How to Tell a LinkedIn Message Was Written by AI

Why This Matters More Than People Think

Most people treating this as a complaint are missing the point. If 81.2% of long-form LinkedIn posts were flagged as likely AI-generated in a July 2026 study by Originality.ai across 5,000 posts, that means the people doing human outreach are standing in an almost empty room. That's not a problem. That's a position.

I run LinkedIn appointment setting for small businesses through Veteran Vectors. I've read thousands of cold messages, sent by clients, sent to clients, sent to me. The AI-written ones aren't hard to spot. They follow a handful of patterns so consistently that I can usually identify one in the first two sentences. Here's what to look for.

The Six Concrete Tells

1. Front-Loaded Generic Flattery

The message opens with a compliment that could apply to any account with a profile photo and a headline. Phrases like "I've been following your work for a while and I'm really impressed by what you're building" or "Your journey in the [industry] space is truly inspiring." There's no specific detail. Nothing that proves the sender read anything you actually wrote.

A real person who did their homework leads with the specific thing they noticed. "Saw your post last Tuesday about losing a client over billing delays, that hit close to home." That's specific. AI doesn't do specific without being told exactly what to say.

2. The Negation Triad

This one's almost a signature. The message builds to a pitch using a three-part negation: "This isn't just about [X]. It's not about [Y]. It's about [Z]." Sometimes it's two negations, sometimes four, but the structure is the same. It's a rhetorical device that sounds punchy in writing prompts and sounds robotic in a DM.

If you see "not just... not just... but truly," stop reading. You're holding AI output.

3. The Contrast Reframe

Somewhere in the middle of the message, there's a pivot sentence that introduces a reframe: "Most [job titles] focus on [common thing]. But the best ones know that [insight the AI just invented]." It sounds like wisdom. It's a pattern. The contrast is manufactured to make the sender seem thoughtful. It doesn't come from observation, it comes from a prompt that said "sound authoritative."

4. Over-Explaining the Value Prop

Human cold messages are usually short. The sender knows their time is limited and so is yours. AI messages explain the entire product, anticipate every objection, and close with three CTAs. They're long because the person who wrote the prompt said "be thorough" and the model obliged.

If a cold message to a stranger reads like a landing page, it wasn't written by a person who respects your time.

5. Uniform Sentence Length

Read the message out loud. If every sentence takes roughly the same amount of time to say, it's AI. Human writers vary their rhythm. Short punch. Then a longer sentence that carries the weight of an idea and lands somewhere unexpected. Then short again. AI writes in metronomic paragraphs where everything flows at the same cadence because that's what training on "good writing" produces when nobody pushes back on it.

6. The Hedge Cluster

Phrases like "I thought it might be worth reaching out," "I wasn't sure if this was relevant," and "feel free to ignore this if it's not a fit" tend to cluster together near the close. A human uses one hedge. AI uses all of them, because the prompt said "don't be pushy" and the model doesn't know when to stop.

Before and After: The Same Pitch, Two Ways

This example is made up. Imagine you sell fractional CFO services to e-commerce founders.

AI version:

"Hi [Name], I've been following your incredible work building [Company] and I'm truly inspired by what you've created. This isn't just about financial reporting. It's not about compliance. It's about building the financial foundation that separates the brands that scale from the ones that plateau. Most founders focus on revenue, but the best ones know that margin clarity is what actually drives growth. I'd love to explore whether there might be a fit. Feel free to ignore this if it's not relevant."

Human version:

"Saw your post about Q1 inventory write-downs. That's a rough quarter to absorb. I work with a handful of e-comm founders on exactly that, cash flow modeling when SKU count is outrunning margin visibility. Worth a 20-minute call to see if it's relevant?"

The human version is shorter, specific, and makes one ask. The AI version is longer, generic, and makes three asks while apologizing for all of them.

The Numbers Are Getting Worse

Originality.ai's July 2026 analysis of 5,000 long-form LinkedIn posts found 81.2% were likely AI-generated. That's long-form posts, the kind people spend time on. Cold messages take less effort to produce, so I'd expect the share there to be higher, though nobody has published that number.

LinkedIn responded. They shipped a "Seems like AI slop" report button, and over one million users hit it within two weeks of launch. The platform is building infrastructure to surface this problem because enough users got loud enough about it that ignoring it became a product risk.

The signal is clear: the volume of AI outreach is so high that the people receiving it are actively asking for tools to reject it.

How to Check a Message You're Not Sure About

1. Read the opening sentence. Is the compliment specific to something you actually published, or is it generic praise for your "incredible work"? Generic is AI.
2. Look for the triad. Find any "not just... not just... but" construction. One hit is a strong signal.
3. Count the CTAs. One CTA at the end is human. Two or three, or a CTA plus an apology for the CTA, is AI.
4. Check sentence rhythm. Read it aloud. Metronomic cadence with no variation is a tell.
5. Search for hedge clusters. More than one softener near the close is a flag.

Why This Is a Buying Signal, Not Just a Complaint

Here's what I want you to sit with. If most of what gets posted on LinkedIn reads like it was typed by a tired language model, and a million people are already using a report button to flag it, then the bar for standing out isn't high. It's just above the floor.

A message that references something specific the person wrote. A message with a single clear ask. A message that sounds like it came from a person who read their profile for two minutes before writing. That message is now unusual. That's where the replies are.

I've built Veteran Vectors' LinkedIn appointment setting service around exactly this. Every outreach script I write is specific to the person receiving it. The list is built by hand for the client's actual ICP. The follow-up is timed and written like a human continuation of a real conversation, not a drip sequence that announces itself.

The AI wave didn't make LinkedIn harder to convert. It made it easier, for the people willing to do it right. If you're a small business owner who wants booked sales calls and you're tired of watching your inbox fill with the same AI patterns you just read about, that's the conversation I want to have.

The back-office side of what we do at Veteran Vectors, automating follow-up, building custom CRMs, connecting the tools you already have in something like n8n, Make, or Airtable, that's a separate service with its own starting point: a paid diagnostic that gets credited in full to the build. Both services exist because small businesses deserve systems that actually work, not just activity for activity's sake.

But start here: the next cold message you get, run it through the six tells. You'll know within thirty seconds whether a person sent it or a prompt did. And then ask yourself which one you're sending.

Frequently Asked Questions

Want this done for you? See how we run LinkedIn outreach, or book a call below.