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, those stereotypes may not be as reliable as they seem. A new report by The Economist analyzed the state of AI writing in 2026, identifying the telltale AI patterns readers should look out for, plus the red herrings that don’t actually point toward AI.

The Economist compared its own articles to versions of the same articles generated by top AI models, including OpenAI’s ChatGPT, Anthropic’s Claude, Google’s Gemini and XAI’s Grok. Its sample also included writing from other news outlets like The New York Times and The Washington Post, as well as excerpts from popular novels published between 1950 and 2022. Altogether, it compared 55,940 sentences and 1.2 million words. 

What not to look for

Perhaps the most infamous sign of AI writing is its overuse of em-dashes. Though they were long a favorite convention of writers, the rise of AI led many people to strike the punctuation mark from their prose, for fear of readers assuming their work wasn’t human-made.

But according to the report, em-dashes are no longer a surefire sign of AI-generated content. Of the major models tested, only Claude used em-dashes more often than human writers.

In actuality, a lack of punctuation is a better indicator of text being AI-generated, says the report. It also found LLMs use fewer commas, semicolons, and parentheses than humans, instead crafting overly long sentences with “and” as their most overused word.

Some stereotypes ring true

AI-generated text is known for being unnecessarily wordy. And The Economist’s report specifically found that LLMs used rarer words and scientific lingo more often than humans, also favoring polysyllabic words and nominalizations (nouns and adjectives based on verbs, like “nominalization” from “nominalize”).

LLMs are also notorious for rhetorical conventions like “it’s not X, it’s Y” and the rule of threes. This contributes to the way they structure sentences and paragraphs: AI-generated writing tends to feature long sentences with little variety in length, leading to blocky paragraphs with uniform sentences, says the report.

Another unfortunately accurate perception of AI is that it learns fast. LLMs are constantly evolving and being trained on human writing to make these differences harder and harder to spot. Not long ago, ChatGPT was a major abuser of em-dashes. But now, it uses them less than any other model (and far less than humans).

The rules for identifying AI-generated content are constantly shifting—but according to The Economist‘s report, for now, verbose, punctuation-light writing is the most likely culprit.