linkedin-post-writer

LinkedIn Post Writer

Overview

Drafts long-form LinkedIn posts using 16 hook formulas that were reverse-engineered from posts that outperformed their authors' baselines in 2025-2026, each with a reference engagement number. Instead of asking "what should I write", the workflow asks "what should this post earn" (comments, reposts, likes, or saves), shortlists 2-3 matching formulas, fills the chosen skeleton with the user's voice, then scrubs the draft for AI tells before it ships.

This is the flagship skill from sergebulaev/linkedin-skills, a 10-skill LinkedIn bundle (writer, humanizer, pre-publish audit, comment drafter, reply handler, hook extractor, content planner, profile optimizer, engager analytics, thread monitor) installable as a Claude Code or Codex plugin. This standalone version covers the drafting workflow; scheduling and publishing automation live in the full bundle.

When to Use This Skill

• Use when the user says "write me a LinkedIn post about X"
• Use when the user has a topic and a rough angle but needs a hook and structure
• Use when the user wants to pick from illustrative post formats instead of improvising
• Use when a draft exists but the hook is weak and needs a formula-based rebuild
• Not for replying to comments or optimizing profiles; this skill only drafts posts

How It Works

Step 1: Gather inputs

Collect: topic, angle, target audience (founders, operators, marketers), desired length (short 300-500, medium 900-1,300, or long 1,500-1,900 characters), and any raw material the user already has (numbers, anecdotes, names).

Step 2: Pick the formula by engagement goal first

Ask (or infer) what the post should earn, then shortlist:

The full set of 16, with reference engagement:

Important caveat: F1-F10 references are engagement counts or format multipliers against the author's own baseline; F11-F16 references are raw corpus reach, often inflated by a famous author or a reshare. The two groups measure different things, so never rank formulas across groups by number.

The full skeletons for all 16 formulas are bundled with this skill in references/hook-formulas.md, along with per-formula warnings and the "Never do" list.

Step 3: Draft the post

Read the chosen formula's skeleton from the bundled references/hook-formulas.md and fill it with the user's voice and material, using these optional formatting conventions:

• Place the hook early; the see-more fold varies by device, layout and post type.
• 900-1,300 characters is one drafting target; choose length from the audience and content.
• Double line-breaks between ideas, never walls of text
• 0-2 hashtags, placed at the end
• Include source links where useful; no blanket algorithm penalty or first-comment advantage is established here.
• Prefer "How I" over "How to": first-person experience is appropriate only when the user actually had that experience
• Put one specific number in the first sentence ($873.47 beats $900; 40,000 beats "a lot")

Step 4: Scrub AI tells

Before showing the draft: strip em dashes, AI vocabulary ("game-changer", "deep dive", "delve"), rule-of-three lists without receipts, and generic openers like "In today's fast-paced world". Add human fingerprints: specific numbers, names and personal details only when supplied or verified; never fill a quota by inventing them. Vary sentence length aggressively (mix 3-word and 25-word sentences).

Step 5: Present the result

Show the user: formula used, full draft, character count, and a suggested posting window only when audience-specific analytics/timezone justify it; otherwise propose an explicitly untested slot.

Examples

Example 1: Founder build-log (F7 Odd-Precision Money Ledger)

User:  write a post about what my bootstrapped SaaS actually costs to run
Agent: Goal check: this reads like a saves play (people screenshot cost
       breakdowns). Best fit is F7 Odd-Precision Money Ledger (9.4x baseline).
       Draft hook:

       "My SaaS costs $412.83 a month to run. Here is every line item."

       Then the ledger: hosting $94.20, email $35.00, monitoring $18.63...
       exact numbers only, totals that actually add up, one lesson at the end.

Example 2: Goal-first selection

User:  I want a post that gets real comments, topic: why I stopped doing demos
Agent: Comment-earners are F4 Confession, F10 Contrarian, F12 Permission Slip.
       For "I stopped doing X" the strongest is F10 Contrarian + Receipts:
       open with the unpopular claim, back it with 2 historical parallels,
       close with a question that invites a substantive response. Reference: 3,083 eng.

Best Practices

• ✅ Pick the formula by engagement goal first, topic second
• ✅ Lead with a real failure or a specific number in the first 3 lines
• ✅ Include one moment of genuine vulnerability or concrete stakes; do not infer reach from narrative form alone
• ❌ Don't blend two hook formulas in one post; it dilutes both
• ❌ Don't use F5 Self-Proving Meta unless the user will actually keep the promise
• ❌ Don't pair F7 Money Ledger with rounded or invented numbers; readers notice
• ❌ Don't open with an all-caps line ("THIS CHANGED EVERYTHING")
• ❌ Don't frame LinkedIn as inferior inside a LinkedIn post

Limitations

• Reference engagement numbers describe the 2025-2026 corpus the formulas were extracted from; they are priors, not guarantees, and LinkedIn's ranking changes over time.
• The skill drafts text posts; it does not generate images, carousels, or video scripts.
• This standalone version does not schedule or publish. Scheduling, comment drafting, reply handling, and engagement analytics require the full bundle from the source repo.
• Voice quality depends on the raw material the user provides; a formula cannot invent authentic anecdotes, and the skill should ask for real details rather than fabricate them.

Common Pitfalls

• Problem: The draft sounds like every other AI-written LinkedIn post. Solution: Run Step 4 ruthlessly. Cut em dashes, cut "game-changer" vocabulary, and force one concrete first-person detail per 100 words.
• Problem: The hook is buried in paragraph two. Solution: The first 210 characters must carry the hook; everything before the fold decides the expand rate.
• Problem: Comparing F11's raw reach to F8's 19.64x multiplier and picking F11 "because the number is bigger". Solution: The columns measure different things. Match formula to goal and topic, not to the largest number.
• Problem: Post gets reach but zero comments. Solution: The formula was picked for the wrong goal. Comment-earners end with a question or a side-picking claim, not a summary.

Related Skills

• @linkedin-content-generator - broader LinkedIn content suite (carousels, newsletters, calendars)
• @linkedin-profile-optimizer - profile and authority optimization rather than post drafting
• @social-post-writer-seo - multi-platform social copy when LinkedIn is not the only target

Additional Resources

• Source repo with all 16 formula skeletons and worked examples
• Full 10-skill bundle install (Claude Code / Codex plugin)

Evidence and publication boundaries

The corpus, baselines, dates and metrics are not bundled or independently auditable. Treat all reference counts, multipliers, algorithm claims and formula-to-goal mappings as upstream anecdotes, not predictions or causal evidence. Saves cannot be inferred from public repost counts. Do not fabricate precise costs, failures, quotations, named endorsements, personal stories or scarcity to satisfy a skeleton. Avoid deceptive bait-and-switch and comment gates; offer useful direct access and an honest CTA. Respect confidentiality/consent when naming people and source factual claims. Drafting never authorizes publishing, tagging people, commenting or DMs. No post was published and no analytics were measured in this review.