---
title: "Maximizing AI for LinkedIn: Avoiding Common Pitfalls"
url: https://stacklist.com/card/c6ba5f83-b1a2-45cb-b77e-5e798b47a66e
source_url: "https://x.com/paolo_scales/status/2082330528506675418?s=12"
stack: https://stacklist.com/c/business/stack/7c5a8832-3368-4e35-961e-f9273bcef8cc
summary: "Paolo Scales reveals why AI-generated LinkedIn content gets suppressed by the algorithm and provides a systematic framework to fix it. The solution involves three critical inputs (skill file, proof assets, customer voice data) and a five-step sprint with an anti-slop editing pass that catches structural AI patterns most founders skip."
tags: "ai-content, linkedin-strategy, algorithm-optimization, prompt-engineering, content-creation, anti-slop, founder-growth"
key_entities: "Paolo Scales (person), Paolo Trivellato (person), Claude (technology), Sonnet 5 (technology), LinkedIn (organization), LinkedIn Algorithm (concept), AI Slop Detection (concept), ICP Voice-of-Customer Data (concept)"
classification: "framework"
content_hash: "sha256:1000d3224a329394b789b97844dc35dd5945c63a80af54de4ef0888510d94e1f"
acp_version: "0.2"
token_counts_approximate: 955
visibility: public
agent_accessible: true
status: "final"
---

# Maximizing AI for LinkedIn: Avoiding Common Pitfalls

paolo trivellato @paolo_scales everyone using AI for linkedin is getting buried right now and they don't know why. i post 30 AI-written pieces a month and the algorithm pushes every one to cold audiences. the difference is one step they're all skipping. most founders who try AI for linkedin get slop. generic hooks. corporate body copy. every other line starting with the same word. linkedin's 2026 algorithm detects that instantly and buries it. the problem was the input and not the actual AI feeding claude a bare topic and expecting founder-quality content is like handing a session musician a guitar and expecting hendrix before they've heard a single recording. the model needs three things before it writes a word worth posting. first, a skill file. not a one-line instruction. a full operating system. the platform rules (dwell time over engagement, the "see more" gate, AI slop suppression), the writing constraints (lowercase, short paragraphs, escalation over repetition), the banned constructions, the funnel split, the CTA template. 1,500 words loaded once into the project. governs everything automatically after that. second, proof assets. every case study, every revenue number, every client outcome, pasted in at the start. the model needs specific proof to pull from, because the biggest AI tell is a vague claim backed by nothing. "we helped a client grow" is slop. "$89,000 in a single month at 93% margins" is proof. third, ICP voice-of-customer data. the exact words prospects use on sales calls to describe their problem. "i've been posting for six months and haven't gotten a single call." when the reader sees their own words in your hook, the recognition trigger fires immediately. AI with the right inputs writes hooks that feel human because the language already is human. it came from a real conversation. the sprint runs in five steps. minutes 0-5. generate 20 hook variations per post across 7 posts. 140 hooks in 90 seconds. each under 280 characters, each with a curiosity gap that doesn't resolve before "see more." minutes 5-10. score every hook on curiosity, clarity, novelty, and click-through. pick the winners. minutes 10-25. generate the full body for all 7 posts in one continuous run. sonnet 5 holds quality from post 1 to post 7 because of how it sustains focus. older models degraded by post 4. minutes 25-30. the anti-slop pass. a final prompt audits every post for parallel repetition, banned openers, monotone rhythm, rhetorical transitions, and vague claims. it rewrites every violation on the escalation principle: each line adds new information, raises the stakes, or deepens the implication. no echoing. that pass catches 60-70% of residual AI patterns. the last 30% gets caught in a 5-minute manual read. the step most founders skip is the edit. and the edit is what separates content the algorithm pushes from content it suppresses. a post that "sounds AI" doesn't have a writing quality problem. it has a structural pattern problem. three sentences of the same length in a row. two lines opening with the same word. a rhetorical question bridging two sections. those are the tells, and the anti-slop prompt catches them systematically. the whole system runs $2-10 a week in API tokens and produces 7 posts that would take 15-20 hours to write by hand. the tradeoff was never quality. it's time. the founders who win back those 15-20 hours a week put them where the business actually needs a human. that's the entire system. skill file, proof, real customer language, then a sprint with an edit pass that most people are too lazy to run. the ones who run it stop sounding like a robot and start pulling calls. 5:01 AM · Jul 29, 2026 2.4K Views 2 4 34 57