Forget em-dashes: A viral report on AI-generated writing has surprising new clues
As AI -generated content takes over the internet, humans are becoming more and more vigilant about telling bot-written copy from genuine human writing. Last week, LinkedIn even became the first major platform to add a button that lets users report “AI slop.” But how can users be sure what they’re reading is actually the product of an LLM? While there are some widely regarded tells of AI writing,…
As AI-generated content dominates the internet, individuals are increasingly vigilant in distinguishing bot-written material from authentic human writing. Last week, LinkedIn announced the introduction of a button to report "AI slop." However, the reliability of certain indicators of AI writing may not be as consistent as previously believed.
A recent report by The Economist examined the current state of AI writing in 2026, unveiling the telltale patterns of AI writing and the misleading red herrings that do not actually point towards AI. The Economist compared its articles to versions generated by leading AI models, including OpenAI's ChatGPT, Anthropic's Claude, Google's Gemini, and XAI's Grok.
The analysis also encompassed content from other notable news organizations such as The New York Times and The Washington Post, as well as excerpts from well-known novels published between 1950 and 2022, totaling 55,940 sentences and 1.2 million words.
What not to scrutinize is em-dashes. Once a popular punctuation mark among writers, AI has led many to avoid using em-dashes to prevent readers from suspecting their work was generated by a bot. However, this punctuation mark is no longer a definitive indicator of AI-generated content. Only Claude, one of the major models tested, used em-dashes more frequently than human writers.
In fact, a lack of punctuation is a more reliable sign of AI-generated text. Moreover, AI models employ fewer commas, semicolons, and parentheses compared to humans, often producing excessively long sentences with "and" as their most frequently used word.
Despite the common perception, AI-generated text is known for its verbosity. According to The Economist's report, LLMs tend to use rarer words and scientific terminology more often than humans. They also favor polysyllabic words and nominalizations, which are nouns and adjectives derived from verbs, such as "nominalization." Additionally, AI models are notorious for employing rhetorical conventions like "it's not X, it's Y" and the rule of threes, which contribute to their sentence and paragraph structure.
AI-generated writing comprises long sentences with minimal variation in length, resulting in blocky paragraphs with consistent sentence structures.
Another potentially accurate perception of AI is that it learns quickly. AI models continuously evolve and are trained on human writing, making it progressively challenging to differentiate between human and AI-generated writing. Not long ago, ChatGPT was a significant user of em-dashes. However, it now uses them less frequently than any other model and significantly less than humans.
The guidelines for identifying AI-generated content are continually evolving—but according to The Economist's report, as of now, verbose writing with minimal punctuation remains the most probable cause.
Written by urgent.news from Fast Company's reporting — not their text. Machine-written — it may contain errors, so check the original before relying on it.
