Why AI-written LinkedIn posts sound the same, and how to fix it
AI-written LinkedIn posts sound alike because most are made from the same thing: a topic and nothing else. The fix is something only you could write. Asked for “a post about leadership”, a language model has to supply the reader, the point, the example and the tone itself, and it supplies the most usual version of each — for you and for everyone else who asked. This guide explains where the sameness comes from, lists the tells so you can find them in your own drafts, and walks through the edit that removes them.
Understand the cause: the model fills every gap with the usual
A language model produces text from patterns in what it was trained on; that is how OpenAI itself describes ChatGPT. Give it only a topic and those patterns have to provide everything else as well — who is reading, what the point is, which example, what attitude. Left to choose, a model reaches for what is common, and everyone who leaves the same gaps gets a neighbouring post. The sameness is not a fault in the tool. It is a fair picture of how little the tool was told.
Learn the tells in the wording
Some phrases now announce a post's origin before the reader has taken in what it says: “In today's fast-paced world”, “Let's dive in”, “Here's the thing”, “game-changer”, “unlock”, “it's not just X, it's Y”. None of them is wrong, and people wrote every one of them long before chatbots did. The problem is frequency: once a reader has met a phrase in a hundred posts, the hundred-and-first is skimmed. Keep a list of the ones you notice and tell your assistant never to use them; a banned list is the cheapest instruction there is.
Learn the tells in the shape
Structure gives a draft away more reliably than vocabulary. A run of one-line paragraphs, each a finished aphorism. Everything in threes. A bold claim followed at once by “But here's what nobody tells you”. A list in which every item has the same grammar and the same length. A last line that asks “What do you think?”. Writing by a person is lumpier: one long sentence where the thought needed it, a list of two because there were two, an ending that simply stops. If your draft is perfectly regular, the regularity is the tell.
Notice what is missing: anything that could be checked
The deepest sameness is in the content. A generic draft contains no name, no date, no place, no price, no quotation and nothing that went wrong, because a model given only a topic has none of these — and if it supplies them anyway, they are invented. So it writes in abstractions: “communication is key”, “aligned teams win”. Underline everything in your draft that a stranger could verify or dispute. If nothing is underlined, there is no post in it yet, however polished it looks.
Repair it with one specific and one position
Take the most abstract sentence and replace it with what happened. “Clear communication prevents project delays” becomes “We lost nine days in March because two people each thought the other had sent the brief.” Then add what a model avoids: a view someone could disagree with. “So client briefs no longer go through a shared inbox” is a position; “communication matters” is not. One real detail and one real opinion change a draft more than any amount of rephrasing, and they are the two things only you can supply.
Cut the scaffolding
Now delete. The first line, if it announces the subject instead of starting it. Every sentence that says what the next sentence is about to say. The paragraph that restates the opening as a conclusion. The closing question, unless you want the answers. Then break the rhythm on purpose: join two of the one-line paragraphs, cut a list of three down to the two that matter. The draft gets shorter and reads as though more thought went into it — which, by now, is true.
Show the model your sentences, not adjectives about them
To stop the sameness at the source, show how you write instead of describing it. “Direct, warm, a bit dry” fits half of LinkedIn. Three posts you wrote yourself fit only you. Anthropic's prompting guidance calls examples one of the most reliable ways to steer tone and structure and suggests three to five, and nothing in that advice is specific to Claude. A full structure for the hand-over — reader, point, material, voice — is at media-pilot.io/guides/brief-ai-to-sound-like-you.
Run the swap test before you post
One question settles whether the edit worked: could a competitor publish this post under their own name without changing a word? If so, it is still a post about a topic. If they would have to take out the nine days in March, the shared inbox and the opinion about briefs, it has become a post about your work, and it will not sound like anyone else's — whoever, or whatever, typed the first draft.