The Tells That Make a Post Read Like AI

Readers have become quick at this. Not analytically — almost nobody could tell you what tipped them off — but the judgement arrives within a line or two, and once it has arrived, the post is over. Whatever you were arguing is now being read as output.

The judgement is triggered by a small set of recognisable patterns. They are checkable, they are mostly removable, and none of them require you to know anything about how the models work. Here they are, concretely, with what to do instead.

One thing up front, because it will otherwise nag: several of these are ordinary writing devices. The tell is almost always density, not presence. One is style. Four in a row is a signature.

1. The scaffold opener

“In today's fast-paced business environment…” “In an era where technology is reshaping how we work…” “As businesses navigate an increasingly complex landscape…”

A first sentence that could open a post about literally anything. It exists because a generator needs to start somewhere and general truths are always available.

The fix: delete the first sentence. Genuinely — try it on your last five drafts. In most cases the second sentence was the real opening, and the post gets better immediately.

2. The not-X-it's-Y reversal

“It's not a tool. It's a teammate.” “This isn't about efficiency — it's about focus.”

A rhetorical move that manufactures the feeling of insight by negating something nobody claimed. One in a post is fine and sometimes earns its place. Three is the single most reliable tell in circulation right now, and readers have started naming it out loud.

The fix: keep at most one, and only where the X is something a reader actually believes. If nobody was arguing for X, you've built a straw man and dressed it as a revelation.

3. Over-symmetry

Everything in threes. Three-item lists, three-clause sentences, three paragraphs that open the same way. Balanced constructions where each half mirrors the other.

Human writing is lumpy. People make one point at length, then another briefly, then remember something and add it. Machine writing is even, and evenness at scale reads as manufactured — often before the reader can say why.

The fix: count your lists. If everything arrives in threes, make one of them two items or five. Break a parallel structure on purpose.

4. Reflexive hedging

“This can help teams potentially improve…” “It may contribute to better outcomes…” “Many businesses often find that…”

Multiple hedges stacked in one sentence. It happens because a generator has no stake in the claim and defaults to safety. A founder with a real opinion states it, and takes the consequences of being wrong.

The fix: find every can, may, might, often, tends to, could potentially. Delete all but the ones carrying genuine uncertainty. The sentence usually gets shorter and much better.

5. Listicle cadence where it doesn't belong

Every idea rendered as a numbered item with a bolded label and two supporting lines. Fine in a blog post like this one, where the reader is scanning for structure. On LinkedIn it's a tell, because a post should usually be one argument, and formatting five ideas as a list is what you do when you haven't decided which one you're making.

The fix: on LinkedIn, if you've written a list, ask which item you actually believe most and whether the other four are there because five felt like the right number.

6. Em-dash density

This series uses em dashes freely, so let's be straight about it: the dash isn't the tell. Several per paragraph, every paragraph, is — particularly the interrupting kind used to insert a qualifier into the middle of every sentence. I checked these eight posts against the rule below while writing them: of two hundred-odd paragraphs, none carries more than two.

The fix: count them. If a paragraph has more than two, some of them are full stops.

7. The three-beat structure

This is the one I catch first, and it's structural rather than verbal — which is why it survives a careful edit that removes every item above it.

Statement or question. Anecdote. Statement or question. Then the next section does it again, and the one after that. The rhythm is so consistent across generated writing that you can feel it before you've read a full sentence, and once you've noticed it you can't stop noticing it.

The fix: look at the shape of your paragraphs rather than the words in them. If every section opens the same way and turns at the same point, the problem isn't the sentences. Let one section be three lines and the next be twelve. Start one in the middle of an argument with no setup at all.

8. The empty superlative

“Game-changing.” “This is huge.” “A powerful new way to…”

Intensity substituted for specificity. It's a tell for the same reason it's bad writing: it tells the reader how to feel instead of giving them the thing that would produce the feeling.

The fix: replace the superlative with the specific fact underneath it. If there isn't one, that's information — the sentence was empty.

9. The everyone-agrees claim

“Quality matters.” “Customers value authenticity.” “The best teams communicate well.”

Statements that are true of every company in every market, which is exactly why a generator produces them: they are safe, and they're what's most abundantly available in anything trained on the public web.

The fix: the competitor test from the third post in this series. If your closest competitor could post the sentence verbatim without contradicting themselves, cut it.

10. Perfect competence

No aside. No hesitation. No admission that a bit of it is unresolved. Every argument tidily closed.

Real thinking has loose ends, and a founder writing about their own market should have at least one thing they're not sure about. The absence of any is itself the signal — and this one is hard to fix by editing, because it's a property of what you're willing to say rather than how you said it.

The fix: leave the bit you're uncertain about in, and mark it as uncertain.

11. The invented specific

This one is different from the other ten, and it's the one to be strictest about.

It happens when someone — or something — notices a claim is thin and reaches for a concrete detail to shore it up. “We saw a 40% lift.” “One client told us it saved them six hours a week.” A number that was never measured, a customer who didn't say that.

It's a tell, because invented specifics have a texture — suspiciously round, suspiciously well-suited to the sentence they're supporting. But that isn't the main problem. A fabricated detail is the tell plus a lie, published under your name, on a profile whose entire job is credibility.

The rule: when you don't have the fact, you have two options — go and get it, or cut the sentence. There is no third option. This applies with full force to anything you use to help you write: an assistant that invents a plausible statistic to make a paragraph land has done the worst possible thing to you, because it looks like the best possible thing.

Worth drawing a line here that gets blurred, because I've spent a career on the right side of it. Every marketer frames things favourably. Your competitor is strong in one area and you're strong in another, and the craft is making the case that your area is the one that matters more. That's positioning, it's legitimate, and anyone who tells you they don't do it is selling something.

Framing a real fact advantageously and inventing a fact that flatters you are different acts. The first is argument. The second is the tell, plus a lie. In twenty years nobody ever asked me to cross that line — which is worth saying, because the tools now available will cross it for you, cheerfully, without being asked.

The check that actually catches these

Read it aloud. Almost every one becomes obvious the moment you hear it, and most are invisible on screen. It takes ninety seconds. The three-beat structure is the exception — that one you have to look at rather than listen to.

Then one question: could anyone else in my industry have posted this, word for word? If yes, the tells are the least of the problem.

The limit of all this — and it's a real one

Removing all eleven makes your post stop reading like AI. It does not make your post contain anything only you could have said. Those are two different properties, and it's worth being honest that this checklist only delivers the first.

A post can be clean of all ten patterns and still be assembled entirely from public, well-known material — which is, more or less, the definition of what a generator produces. Polishing the surface doesn't change what's underneath.

The only cure for the second property is substance you actually have: something that happened to you this month, a number you genuinely measured, a thing a customer really said, a position you'd defend in an argument. Style suppression is a floor. Substance is the ceiling, and it can't be manufactured — which is precisely why the last item on that list is a hard rule rather than a preference.

All of which assumes you're publishing often enough for any of it to matter, which turns out to be the hardest part of the lot.