A few weeks ago, I got curious about something.
I'd been scrolling LinkedIn — as one does when procrastinating — and I kept having this weird sensation. Post after post felt... familiar. Not the topics. The texture. Like they were all written by the same person wearing different profile pictures.
So I did what any reasonable person would do: I turned it into a research project.
I collected 500 LinkedIn posts from solopreneurs, founders, consultants, and creator-types. All published in the last 90 days. All from people who clearly use AI tools (you can tell — more on that in a moment). I tagged every post for structure, vocabulary, formatting patterns, and engagement metrics.
What I found was both unsurprising and deeply alarming.
The Methodology (Keep Me Honest)
Before you ask: no, this wasn't a rigorous academic study. I'm not publishing a paper. Here's what I actually did:
- Collected 500 posts from LinkedIn feeds across niches: marketing, SaaS, coaching, dev tools, freelancing
- Filtered for posts that showed strong AI-generation signals (I'll explain the signals below)
- Categorized each post by structure type, opening line pattern, vocabulary, formatting, and engagement (likes + comments)
- Compared against a control set of 100 posts from creators known for distinctive human voices
It's directional data, not gospel. But the patterns were so consistent that the exact numbers almost don't matter. You'll recognize every single one.
Finding #1: 82% Use the Same 3 Opening Structures
Out of 500 posts, 411 opened with one of these three patterns:
Pattern A: The Contrarian Hook (38%)
"Most people think [common belief]. They're wrong. Here's why."
Pattern B: The Humble Brag Confession (27%)
"I [impressive achievement]. But here's what nobody tells you about [topic]."
Pattern C: The Single-Line Shock (17%)
"[Bold claim or surprising statistic]." (new paragraph) "Let me explain."
That's it. Three templates. Eighty-two percent of posts.
The remaining 18%? Those were the posts getting 3-5x more comments. Not because they had better information — because they opened in ways that didn't trigger the reader's "oh, another one of these" reflex.
One founder opened with: "My therapist told me to stop posting on LinkedIn. She might be right." That post had 847 comments. The AI-template posts around it averaged 12.
Finding #2: The "Line Break Every Sentence" Epidemic
This was the single most consistent pattern across the entire dataset.
91% of the AI-generated posts used the same formatting trick: one sentence per line, with a blank line between each sentence. Like this:
This is the first point.
This is the second point.
This is the third point.
Every single line gets its own paragraph.
Because apparently that's "engaging."
You've seen this a thousand times. And two years ago, it actually worked — it was a legitimate LinkedIn formatting hack for readability on mobile. But when every AI tool started doing it by default, it became a tell. It's the writing equivalent of that one Instagram filter everyone used in 2014.
The control group posts? They varied their paragraph length. Some had dense three-sentence paragraphs followed by a punchy one-liner. Some used actual paragraph structure. The rhythm was irregular — which is exactly what made it feel human.
Finding #3: 73% Contain "Permission Words" That Nobody Actually Uses
I built a vocabulary list from the 500 posts and compared word frequency against normal professional writing. Some words appeared at 10-40x their natural frequency:
| Word/Phrase | Frequency vs. Normal | AI Tell Rating |
|---|---|---|
| "Here's the thing" | 34x | Extreme |
| "Let that sink in" | 28x | Extreme |
| "Read that again" | 22x | High |
| "Game-changer" | 19x | High |
| "It's not about X, it's about Y" | 17x | High |
| "And honestly?" | 15x | Medium |
| "Full stop." | 14x | Medium |
| "The truth is" | 12x | Medium |
"Here's the thing" appeared in 170 out of 500 posts. One in three. In normal professional writing, that phrase shows up maybe once every 50 articles.
These are what I call "permission words" — phrases that exist to make the writer sound confident and conversational without actually adding meaning. AI tools scatter them everywhere because they pattern-matched on viral posts from 2022-2023 where these phrases were genuinely novel. Now they're noise.
The most damaging one? "Read that again." It showed up after mundane observations like "Your network is your net worth. Read that again." Sir, I read it the first time. It wasn't that profound.
Finding #4: The Listicle-Story Hybrid Has Become a Monoculture
I categorized each post by structure. The breakdown:
- Numbered tips/lessons ("7 lessons I learned from..."): 34%
- Story → lesson → CTA: 29%
- Contrarian take → evidence → reframe: 21%
- Question → answer → engagement bait: 11%
- Something genuinely different: 5%
That last 5% is doing all the heavy lifting. Those posts had original structural ideas: a post written as a fake job description, a post formatted as a conversation transcript, a post that was just one long paragraph with no line breaks at all (it went viral specifically because it broke every "rule").
Here's what's happening: AI tools have been trained on LinkedIn's greatest hits. The structures that worked in 2023 are now the default output of every AI writing tool. Which means they're also the default structure of every post on the platform. Which means they no longer work — because distinctiveness was the entire reason they worked in the first place.
It's a content ouroboros. The snake eating its own tail.
Finding #5: Engagement Is Inversely Correlated with "AI Polish"
This was the finding that surprised me most.
I scored each post on an "AI polish" scale of 1-10, based on how many of the above patterns it exhibited. Then I plotted that against engagement (normalized for follower count).
The correlation was negative. The more "polished" a post looked by AI-template standards, the less engagement it received.
Posts scoring 8-10 on the AI polish scale averaged 0.4% engagement rate. Posts scoring 1-3 averaged 2.1% — more than 5x higher.
The highest-engagement post in my entire dataset was riddled with what a writing teacher would call "errors." It started mid-thought. It had a paragraph that was just "Anyway." It ended without a neat conclusion. It felt like someone thinking out loud. It had 2,300 comments.
The implication is uncomfortable but clear: the things AI tools optimize for — structure, clarity, polish — are actively working against the thing that drives engagement — distinctiveness.
Why This Is Actually a Huge Opportunity
I know this reads like a doom-and-gloom piece, but I actually think the opposite is true.
If 95% of content on a platform converges on the same patterns, then being different has never been easier. You don't need to be a better writer. You just need to not sound like the default output of the tools everyone else is using.
Some concrete things that worked in the control group:
1. Write your opening line last. Most AI tools front-load the hook. If you write your post first and then figure out how to open it, you'll naturally avoid template hooks because you're working from your actual point, not a formula.
2. Leave in one "imperfection." A sentence fragment. A tangent. A parenthetical that goes on too long. These are voice signals. They tell the reader's brain: a human made choices here.
3. Use vocabulary from your actual life. The control group creators referenced specific tools by name, used industry jargon naturally (not as keywords), made jokes that only their niche would get. AI writes for everyone. You should write for someone.
4. Vary your structure between posts. If your last post was a numbered list, make the next one a narrative. If you always open with a story, try opening with a question. Pattern-breaking across posts is just as important as within them.
5. Map your actual voice — then protect it. This is the hard one. Most people don't know what their writing voice actually is until it's gone. They can't describe their sentence patterns, their go-to transitions, the rhythm that makes their writing theirs.
This last point is what led me to start building VoiceForge. The core idea: feed it your best existing content, and it extracts your "Writing DNA" — the measurable patterns and choices that make your writing distinctly yours. Then when you use AI to create content, it generates from your voice profile, not from the same default register producing a million identical LinkedIn posts every day.
We're in early access now. If you're tired of sounding like everyone else's ChatGPT — or you've noticed the engagement decline I described above — check it out.
The Bottom Line
500 posts. 82% identical openers. 91% same formatting. 73% same vocabulary. And engagement that drops as AI-polish increases.
The content sameness problem isn't coming. It's already here. And it's creating a massive opening for anyone willing to sound like themselves.
The question is whether you'll notice the drift in your own content before your audience does.
I'm curious — have you noticed the LinkedIn sameness effect in your own feed? And more importantly, have you caught yourself falling into any of these patterns? Would love to hear your experience in the comments.
Top comments (0)