| name | create-article |
| description | Creates a LinkedIn Article (long-form Pulse, 1000-2000 words) optimized simultaneously for AI-search (GEO) and the LinkedIn algorithm. Use when long-form content is needed with the goal of LLM citations. 8-step pipeline with a mandatory GEO checklist. |
Create LinkedIn Article
An Article is long-form pillar content (1,000–2,000 words) optimized for AI-search engines (ChatGPT, Perplexity, Google AI Overview) + the LinkedIn algorithm. Unlike a post, it has a permanent URL and gets indexed for LLM retrieval. One Article per week + 2–3 posts is the standard cadence.
When to invoke
- The user says "write an Article", "write long-form", "do a Pulse"
- The topic is "big" — won't fit in a 1,200–1,800-character post, needs a framework / multiple H2s
- The topic is phrased as a question the audience would ask ChatGPT (classic GEO signal)
- You want long-tail discovery through LLM engines, not a one-shot engagement spike
When NOT to invoke
- Topic is short, single insight =
/create-post (LinkedIn post)
- Personal story / reflection without an actionable framework =
/create-post
- Topic doesn't match the author's headline (Profile-Content Alignment gates reach)
- Author has already had this topic in an Article in the last 3 months — pick a different angle
Pipeline (8 steps)
Step 1: Load context
Read the mandatory items:
Knowledge-Base/01-Identity-Profiles/[Author]-DIP.md — voice, expertise clusters, viral angles
Knowledge-Base/10-Articles-and-GEO/LinkedIn-Articles-GEO-Framework.md — 6 GEO principles
Knowledge-Base/02-LinkedIn-Algorithm/Algorithm-Intelligence.md — LinkedIn algorithm
Articles/Articles-Plan-Template.md — structural template
Hook-Bank.md — for choosing the hook angle
Optional:
Knowledge-Base/01-Identity-Profiles/_Voice-Vocabulary/[Author]-Voice-Vocabulary.md (if it exists)
- Recent posts from
Posted/[Author]/ (last 5) — to avoid repetition
- Recent Articles from
Articles/Posted/[Author]/ (if any) — to avoid repetition
Step 2: Phrase the title as an LLM query
The title must be the exact phrasing of the question a user would ask ChatGPT/Perplexity.
Test: type the title into Google. "People also ask" should surface a similar question.
Pattern examples:
- "How [does/do] X [verb] Y?" → "How AI Agents Manage Ad Budgets"
- "When to [action]: [framework]" → "When to Kill Underperforming Ads"
- "What Is [concept]? [Numbered framework]" → "What Is an AI-Native Company? 12 Steps"
- "Why [observation about industry] [+ counter-take]" → "Why Your CRM Is Missing 30–40% of Lead Sources"
Step 3: Write the Lead Paragraph (direct answer)
The first 2–3 sentences = DIRECT ANSWER to the question in the title.
❌ "Let me set up the context. In today's marketing world..."
✅ "AI agents manage ad budgets by monitoring campaigns in real time, detecting anomalies before metrics shift, and either alerting humans or auto-adjusting within guardrails. In practice at $20M+ scale, this looks like..."
This is the most important paragraph of the article. The LLM will pull it as the summary.
Step 4: Design 4–6 H2 sections
Structure pattern:
H2: Why this matters / The problem
H2: What [your concept] actually does / The framework
H2: Real results / Real cases (with concrete numbers)
H2: How to start (even without [your product])
H2: Where most teams get stuck (optional)
Each H2 must work standalone (an LLM can extract one).
Step 5: Create ONE unique concept with its own name
Every Article should have 1 named concept your audience can find.
Examples from practice:
- "Trust adoption curve" — chat → autonomous progression
- "5 levels of AI ad management" — numbered scale
- "The 30–40% CRM attribution gap" — statistic-as-phrasing
Test: is the phrasing unique? If others say the same — you need a new angle.
Step 6: Pack with concrete numbers
Minimum 5 fact-rich points in the article:
- ❌ "AI saves time" → ✅ "Analysis that took 90 minutes now takes 10"
- ❌ "Most leads are tracked" → ✅ "30–40% of leads hit CRM with no source data"
- ❌ "We test creatives" → ✅ "We analyzed 1,180 creatives — optimal kill threshold isn't $5 or $100"
Numbers must come with context: what, at what scale, compared to what.
Step 7: Internal GEO checklist
Before publication, walk the checklist:
Step 8: Save + automatic validation
Save to Articles/Posted/[Author]/YYYY-MM-DD-[slug].md with frontmatter:
---
title: "[Title]"
author: [Author]
published_date: YYYY-MM-DD
publication: LinkedIn Pulse
direct_url: TBD
companion_post: TBD
named_concept: "[your unique concept name]"
geo_target_queries:
- "[query 1 — branded]"
- "[query 2 — topic]"
- "[query 3 — mid-funnel]"
hook_bank_refs:
- "#N"
sources:
- "[transcript / podcast / call from which content derived]"
status: draft / published / audited
---
Then invoke /validate-content [path] --type article for:
- check-facts
- check-generic (no AI markers)
- check-sources (each claim → source)
- detect-ai
If everything is clean → ready for publication.
Output
A file at Articles/Posted/[Author]/YYYY-MM-DD-[slug].md + alternative leads (3 variants) + timing recommendation + companion post draft (1,200–1,500 characters, teaser hook + link to the Article).
Worked examples
See Articles/_Examples/ in this repo — 5 published Articles that went through the full pipeline. Study them before running the skill for the first time.
Post-publication
After 4 weeks — invoke /audit-llm-visibility [article-path] to check:
- Direct URL discoverability
- Content citation in AI summaries
- Brand share-of-voice on ICP queries
Results — in Articles/_Audits/YYYY-MM-DD-[slug]-audit.md.
Anti-patterns
| Anti-pattern | Why it's bad |
|---|
| Clickbait title without substance | LLMs skip clickbait |
| Lead paragraph as intro-anecdote | Wastes the main slot |
| 2 articles per week from one author | LinkedIn cuts distribution (40% penalty) |
| Marketing CTA in conclusion | Lowers LLM-trust, citation rate falls |
| Topic doesn't match author headline | Profile-Content Alignment gates reach |
| Article = rewritten post | Wastes the Article cadence |