How to write LinkedIn posts with AI: 8 prompts, and the rules LinkedIn set after it cut its own AI writer

How do you write a LinkedIn post with AI?

You give a chatbot your raw material and your voice, then cut what it adds. These five steps work the same in ChatGPT, Claude, Gemini or Microsoft Copilot, free or paid.

1. Paste three posts you wrote yourself and ask for the rules of your voice (prompt 1 below).
2. Save those rules where the chatbot keeps them: ChatGPT custom instructions, Claude's Instructions for Claude, or a Gemini Gem.
3. Give it your notes: the result, the number, the person, what went wrong. The model can't know any of it.
4. Ask for a draft whose main point sits in the first 150 characters, and check the length yourself.
5. Edit it by hand, then run the slop check (prompt 7) before you post.

Prompt 1, the voice file every later prompt leans on: "Here are three LinkedIn posts I wrote myself: [paste]. List the rules that describe how I write: how long my sentences run, how I open, words I use often, words I never use, and how I end. Write them as a list I can paste into custom instructions. Don't praise the posts, and don't add rules you can't point to in them."

LinkedIn's own numbers belong in every prompt. Each row below comes from a LinkedIn help page or a LinkedIn executive, read on 12 October 2026.

LinkedIn doesn't publish where an ordinary post folds behind "see more". Its ad specs do: "Use up to 150 characters including spaces, emojis, and punctuation to avoid truncation." Treat 150 characters as your hook budget. That's about 28 words by OpenAI's rule of thumb that "1 token is approximately 4 characters" and three quarters of a word.

Don't trust a chatbot's own character count. Models read tokens, not letters, so they guess at length. A 3,000 character post is about 750 tokens. Paste the draft into a character counter before you post it.

Does LinkedIn penalize AI-written posts?

No, LinkedIn doesn't penalize a post for being written with AI. It cuts the reach of posts that read as generic, and on 30 July 2026 it said it was removing its own AI writer.

Laura Lorenzetti, LinkedIn's VP and executive editor, set the rule in a post on 20 May 2026: "It's ok to use AI to help you write, but your posts and comments need to represent your voice and your perspectives." In the same post she said LinkedIn's systems correctly identify generic content 94% of the time in initial testing. LinkedIn hasn't published a false positive rate.

Flagged posts aren't deleted. LinkedIn shows them less to people outside your immediate network, so your connections may still see a generic post while strangers don't.

Ten weeks later, chief product officer Hari Srinivasan went further. On 30 July 2026 he said members can now report a post or comment that "seems like AI slop". LinkedIn will also test a private flag in your analytics dashboard when readers find a post inauthentic, and it was removing the "enhance your post" feature. Its replacement proofreads your words and doesn't write new ones. The old feature was gone by August, Social Media Today reported on 20 September 2026.

The tool had been struggling for a year. CEO Ryan Roslansky told Semafor in July 2025 that people who post AI content "get called out", and that "you don't want to get called out in your professional context."

So let AI write sentences. Keep the experience, the numbers and the opinion yours, because those are exactly what a generic post lacks.

How many LinkedIn posts are already written by AI?

Between 40% and 81% of long LinkedIn posts, depending on whose detector you believe. Two detector companies measured LinkedIn in mid 2026 and got answers a factor of two apart.

They count different things. Pangram flags only text it judges fully AI-generated. Originality.ai uses what it calls a 15% AI Allowance and doesn't define it further. Both companies sell the detectors they used.

Pangram's other number is the sharper one. LinkedIn posts were a third of everything its users scanned, yet LinkedIn produced 62% of all the AI content it flagged. Neither study measured engagement, so any claim that AI posts get some percentage less reach doesn't come from these datasets.

That's the feed your post lands in. A draft that sounds like the median long LinkedIn post sounds, by both counts, like a machine.

Which prompts write a good LinkedIn post?

The prompts that work hand the model your facts and forbid it from inventing more. These eight cover the job from a blank page to the last edit. Longer lists exist, but most of their entries are topic ideas, and the voice and fact rules matter more than the topic.

Prompt 2, interview me first: "I want a LinkedIn post about [project or event]. Before you write anything, ask me five questions, one at a time, about what happened, what I learned and what number proves it. Then write the post using only my answers."

Prompt 3, draft from notes: "Write a LinkedIn post from my notes below, using my voice rules. Put the main point in the first 150 characters. Keep it under [200] words, with short paragraphs. Don't invent numbers, names, quotes or results that aren't in my notes. Notes: [paste]"

Prompt 4, five hooks: "Give me five opening lines for this post, each under 150 characters. Each one has to state a specific fact, number or result from the draft. No questions and no announcements about how excited I am."

Prompt 5, repurpose: "Turn this [article, talk transcript, report] into one LinkedIn post. Keep the single strongest finding and the number behind it. Drop everything else. Tell me which claim you chose and why."

Prompt 6, proofread only: "Edit my draft in three passes. Fix spelling, grammar and punctuation. Cut it by a third without losing a fact. Rewrite any sentence over 25 words so it's easier to read. Show every change, and don't add claims, hashtags or emoji." That's the same Review, Shorten and Clarify trio LinkedIn sells to Premium members, in any language.

Prompt 7, the slop check: "Read this draft as a skeptical LinkedIn reader who sees AI posts all day. Quote every phrase that sounds generic or machine written, like a stock opener, a not this but that line, or a run of three adjectives. Give a plainer version for each, using words from my own posts."

Prompt 8, hashtags: "Suggest up to three hashtags that are closely related to the topic of this post, or tell me it needs none." A LinkedIn statement quoted by Social Media Today says hashtags work best when they're "closely related to the topic of the post", so the prompt asks for exactly that.

For the reasoning behind prompts like these, how to use AI for writing covers the general method and what a system prompt is explains why saved instructions change every answer.

Which AI is best for LinkedIn posts?

The best AI for LinkedIn posts is the one that keeps your voice rules saved, because no public benchmark scores LinkedIn posts. ChatGPT, Claude and Gemini all draft a short post on their free plans. Where they differ is where your rules live and how much fits.

ChatGPT Free is the tightest fit. OpenAI's custom instructions page caps Free and Go at 1,500 characters, which holds a voice file but not much else. Claude applies its account instructions to every chat, per Anthropic's personalization guide. Google's Gems page lets you add files under Knowledge, so a Gem can hold your best past posts as examples.

LinkedIn's Post proofreader is the only one inside LinkedIn, and it's the most limited. Its help page covers English drafts only, one suggestion at a time, with no undo or version history.

Whizi, which publishes this page, applies a Skill to whichever model you pick, so one set of voice rules runs on Claude and ChatGPT alike. On Pro you can draft one post in two models side by side and keep the better one. It loses on the free question: a free Whizi account gets 10 messages a day, on Whizi AI, Auto and Claude Haiku 5.5 only.

What shouldn't you hand to AI on LinkedIn?

Don't let AI post or comment for you. LinkedIn's User Agreement bans using "bots or other unauthorized automated methods" to "create, comment on, like, share, or re-share posts". Srinivasan wrote that LinkedIn now catches hundreds of thousands of automated comment attempts every day.

Don't let it invent your results either. A model asked for a success story fills the gaps with plausible numbers, and on LinkedIn your name is on them. Prompts 2 and 3 above exist to stop that.

AI images are a separate case. LinkedIn shows Content Credentials on images and video that carry C2PA data, and the label can show whether AI was used to make them. Text carries no such label, so readers judge your words by how they sound. If you write for clients as well as yourself, AI for copywriters covers who owns the copy.