Quick disclosure before we start: this list has five points, and yes, our posts often land on three key bullets in our other stories. That’s on purpose. Three is a classic rule from old-school copywriting and journalism, long before any chatbot existed, because it’s the number a reader’s brain holds onto best. If a memo you get looks suspiciously tidy in threes, don’t assume a robot wrote it, because this is an example of where it didn’t. 

1.        Everything comes in threes, and both sides get equal time. AI-generated text defaults to triads: three reasons, three examples, three takeaways, whether the topic breaks down that neatly. It also tends to present “both sides” of something even in a quick internal memo where nobody asked for balance (“while this approach has benefits, it’s worth considering…”). A person who needs something done usually just says the thing.

2.        Formatting is doing the job that words should be doing. Watch for constant em dashes (these things: “—”) used as a tic, bolded mini-headers, and bullet lists breaking up what should be a normal paragraph. A real memo does occasionally use the “—” for emphasis. AI output often looks like a slide deck got compressed into an email.

3.        The rhythm is too smooth. Every sentence is roughly the same length, every paragraph runs three to four sentences, and nothing sounds like a person talking. Real writing, especially something typed quickly between meetings, has bumps: a one-line paragraph next to a long one, a fragment tossed in for emphasis, maybe even a typo or too.

4.        The vocabulary gives it away. A handful of words show up constantly in AI writing and rarely anywhere else. Here’s a list of the most common: delve, underscore, leverage (as a verb), robust, tapestry, pivotal, testament, foster, streamline. Same goes for stock phrases like “in today’s fast-paced world,” “it’s important to note that,” and signoffs like “let’s leverage this opportunity to drive success.” 

5.        It’s polished, generic, and a little hollow. This is the hardest one to fake around and the most reliable. AI output hits every expected point on a topic without the specific detail, opinion, or lived-in friction that someone with real context would naturally include. No mention of yesterday’s meeting, no office shorthand, no personal read on the situation. It reads like it could have been written about any company, because it probably could have been.

One honorable mention worth a quick scan: unedited leftovers like “[Insert Name Here]” or raw markdown symbols like “***” that never got cleaned up. That’s embarrassing.

None of these points prove anything on their own. A careful writer can hit all five naturally, and someone using AI can edit every one of them out before sending. It’s the pattern, not any single sign, that’s worth trusting.