AI LinkedIn posts sound the same because most of them start from a topic, like “write a post about leadership,” instead of from something that happened to the writer. With nothing specific to work with, the model falls back on the most common LinkedIn shape: a hook line, one-sentence paragraphs, a lesson, and a question at the end.
You can fix most of it before you press the button. Put the facts only you know in your notes, choose the voice on purpose, then cut the formula out of what comes back.
What makes a LinkedIn post read as AI?
You’ve probably seen these, often all in one post:
- An opening line that teases instead of saying anything: “I almost made a huge mistake.”
- One sentence per paragraph, all the way down.
- A turn announced in advance: “Here’s the thing.” “Here’s the twist.”
- A lesson spelled out at the end, repeating what the story already showed.
- A closing question nobody needs to answer: “Agree?” “What do you think?”
- “Thrilled to announce,” “humbled,” and an emoji at the end of every other line.
None of these is wrong on its own. People wrote this way on LinkedIn before AI did, and that is part of the problem: the model learned the format from those posts, so it is the safest guess when you give it nothing else.
Why does every AI post come out the same?
Mostly because the input is thin. If your prompt is a topic and a tone, the model has to invent the content, and invented content is general by nature. It can’t know that the feature took three weeks, that you nearly built the wrong thing, or what the customer actually said.
The same prompts also go around. “Write a viral LinkedIn post about X” gives thousands of people a similar post about X.
And many posts go out without an edit. The model’s draft is a reasonable first pass. Published as is, it reads like everyone else’s first pass.
What should you put in your notes?
Write down what happened, with names, numbers and dates, and what you think about it. Bullet points are enough. For example:
Shipped CSV import this week
- most requested feature, 3 weeks of work
- almost built it wrong: users wanted mapping, not upload
- next: Google Sheets syncFour lines, but each one gives the post something to stand on: what shipped, how long it took, the mistake, and what comes next. The mistake line does the most work. It is the part nobody else could have written, and it gives the post a point. How to write a LinkedIn post about a feature you shipped turns these four lines into a template.
If you want more of you in it, add the detail you’d tell a friend: what a user said that changed your mind, or what the first version looked like. The generator is told to take its facts only from your notes, and so should any prompt you write for ChatGPT or Claude. The more specifics you give, the less room it has to be generic.
Does the voice you pick matter?
Yes. It decides the shape of the post, and it is the part most people leave on the default. Here is how the same four lines opened in each of the generator’s personas (real outputs from Claude Opus 5.5, shown on the tool page):
| Persona | How it opened |
|---|---|
| No persona | ”This week we shipped CSV import, and I nearly built the wrong thing.” |
| Builder | ”CSV import is live 🎉“ |
| Marketer | ”We almost built the wrong thing. 😅“ |
| Journalist | ”We shipped CSV import this week.” |
| Storyteller | ”The upload button worked. You picked a file, it went in, the little bar filled up, and for a moment I thought the hard part was done.” |
All five kept the facts: three weeks, mapping not upload, Google Sheets next. What changed was everything around them. The Marketer post is the familiar LinkedIn formula, with one-line paragraphs, “Here’s the twist:”, and a call to action at the end. That is what the persona asks for, and it suits some people. If it isn’t how you talk, don’t pick it.
The Journalist post is the shortest, about 700 characters, and it ends like this:
The request named the feature. It did not describe the need. We caught the difference before it shipped, which is the only reason this is a short post and not a postmortem.
It works because it says one specific thing and stops.
How do you edit the AI sound out?
Read the post once as a stranger would, then make these cuts:
- The hook. If the first line teases, replace it with the news or the point. “We shipped CSV import this week” is a fine opening.
- The lesson paragraph. If the story already made the point, delete the paragraph that explains it.
- The closing question. Keep it only if you want answers to it. Otherwise end on the last real sentence.
- Joined-up claims. Check each sentence against your notes. A model can join two of your points in a way you didn’t mean, or make a cautious opinion sound certain.
To find the patterns faster, run the post through the Humanizer skill, which flags common AI writing habits. For the longer version of this process, including how to get your own views into a draft, see How to Write with AI Without Losing Your Voice.
How do you make it sound like you every time?
Show the model how you already write. Keep a few of your past posts that you’re happy with, and give them to the model with the new notes: “Match the voice of these posts. Take the facts only from my notes.”
That works in any chat, but you have to paste the samples in each time. If you post often, keep your past posts, notes and drafts in one folder so an agent can read them directly.
That is what we built StashBase for. It’s our app: a free, local-first writing workspace where Claude Code, Codex or Antigravity works beside your own files. The Persona picker has Marketer, Journalist and Storyteller presets, and a Custom entry where you keep a persona prompt worked out from your own past posts. You write that prompt yourself, with the agent’s help; it is not built in one click. After that, each new post starts from your notes and your voice instead of the average LinkedIn post.
Frequently asked questions
How long can a LinkedIn post be?
Up to 3,000 characters, according to LinkedIn Help. Most posts that get read are much shorter.