# AI Experts' LinkedIn Strategy Analysis

Miguel Perez Pellicer's agent wrote an 80-word LinkedIn post that got 41 reactions and 21 comments in six hours. A line-by-line teardown, what LinkedIn's engineers say they rank, and a skeleton to reuse.

By one detector's count, 4 in 5 long LinkedIn posts now read as written by AI ([Originality.ai, July 2026](https://originality.ai/blog/ai-content-published-linkedin)). Most of them sink. Miguel Perez Pellicer's agent wrote 80 words that got 41 reactions and 21 comments within six hours. I commented under it: "Handing this post to my agents for reverse engineering and automation, thanks for the exemplar." He replied: "Send me what your agents find. I'd read that teardown. P.S. Yes, an AI drafted and posted this :)"

So here's the teardown, Miguel. It covers what each line does, the research behind it, what LinkedIn's engineers have published about ranking, and a skeleton anyone can reuse. My agents read the 12 studies that a first draft cited, and checked every LinkedIn claim against LinkedIn's own pages.

## The post

> Nobody cares about your Claude certification.
>
> I've spent 12 years building software for small businesses and big corporations.
>
> Here's what I learned: only devs care about the tech stack.
> Business owners don't. Today it's this tool. Tomorrow it's that one.
>
> They care about the thing slowing them down.
> Usually it costs time. Often money. Sometimes both.
>
> That's what I build. AI agents and automations that kill it.
> If it doesn't save you time or money, you get your money back.

[Miguel's post on LinkedIn](https://www.linkedin.com/posts/miguelperezpellicer_nobody-cares-about-your-claude-certification-share-7527369399050301440-DL50/)

## What each line does

**"Nobody cares about your Claude certification."** A negative first line, aimed at a badge, not a person. Negative words pull clicks. Across 22,743 randomized headline tests on Upworthy, each extra negative word raised click-through by 2.3% ([Robertson et al., Nature Human Behaviour, 2023](https://doi.org/10.1038/s41562-023-01538-4)). Those were news headlines, but a first line on LinkedIn does a headline's job.

**"I've spent 12 years building software for small businesses and big corporations."** One exact number, and two kinds of client. It answers "who are you to say so?" before the reader asks.

**"Here's what I learned:"** The colon opens a gap. Curiosity is "a form of cognitively induced deprivation that arises from the perception of a gap in knowledge" ([Loewenstein, Psychological Bulletin, 1994](https://doi.org/10.1037/0033-2909.116.1.75)). The reader keeps going to close it.

**"only devs care about the tech stack. Business owners don't."** Now every reader has to pick a side. The badge holders from line 1 are on the wrong one.

**"Usually it costs time. Often money. Sometimes both."** Three beats. When the writer has something to sell, more claims help "until the fourth claim, at which point consumers' persuasion knowledge causes them to view all the claims with skepticism" ([Shu and Carlson, Journal of Marketing, 2014](https://doi.org/10.1509/jm.11.0504)). Three is the most a seller gets.

**"If it doesn't save you time or money, you get your money back."** A guarantee. In an economic model, a money-back guarantee signals quality because a weak product would be returned more often ([Moorthy and Srinivasan, Marketing Science, 1995](https://doi.org/10.1287/mksc.14.4.442)).

**All 80 words, short ones.** Needless long words make writers seem less intelligent. In Oppenheimer's experiments, complex texts got lower ratings of the author's intelligence ([Applied Cognitive Psychology, 2006](https://doi.org/10.1002/acp.1178)).

## What LinkedIn says it rewards

LinkedIn's engineers have published how the feed ranks posts. Here's what they say, in their own words.

1. **Time spent reading counts.** LinkedIn "starts measuring when at least half of a feed update is visible," and ranks down posts people scroll past fast ([LinkedIn Engineering, May 2020](https://www.linkedin.com/blog/engineering/feed/understanding-feed-dwell-time)). In 2024 it began treating a long read "as a positive signal," judged against similar posts, while "limiting 'click bait' or 'dwell bait' content" ([LinkedIn Engineering, October 2024](https://www.linkedin.com/blog/engineering/feed/leveraging-dwell-time-to-improve-member-experiences-on-the-linkedin-feed)).
2. **Advice beats virality.** "More reach when the content is focused on sharing knowledge and insights." And: "LinkedIn is not designed for virality" ([Alice Xiong, LinkedIn, June 2023](https://www.linkedin.com/pulse/how-were-surfacing-knowledge-linkedin-alice-xiong)). Comments from people in your niche are "a strong signal," but "not generic, one word comments" ([Social Media Today, June 2023](https://www.socialmediatoday.com/news/linkedin-new-insights-latest-algorithm-updates/654091/)).
3. **Pure selling and bait lose.** LinkedIn discourages posts that "solely attempt to sell products or services without offering useful insights", engagement bait like "reaction-vote polls, tagging for attention," and "dismissive, derisive, or unconstructive conversations" ([Search Engine Journal, November 2024](https://www.searchenginejournal.com/linkedin-shares-dos-donts-for-better-post-visibility/533571/)).
4. **The feed is now ranked by an LLM.** In March 2026 LinkedIn moved to a ranking system "powered by LLMs and GPUs." Each post's prompt carries its engagement as percentiles, and "when a post is published or gains engagement, we update its item prompt" ([LinkedIn Engineering, March 2026](https://www.linkedin.com/blog/engineering/feed/engineering-the-next-generation-of-linkedins-feed)).

Read Miguel's post against that list. It's a sales post in the shape of a lesson ("Here's what I learned"), which is LinkedIn's own fix for a pure pitch. It asks for nothing, so it isn't bait. Its sting lands on a certificate, not a person, so it isn't derisive. It's short, so anyone who starts it finishes it.

## What we're testing next

Some of what makes this post work is still a hunch. My guess is that the first line pulls the badge holders into the comments while the business owners read quietly to the end, and LinkedIn counts both. Nobody has published a test of that. It's my next experiment: tag a post's commenters by role and see who answers a sting.

## The skeleton

> Nobody cares about your [BADGE YOUR PEERS ARE PROUD OF].
>
> I've [DONE THE WORK] for [BUYER A] and [BUYER B]. [One exact, true number.]
>
> Here's what I learned: only [PEERS] care about [THE BADGE].
> [BUYERS] don't. Today it's this [TOOL]. Tomorrow it's that one.
>
> They care about [THE THING SLOWING THEM DOWN, in their words].
> Usually it costs [COST]. Often [COST]. Sometimes both.
>
> That's what I build. [WHAT YOU BUILD] that [FIXES IT].
> [A guarantee, or one measured before-and-after.]

Three rules keep it inside what LinkedIn has published. Sting a badge, never a person. Ask for nothing. Keep the number true.

## The skeleton with my facts

> Nobody cares how many agents you can run.
>
> I bootstrapped GPTBoss to 60,000 signups and $240k in revenue with $0 spent on ads.
>
> Here's what I learned: only builders care about agent counts.
> Business owners don't. Today it's this model. Tomorrow it's that one.
>
> They care about the job that eats their week.
> Usually it costs time. Often money. Sometimes both.
>
> That's what I build. AI that takes that job off their hands.
> And I measure it. When I rebuilt BuildCanada.com, pages went from 20 seconds to load to under 2.

An agent can fill in every slot but one. The true number has to come from work you did.

Miguel, your agent did fine. I'm keeping the skeleton. If you're an owner with a job that eats your week, [tell me which one](/contact).

mackenzie@mackenziebowes.com or @_mackenziebowes.

---

Published 2026-11-14. Web page: https://www.mackenziebowes.com/posts/ai-experts-linkedin-strategy-analysis
