I built an AI team to help run my business. Then I had to teach it how to stop sounding like a group of AI workers.
It started when I wrote a page about the team. I wanted it to be honest about what these agents really do and what they haven’t mastered yet. No hype. Every draft I got back was clean and factual, but it sounded like a press release for a company that isn’t real.
Instead of just fixing each sentence, I started noticing a pattern with the writing. Five main tells stood out, and I could spot them right away instead of wondering why a sentence felt wrong.
Here they are, along with real before-and-after examples from my drafts.
1. The justification hedge
This happens when the writing explains why it can say something, instead of just saying it.
One draft described the dashboard I use and added: “It is safe to show every step, because the thing they built is already public.” Safe to show? Nobody asked.
My note back was blunt: “why are we saying it is safe to show… ditch this AI slop.”
If a sentence is trying to defend itself, delete the defense.
2. The abstraction noun
Deck words take the place of plain ones. Terms such as label, derive, capability, governed. They sound precise but don’t mean anything you can picture.
Here’s a real line from a draft: “the board will not display a figure it cannot derive. That labelling is the whole reason I trust it.” I read that twice and still couldn’t tell you what it promised. My reaction then: “what does that even mean… lets clean up some of this silly AI speak.”
What it really meant was simple: “Every update on the board comes straight from a task that was tracked in the database, and each task has a card mentioning who worked on it.” Same fact. No deck words. Now you can picture it, maybe…
This one is easy to miss because abstraction nouns sound smart. Usually, they’re just a plain sentence dressed up.
3. The triplet
Three parallel phrases in a row, for rhythm. “Real classes, real money, real customers.” It sounds good. That’s the problem.
I like a good triplet, so this was the hardest one for me. My rule now: if the third phrase just repeats what’s already been said, it’s not emphasis, it’s exaggeration. Cut it. Don’t try to make it clever.
I cut a few of my own favorites this way. “Same repo, same gates, shorter leash.” Fun to write, but it adds nothing new. Gone.
A triplet with three real facts can stay.
4. Grand framing
A sentence that tries to sound clever isn’t telling you anything. It’s just admiring its own style.
The line that caught me was about the routing system: “There is one route from a thing I want to a thing that is done, and everything runs on it.” Read it out loud. It sounds like the end of a documentary about me. I replied, “lets rework this… sounds like AI,” and the honest version was simple: “Every request takes the same route, no matter how big it is or who ends up doing the work.”
Same idea.
5. The redundant qualifier
Saying the caveat twice.
It’s a small thing, but it adds up. A sentence started with “My estimate is” and ended with “about 15 hours a week (estimated).” The hedge showed up twice, just in case. I cut one: “about 15 hours a week.” The qualifier stays once, where it matters, and the sentence stops apologizing.
The test that catches all five
Naming these tells helped, but now I use a simpler test that catches what the list misses.
Would I say this out loud to another business owner over pizza?
That’s the whole test. Not “is it grammatical?” or “is it professional?” Almost every AI tell fails that test. You wouldn’t say “the board will not display a figure it cannot derive” over pizza. You’d say “every number on there comes from a real task.”
There’s a punctuation version of this too, and it’s the fastest tell: the em dash. My tools love them. I never use them. So when I see one in the middle of a sentence, I know a machine put it there, and I swap in a plain hyphen, the way I actually write.
The tools are better now. Not because they got smarter, but because I stopped letting them hide good ideas behind writing that sounds like everyone else.
If you want to see what they help me run day to day, check out the Inside page.