How to write LinkedIn posts with ChatGPT
ChatGPT writes a good LinkedIn post when you hand it your own material and make it ask before it writes. Given only a topic, it writes everyone's post. This guide is a working method — what to give it, how to iterate, where to keep the context you would otherwise retype — and an honest account of where it stops: ChatGPT does not know your numbers, can be wrong with confidence, and on its own publishes nothing.
Hand over material, not a topic
“Write a LinkedIn post about client onboarding” gives ChatGPT nothing that only you know, so it fills the gaps with what is typical — and typical is what your readers scrolled past all week. Start with what happened: your notes from the call, the sentences you wrote to the client afterwards, the number and where it came from. Paste it rough; a draft can only be as specific as what went in. OpenAI's own advice is the same: be clear and specific, and provide enough context.
Say who reads it and what they should think afterwards
Two sentences change a draft more than any instruction about style. Who is it for: “finance leads at agencies of twenty to fifty people”, not “B2B decision makers”. And what should they think at the end: “late invoices are a process problem, not a client problem”. Without a reader and a point, ChatGPT covers every angle a little, which is how a post ends up as a summary of a subject rather than a view on it.
Make it ask before it writes
Add one line: “Before you draft, ask me up to five questions about anything you need to know.” They tend to be the questions a good editor would ask — what did it cost, what did you try first, who disagreed. Answer briefly, in your own words; phrases from those answers often belong in the post unchanged. If the questions come back generic, the material was too thin, and it is quicker to go back a step than to press on.
Ask for the first line on its own
The opening decides whether the rest is read, so do not settle for the one that arrives with the draft. Ask for eight first lines, each under about 80 characters and each of a different kind: a number, a contradiction, a mistake, a question the reader already has. Most will be poor; the best is often one you assemble from two of them. Then have the body rewritten to follow it. What makes an opening work is covered at media-pilot.io/guides/linkedin-hooks-that-work.
Revise one instruction at a time
“Make it better” returns a different draft, not a better one. Give instructions that can be checked: cut the second paragraph, replace the list with the example, end on the number instead of a question. ChatGPT can put a draft in an editable block in the conversation, where you can change the text yourself or have a selected passage revised — a full rewrite quietly discards the sentences you liked. And type at least one sentence by hand. Nothing makes a draft sound like its author faster.
Keep the standing context out of the prompt
What is true of every post — who you write for, how you spell, what you never say — should not be retyped each week. Custom instructions apply to all your chats. A project keeps chats, files and instructions together, and its instructions override the general ones inside it: the natural home for a few posts you wrote and still like, a paragraph on your reader, and rules a draft can be checked against. Memory carries some preferences between chats, but OpenAI says plainly that it does not retain every detail. And do not start a custom GPT: OpenAI is retiring them — its help centre currently plans 11 December 2026 for Enterprise workspaces and expects other plans to follow — and personal accounts can no longer create new ones.
Check every fact before it carries your name
ChatGPT can state something false in exactly the tone it uses for something true. OpenAI's help centre says so itself, and lists invented quotes, studies and citations among the things to watch for. On LinkedIn that matters twice: the post carries your name, and your readers include people who know the field. Any number, quotation, date or “research shows” that did not come from your own material gets checked at the source or deleted.
Know where ChatGPT stops
A finished draft in a chat is not a published post, and on its own ChatGPT goes no further: it cannot see how your earlier posts did, keeps no queue of what goes out when, and does not post to LinkedIn. For one post a week, copying the text across by hand works perfectly well. Beyond that you need a publishing tool. MediaPilot is one, and it can be connected to an AI assistant over MCP (media-pilot.io/product/mcp) — but whether your ChatGPT plan allows a connection that acts rather than only reads is OpenAI's decision and has changed more than once, so check its help centre first.