, Google's Gemini, and xAI's Grok - to human-written articles. The AI models were given summaries of The Economist's articles as prompts to generate their own versions. This created a large dataset of human and AI-written sentences and words for analysis.

The study found that AI writing can be distinguished by its word choice, punctuation, sentence structure, and paragraph organization. However, these quirks are not immediately obvious and have changed over time due to software updates. Unlike some expectations, AI writing is not marked by overly complex language or the use of Latin or Greek words. Instead, it tends to use more Latinate suffixes than human writing.

One notable difference is in punctuation. The study found that AI models, particularly Claude, use em-dashes more often than humans. On the other hand, AI tends to use fewer commas, semicolons, and parentheses. This is likely because AI generates longer sentences, relying more on "and" as a connector, and does not quote experts as often as humans do.

To avoid writing like AI, human writers should aim to use clear and concise language, avoiding pretentious diction and overuse of Latin or Greek words. They should also incorporate punctuation more extensively, such as commas, semicolons, and parentheses, to give their writing structure and clarity. Additionally, writers should strive to balance sentence length and variety, avoiding overly long sentences that may resemble AI-generated text.