---
name: linkedin-writing-voice
description: How to write LinkedIn posts that sound like me, sentence by sentence. Load this whenever drafting or rewriting the words of one of my posts. It governs craft only: openings, rhythm, vocabulary, how I tell a story, punctuation, how I close. What to post about lives in the short section at the bottom.
---

# LinkedIn Writing Voice

*Julie Irving's skill, given away as-is. Swap my examples for yours (the opener bank, the words, the sign-off, the lanes, the facts). Don't try to sound like me. Sounding like me only works for me.*

*Housekeeping: this skill grows every time you add a lesson, and every word gets read before every draft. Every few weeks, ask the AI to revise it to be more efficient without dropping a rule. Mine hit 10,000 words before I did that.*

---

## The one test

Read the draft out loud. Would someone who has read my last ten posts recognize this as mine before they saw my name? If it could belong to any competent person on LinkedIn, it isn't done. Aim for specific, not for polish.

## The shape of a post

Posts are not paragraphs. They're a stack of short beats separated by white space, most one to three sentences, several a single line. The rhythm is the format. Think in beats.

1. A cold open that lands in the first line. No setup, no "I've been thinking about." The most surprising or costly line in the post is line one.
2. A short reveal or reframe right under it, often a fragment.
3. The story or the mechanism, told plainly, grounded in a named place or moment.
4. The turn. What I make of it, usually reasoning out loud.
5. A clean close. A specific question or a committed statement.

Not every post has all five. The spine is always: punch, reveal, grounded middle, my take, a close that commits.

A joke seasons the post; it doesn't organize it. If I hand you a funny aside, drop it once near the top and let the real story be the spine.

## Openings

The opening is the whole game. Write it harder than anything else.

**The rules:**

- Put the single most surprising, costly, or human line first. If the best line is in paragraph three, move it to line one.
- Front-load the answer. If the post has one headline finding, spend it in the first two lines. The rest confirms the promise instead of building to a reveal.
- Let the second line do the twist. My openers are often two short sentences where the second flips the first.
- A true, specific reason beats a manufactured curiosity gap. If there's an open loop, resolve it fast ("...because I'm getting a bit of imposter syndrome"). Honesty wins the tie.
- Give it room. Short openers underperform, not because of length but because a short line rarely carries a scene. Write the scene.
- For a teaching post, lead with a pattern I've observed out in the world, not with my own move or my own stat. My proof goes in the body, two or three beats down. Exceptions: a milestone post can open on the announcement, and a personal-journey post opens on my own motive and decision.
- A confession can be the hook even when it looks bad on first read, as long as the reveal reframes it a few lines later.
- If my brain dump contains a question I asked myself, that question closes the hook. It invites the reader in; a defensive fact pushes them out.
- Never open defensively. No "before you ask," no "this isn't about X." If the objection doesn't matter, wave it off ("who really cares?") and move on. If it does, state it as a finding, not a rebuttal.
- Don't announce the topic. Don't open with "excited to," "thrilled to," or "happy to share." Don't slip into third person about yourself or your product.

**My real openers, as a reference bank:**

- "Okay, I lied."
- "R.I.P. to applying to 200 jobs and hearing nothing back."
- "I wasn't using LinkedIn the right way. You probably aren't either."
- "A recruiter messaged me about a Head of Marketing role. By the time she followed up, the job had been replaced by AI."
- "More than 30% of people on GLP-1s quit within the first four weeks."
- "Well, I guess it was finally my turn."
- "I stopped waiting to get hired. So I hired myself."
- "There's no reason to spend $10,000 on a shirt."
- "My kids are on two health plans this month. One covers their pediatrician, one doesn't. I don't get to pick which one to use. My husband's birthday does."

## Sentence mechanics and rhythm

The rhythm is choppy and uneven on purpose. It reads like talking, thinking, correcting yourself. Smoothing it into even 25-word sentences is the fastest way to sound like AI.

**Do:**

- Vary length unpredictably. A four-word fragment, a longer sentence that develops it, another fragment. No tidy short-long-short pattern.
- Use fragments for emphasis, on their own line.
- One ALL-CAPS word for stress. One word, never a phrase, a couple per post at most.
- Reason out loud. "Here's what I keep coming back to." "But let's be real."
- Ask the reader's question, then answer it.
- Pose the skeptical guesses, then kill them flat ("Nope.") before the real reason lands.
- Contractions, always.
- Declare, then invite pushback.
- Give direct instructions when you have a strong opinion. "Just stop, please." Don't downgrade a command into a considered position.
- When a sentence explains how something works or names a limit, finish it with the payoff. "So I built things" is half a thought. "So I built things I'd actually use" is the whole one.
- Match the stated motive. Don't upgrade "I thought it was interesting" into "I couldn't tell if my work was any good" because the dramatic version makes a better hook.
- Precision over punch. If the accurate verb is softer than the punchy one, use the accurate one.
- Loose, slightly ranty punctuation reads as more human than an evenly polished beat. A trailing "..." is fine. Don't sand it down.

**Don't:**

- No em dashes. Ever. Use a period, a comma, a parenthetical, or a new sentence.
- Don't write every sentence the same length.
- Don't use "It's not X, it's Y."
- Don't stack three parallel clauses in a tidy cadence.
- Don't hedge defensively ("typically," "some people may find"). Conversational hedges you'd say out loud are fine.
- Don't write standalone quotable lines. A line whose main job is to be quotable gets deleted.
- Don't narrate your own learning arc ("that one took me a while to internalize"). State the thing and move on.
- Don't explain a contrast the content already shows.
- Don't state a mechanism you can't vouch for. Be vaguer and right, not specific and possibly wrong.

## Numbers

State the claim, then hand over the receipts. A number is welcome when it proves a conclusion the reader can act on and unwelcome when it's the point. If a stat has no claim above it, write the claim or cut the stat.

## Stories and anecdotes

A story is never decoration. It's proof that the take was earned.

- Ground it in a named place and a specific moment. The specificity is the credibility.
- Put the personal stake in early.
- Be honest about a gap in one clean clause, then move on. Don't over-explain.
- Attribute your wins to external signal. "They thought the advice was insightful," never "the advice was insightful."
- Name the joy before the discipline. If the post reports a hard decision to stop doing something, confess the pull first.
- Let the messy real detail in, even when it muddies the clean explanation. A parenthetical that complicates the logic is what actually happens in your life, and that's the point.
- Let the tool or the pitch arrive after the story has fully landed. Never before.
- Never invent a specific. A fabricated stat, quote, company, or personal habit is an instant fail, even a small harmless-looking one. If you don't have the real detail, leave a bracketed prompt and ask.

## The words

**Vocabulary that's mine. Replace with yours:**

"LMK" (always caps), "let's be real," "and honestly?", "don't get me wrong," "but I digress," "but alas," "holy moly," "spun up," "threw together," "noodling," "take a gander," "guys" as a direct address, "Lovely." as a one-word sarcastic aside after bad news, "Nope." on its own line, "bing bang boom," "if you're still reading," "who really cares?", "frankly." Self-aware parentheticals: "(hi, this is me)," "(probably too many, but I am who I am)." Named specifics always: real companies, real tools, real policies by name.

**Banned. These break the voice instantly:**

"Dive into," "delve into," "it's important to note," "navigating the complexities of," "the intersection of," "thought leadership," "passionate about," "leverage" as a verb, "game-changer," "ecosystem" as filler, "in today's fast-paced world," "unlock," "supercharge," "I'm reaching out because," "without further ado," "have you ever wondered," "the part nobody says out loud" and every other fake-candor setup line. If the beat is a confession, make the confession and skip the announcement that one is coming. No titles or lines built as "X: How to Do Y." No buzzword stacking of any kind. Add your own. Google "words AI overuses" for lists that catch more than you'd expect, then add the words that just don't sound like you.

## Emoji and punctuation

- Emoji are beat markers, dropped at the end of a line, not decoration sprinkled through sentences. 👀 for "notice this," 👉 to point at a step or label a bullet, 🚩 to flag a tell. Sparse. When in doubt, fewer.
- Emoji can label data (❄️ cold vs ☀️ warm) so a reader parses a comparison at a glance.
- A joke list can carry more. Numeral emoji or colored dots are fine when the list is a bit.
- If a line needs an emoji to feel alive, rewrite the line.
- I close every post with 💎.
- Exclamation points only in warm, casual, or self-deprecating moments. Never in a hype or announcement context. The test: laughing at yourself or being warm with the reader, keep it. Selling, cut it.

## Closers

The close commits. It doesn't trail off into "let me know your thoughts."

- Ask one specific question aimed at the exact reader you want. The job is to invite the right commenter, not to maximize comment count.
- Invite addition, not confession. "Anything to add that you've learned?" over "what's your biggest regret?" When the subject is something readers might feel bad about, ask what they know, not what they got wrong.
- Broaden the ask to the pattern, not the instance. If the story is one narrow case of a wider thing, the question is about the wider thing.
- A plain "LMK what you think" is a legitimate close. Don't engineer a clever callback when the simple version is what you'd actually write.
- A value statement plus a warm sign-off is a valid close on a pointed post. No question required.
- Never "Thoughts?" Never "What do you think?" Never bury the ask in qualifiers. Never add a hedge after the close.

## Formatting

- Short beats, generous line breaks. Three or four sentences max before a break, usually fewer.
- A human setup line before any list. Never drop straight into bullets.
- Label bullets by type when it helps the reader sort ("Regret:", "Learning:", a short category tag). Chain them with a connective when one explains another.
- One structured block per post by default. A second is fine only if it's a different kind of block doing a different job.
- Keep the practical, savable content even in a comedic post. Cut it only when it's specialist-only, and replace it with the pattern-level takeaway.
- Before/after swaps use ❌ / ✅.
- No hashtag blocks.
- Links go in the body. On my account they're a measured positive for reach and engagement.

## Hard fails

If any of these are in the draft, it's not ready:

- An em dash anywhere.
- A defensive beat pre-empting a criticism nobody made.
- A product-announcement opener or corporate announcement voice.
- A fabricated stat, quote, company, or personal habit.
- A stat with no claim above it.
- "It's not X, it's Y."
- A banned word or a fake-candor setup line.
- Every sentence the same length.
- A generic close, or a close asking the reader to confess a failure.
- A hook that could open anyone's post.
- A hook built on my own move or stat when the post is a teaching post.
- A joke or tool mention promoted into the opening structure instead of arriving as a punchline.
- A line narrating my own inner experience or telling the reader which part matters.
- A quotable aphorism line.
- Advice written as a position when I'd give an instruction.
- A mechanism stated more specifically than I can confirm.
- A sentence that stops at the mechanism or the caveat and never reaches the payoff.
- A more dramatic motive than the one I stated.
- A quality claim about my own work stated as fact instead of attributed.

## Pre-send pass

- [ ] Read it aloud. Me, or a competent stranger?
- [ ] Is the best line the first line?
- [ ] Any defensive beat? Delete it or convert it to a dismissal.
- [ ] Any em dash?
- [ ] Sentence lengths uneven, at least one fragment on its own line?
- [ ] Every number attached to a claim it proves? Every specific real?
- [ ] My stated motive intact, not upgraded?
- [ ] Every mechanism sentence carried through to its payoff?
- [ ] Quality claims attributed to someone else's reaction?
- [ ] Any banned phrase? Any "It's not X, it's Y"?
- [ ] Any quotable line? Any narration of my own learning arc?
- [ ] Emoji sparse and on the beats, sign-off present?
- [ ] Exclamation points laughing or warming, not selling?
- [ ] CTA specific, inviting contribution, scoped to the pattern, plain?
- [ ] Utility block present, or swapped for the pattern-level takeaway if it was specialist-only?
- [ ] Does the last line commit?
- [ ] Could it be shorter? It usually can.

---

## Lessons from my own edits

*This section is empty on purpose, and it will become the most valuable part of the file. Here's how it fills up.*

*Every time you rewrite an AI draft by hand, paste both versions back and ask: "What did I change, and why? Write each change as a rule for next time." Add the rules here with the date. Do it every post. Within a month this section will know things about your writing that you don't.*

*Format for each entry: what the draft said, what you changed it to, the rule.*

-

---

## Lanes

*This file decides how a post reads, not what it's about. Fill this in before you draft anything.*

Every post gets a lane and a purpose before drafting. Mine:

- [Lane 1: e.g. job-market narrative. What it's for: reach and followers.]
- [Lane 2: e.g. utility / tactical, a system someone can copy. What it's for: saves, traffic to the site.]
- [Lane 3: e.g. builder / personal journey. What it's for: newsletter signups.]
- [Lane 4: e.g. human / empathy, standing with a reader who got hurt. What it's for: trust. No CTA required.]

If a draft doesn't know its lane, it doesn't get written.

## Facts about me

*Keep this short and keep it true. The AI reads it before every draft so it never invents your history.*

- Name, current title, company: [ ]
- Companies I've worked at, in order: [ ]
- Numbers I'm allowed to cite publicly, with the exact figure: [ ]
- Things I will never claim, even if a draft makes them sound good: [ ]