You’ve heard it by now - the em dash means AI wrote this. The overly structured part too: formal, bullets, lots of bold. Clear hallucinations are of course a sign. Here are some other things I’ve picked up on in the last few years.

Beware of the music. AI writing flows too well. AI would never suddenly say, “beware of the music” in this context because it likes neatly structured, logical flow, with fairly consistent sentence length and a rhythmic quality. It’s why we sometimes tell it to “Mix short, impactful sentences with longer, more detailed ones to make writing more engaging and human.”

Most people know AI makes confident, sweeping generalizations. Like I just did there. (I can’t know that.) Language models are filling in gaps with confidence, and it doesn’t matter if there is no source for their sweeping conclusion. Watch for these mini-hallucinations.

Directly related, AI will tell you the inside of the minds of the people it uses in its writing. Unless we’re writing fiction, we don’t get to see inside the minds of others. It’s like AI is telling us the mental state it is guessing, where a human writer wouldn’t give another person’s mental states, motive, and emotions without direct attribution.

Also, while I’m on this people point: AI quotes are perfect. Real speech is a mess. Even a great orator like Tricia won’t always use the right word, will speak with “so,” “uh,” “anyway,” and the like. But AI quotes are made up on the spot by an expert speechwriter.

“There’s an old saying in Tennessee—I know it’s in Texas, probably in Tennessee—that says, fool me once, shame on—shame on you. Fool me—you can’t get fooled again.” -George W. Bush

Despite the em dashes, that’s a real quote. Similarly, I get suspicious around “studies show” or “cases reveal” where there is no link to such studies or reports. It’s easy for AI to make hallucinatory claims, like illusory quotations, and a quick bit of Googling will often show a lack of real evidence.

Back to writing styles over false claims. Like contrastive focus, emphasis and pacing. Here’s an example: Running laps in PE when they’ve misbehaved teaches children exercise is punishment – real fitness is empowerment. That was written by me… or was it? It’s a way of simulating rhetorical flair, and wow is it most LLMs favorite. It’s fine occasionally, but every paragraph doesn’t need a twist.

We’re at a crossroads because what I’ve shared is the tip of the iceberg. They’re not just fluff (lol), they’re generic analogies. They are classic metaphors used without critical and intentional thought behind them. Meanwhile, real writers have been taught to avoid cliches like the plague, and President Bush screwed up a common one in his quote above, like a human.

This writing style arises from the fundamental way large language models are trained. They aren’t trained to hold facts or beliefs, these models generate text by predicting the most statistically probable next word based on vast amounts of training data. As a result, the output tends to reflect patterns that are common, coherent, and contextually appropriate. But not original, spontaneous, or based in relevant fact–which happen to be some of my favorite things to read.