How to Write a LinkedIn Post Using ChatGPT That Actually Gets Engagement

Why Most AI LinkedIn Posts Get Ignored

AI-written posts underperform for two mechanical reasons, not one aesthetic one:

• The vocabulary flags them. Certain words cluster in machine-written business prose at rates no hurried human produces.
• The first line is wasted. The opening gets spent on scene-setting, in the only space the feed guarantees you.

LinkedIn published an unusually detailed account of how its feed was rebuilt in LinkedIn’s own engineering post on the next generation of its Feed (March 2026). The short version for anyone who does not write ranking systems for a living: a language model now reads what your post actually says and matches it against what a given reader has shown interest in. A big network no longer guarantees distribution. Relevance to a specific reader does.

That shift is good news for anyone who wants to write a LinkedIn post using ChatGPT and be read. It means the winning move is not volume or timing tricks. It is saying something specific enough that a machine can tell who would care. Generic posts lose under this system by design, because there is no particular reader they are relevant to.

The vocabulary that gives you away

Ask any experienced editor and they will tell you the same thing: the giveaway is never the idea, it is the word choice. Certain words appear at wildly higher rates in machine-written business prose than in anything a human types in a hurry. Readers cannot name the pattern, but they recognise it, and they scroll.

You fix this by banning the words up front rather than editing them out afterwards. The full ban list is in the prompts further down.

Where the feed stops reading

Your post can run to three thousand characters, but the feed shows roughly the first two or three lines before collapsing the rest behind a “see more” link. On a phone that is around 140 characters. LinkedIn does not publish the exact number and it shifts with device and screen width, so treat it as a budget rather than a rule.

The practical consequence is blunt. If your first line is scene-setting, nobody reaches the point.

The Raw Material Method, Explained

The Raw Material Method is a three-layer way of building a prompt: you supply the raw material, ChatGPT supplies the structure, and a fixed set of guardrails stops it drifting into generic business voice. The rule underneath it is that ChatGPT never sources the substance. It only arranges what you already know.

Raw material means one specific thing that actually happened, written as messy notes. A number, a date, a decision you regret, what you changed. Four or five fragments is plenty. If you cannot produce them in ninety seconds, you do not have a post yet, and no prompt will rescue that.

If you have met our Logic System for structuring any AI request, this is that idea narrowed to one job. The master prompt below is the layer-two and layer-three scaffolding, permanently. You only ever edit the notes at the top.

You are an editor helping me turn rough notes into a LinkedIn post. You are not the author. Use only what I give you below. Do not invent numbers, quotes, job titles, or events.

MY NOTES:
- [what happened, one line]
- [the number, date or detail that makes it specific]
- [what went wrong, or what surprised me]
- [what I changed, and what happened next]

WHO I AM: [your role] at [type of company].
WHO I AM WRITING FOR: [the job title of the person you want to reach].

STRUCTURE:
- Hook: one line, under 12 words, stating the most surprising fact from my notes. No question mark, no emoji.
- Context: two or three short lines on the situation and why it was hard.
- Lesson: two or three lines on what I changed and what it produced.
- CTA: one specific question aimed at the reader above, about their own situation.

RULES:
- Under 200 words total.
- No paragraph longer than two lines. Blank line between every block.
- Zero emoji. No hashtags.
- Do not use these words: delve, testament, leverage, unlock, harness, game-changer, revolutionize, navigate, journey, landscape, thrilled, humbled, seamlessly.
- Plain language. Short sentences. No phrase that sounds like a press release.

Return the post only, with no preamble and no explanation.

Hook, Context, Lesson, CTA Explained

Hook, Context, Lesson, CTA is a four-part structure for a professional post: a one-line opening that earns the click, a short setup that establishes stakes, a takeaway the reader can use, and a closing question narrow enough to answer in a sentence. Each part has a job and a length budget.

The hook is the part people over-engineer. It does not need to be clever, and it should not be a question — questions read as bait and give the reader nothing. The strongest hook is usually the single most surprising fact sitting in your notes, stated flatly. If your notes contain a number that made you wince, that is your first line.

The CTA is where almost every AI draft collapses. ChatGPT defaults to a question so broad that answering it requires writing an essay, so nobody bothers. Narrow the question until a busy person could reply from a phone at a bus stop.

How to Make ChatGPT Sound Like You

Voice is not a tone instruction. Telling ChatGPT to “sound conversational” produces the same beige register every time, because that instruction means nothing without samples. Two things genuinely move it: banning what you never say, and showing it what you do say.

Set your constraints once

The ban list in the master prompt does most of the work, and it works because it is a negative constraint rather than a positive instruction — telling a model what to avoid is far more reliable than telling it what to aim for. Put that list somewhere permanent instead of retyping it: ChatGPT’s Custom Instructions apply your rules to every chat automatically, and ChatGPT’s memory can hold your role and audience so you stop restating them.

Give it three real examples

Then hand it evidence of how you write. Three posts you were happy with, or three paragraphs from emails you have actually sent, beat any adjective you could supply. This is few-shot prompting, and for voice matching it is the only technique that reliably works.

Below are three things I have written. Read them and describe my writing voice as a set of rules another writer could follow.

Cover: average sentence length, how formal I am, whether I use contractions, how I open, how I close, words and phrases I clearly favour, and anything I never do.

SAMPLE 1:
[paste something you wrote]

SAMPLE 2:
[paste something you wrote]

SAMPLE 3:
[paste something you wrote]

Return the rules as a short numbered list I can paste into my Custom Instructions. Do not flatter the writing or comment on its quality.

Save the output. Paste it under a “MY VOICE” heading in the master prompt, and the drafts stop reading like a stranger wrote them.

Write a LinkedIn Post Using ChatGPT: Four Prompts

These are the four situations that produce most professional posts. Each one is the master prompt with the notes section pre-shaped, so you can write a LinkedIn post using ChatGPT for any of them without rebuilding the scaffolding. Fill the brackets, paste, edit for two minutes, publish.

Before you paste anything, one boundary worth setting properly.

Full detail on where those lines sit is in our guide to using ChatGPT safely with work information, and the plan-by-plan breakdown is on OpenAI’s own ChatGPT plan comparison page.

The highest-performing professional post is a competence story told through a mistake. It works because the alternative — a clean success story — reads as advertising.

Use the master prompt rules. My notes:

- Project: [what it was, and who it affected]
- What went wrong: [the delay, the resistance, the miss — be specific]
- What I tried first: [the thing that did not work]
- What actually worked: [the change you made]
- The result: [the number or outcome]

WHO I AM: [your role].
WHO I AM WRITING FOR: [job title].

Extra rule for this one: I want to sound like someone who learned something, not someone confessing or defending. Do not apologise on my behalf. Lead the hook with the failure, not the recovery.

You disagree with something your field treats as settled. Written well, this earns more genuine comments than anything else you post. Written badly, it reads as a swipe at people you work with.

Use the master prompt rules. I want to disagree with a common view in my industry without sounding aggressive or superior.

THE COMMON VIEW: [the thing everyone repeats]
WHY I DISAGREE: [your reasoning, in plain terms]
WHAT I HAVE SEEN INSTEAD: [the specific experience behind your view]
WHAT I WOULD DO DIFFERENTLY: [your alternative]

WHO I AM: [your role].

Extra rules for this one: acknowledge the strongest version of the opposing view before disagreeing. Never imply people who hold it are lazy or naive. End by inviting disagreement rather than closing the argument.

Nobody reads the photo-and-gratitude conference post. A recap earns attention only when it hands over the takeaways so the reader does not have to attend next year.

Use the master prompt rules. Turn my event notes into a recap that is useful to someone who was not there.

EVENT: [name and type]
THE THREE THINGS WORTH REPEATING:
- [takeaway one]
- [takeaway two]
- [takeaway three]
WHAT CHANGED MY MIND: [the one that surprised you]

WHO I AM WRITING FOR: [job title].

Extra rules for this one: no praise for the organisers, no mention of the venue, no photo captions. Open with the takeaway that surprised me most. If a takeaway is vague in my notes, say so rather than inflating it.

The fix here is structural, not tonal. Make the client the one who did something and the method the thing worth learning, and the boasting problem disappears on its own.

Use the master prompt rules. Write this as a short case study where the client is the one taking action, not me.

THE SITUATION: [what the client was stuck on — no names, no identifying details]
WHAT WE CHANGED: [the approach, in three steps]
THE RESULT: [the outcome, with a real number if I have one]

WHO I AM: [your role].
WHO I AM WRITING FOR: [job title].

Extra rules for this one: the word "we" appears at most twice. No company names. No claim I cannot evidence. The lesson must be usable by someone who will never hire me.

If you would rather start from a set that is already written out, these sit alongside about a hundred others in our free library of AI work templates.

The Two-Minute Polish Before You Post

Never publish the raw output. Not because it will be embarrassing, but because it will be almost right, and almost right is what readers detect. Whenever you write a LinkedIn post using ChatGPT, budget two minutes for four checks, in this order:

1. Check every fact. ChatGPT rounds, tidies, and occasionally invents. Confirm each number and date against your notes.
2. Scan for the ban list. One survivor is enough to flag the whole post.
3. Fix the line breaks. Blank line between every block, nothing longer than two lines.
4. Add one detail only you know. The weather, the room, what someone said. One concrete thing.

That fourth check is the one that matters most, and it is the one every other guide skips. A structurally perfect post with no specific human detail in it is still recognisably machine-made.

Fact-checking is not optional here, and the reason is worth understanding rather than just obeying — see our explanation of why AI models invent plausible details. A fabricated statistic in a post carrying your name and job title is a professional problem, not a technical one.

When ChatGPT Gets the Post Wrong

Most bad drafts fail in one of five predictable ways, and each one traces back to a missing instruction rather than a model limitation. Match the symptom, add the line, regenerate.

One habit worth building: when a draft is close but wrong, fix the instruction rather than starting again. Our guide to refining AI output without starting over covers the diagnostic version of this — identify which layer failed, change only that layer.

What ChatGPT Cannot Do for You

Being straight about the limits is the difference between a system you can rely on and one that quietly embarrasses you in six months.

There is also a reasonable question underneath all of this: will people know? Some will suspect, and that is fine. What readers actually object to is not assistance with sentences — it is content with no person behind it. A post built from your own raw material passes that test whether or not a model helped you arrange it. The same principle scales, which is the argument in our piece on producing more content without losing your voice.

Start with the retrospective prompt this week, on something that genuinely went sideways. Once you have done it twice, you will be able to write a LinkedIn post using ChatGPT in the gap between two meetings, and it will still sound like you.