The One Punctuation Habit That Gives Away AI-Written Marketing Copy

Open a thread about spotting AI writing and within a few replies someone mentions the em dash. It's become the shorthand tell, the punctuation equivalent of a watermark, discussed across writing forums, marketing Twitter, and think pieces about the future of content. Fair or not, a paragraph stitched together with em dash after em dash now reads to a lot of people as "a model wrote this," whether or not one actually did.

Riibon is an AI-native agency. We are not writing this to dunk on AI-assisted copy. We're writing it because we use AI heavily in our own work, and we've had to get deliberate about exactly this pattern, so it felt worth saying out loud rather than pretending we're immune to it.

It's not random. Large language models are trained on enormous volumes of text where the em dash functions as a grammatically safe multi-tool, a way to slot in a parenthetical, connect two related clauses, or add emphasis without committing to a semicolon or a full stop. It's flexible and it's rarely wrong, which is exactly the kind of choice a model trained to avoid errors will gravitate toward by default. Typical human writing uses it far more sparingly and situationally.

The result is a subtle but real frequency mismatch. One em dash in a paragraph is a stylistic choice. Three or four, doing the job commas and periods used to do, starts to read as a tic rather than a decision.

Ad copy, landing pages, and email subject lines live or die on feeling like they were written by someone who actually thought about the reader. AI-assisted drafting is now standard in that workflow, quietly, across most serious marketing teams, and there's nothing wrong with that. But readers have started developing a subconscious "this feels AI-written" reaction even when they can't articulate why, and overused em dashes are one of the patterns most commonly cited as the trigger.

That reaction is a conversion problem before it's a taste problem. A headline that pattern-matches to "generic AI copy" gets skimmed past faster than one that sounds like a person who understands the product.

The em dash is just the most visible example of a broader failure mode: any stylistic habit deployed reflexively instead of deliberately reads as generic and slightly uncanny, no matter who or what produced it. Starting every third sentence with "Additionally." Reaching for a rhetorical question to open every section. Falling into the same "X, Y, and Z" triadic list rhythm again and again. A human copywriter with a favorite crutch has the exact same problem as a model defaulting to its statistically safest option.

The fix doesn't depend on whether AI touched the draft. Read your own copy back and ask, for every em dash, every rhetorical question, every list of three, whether it's there because it's the clearest way to say the thing, or because it's just what came out. Cut the ones that fail that test. That's the whole exercise, and it works the same whether you wrote the first draft yourself or asked a model to.