AI-written LinkedIn posts: how to keep your voice (and stop sounding like everyone else)

You can spot an AI-written LinkedIn post from the first line: the tidy triplet ("Not X. Not Y. Z."), the "In today's fast-paced world", the em dash doing the work of a thought, the lesson with no story under it, the emoji before every bullet. The feed is full of them, and readers have learned to scroll past.

The problem is not AI. It is AI used without material and without a voice to match. A model given nothing but "write a post about pricing" returns the average of everything ever written about pricing, which is, by definition, what everyone else sounds like. This guide is about the alternative: using AI to write posts that are unmistakably yours, faster than you could alone.

Why generic prompts produce generic posts

Language models predict the most likely next words given the input. Feed them a generic input and the most likely output is the generic post. Three ingredients change the odds:

• Your writing as the style reference, so the model has your rhythm, not LinkedIn's.
• Your facts: the actual numbers, dates, names, quotes, as the only material it is allowed to use.
• Your answers to pointed questions, so the post contains what only you know.

Remove any one of them and the post drifts back toward the average. The workflow below is built around never removing them.

What "your voice" actually is

Voice is not a vibe. It is a set of habits a careful reader (or a model) can measure from thirty of your posts:

• Sentence length and rhythm. Short and punchy, or long and winding. Whether you use fragments. Where you break lines.
• Vocabulary. The words you reach for ("ship", "customers", "figure out") and the ones you never use ("leverage", "synergy", "delve").
• Punctuation and ornament. Emojis or none. Exclamation marks or none. Em dashes or full stops. Hashtags or not.
• Structure. Story first or lesson first. Lists or prose. Whether you open with a scene or a claim.
• Stance. How direct you are, how much you hedge, how often you admit mistakes, whether you address the reader as "you".

Any of these can be learned from your past posts, if the system is asked to learn them and forbidden from overriding them with defaults. "No emojis, ever, because Tom never uses them" is a rule; "professional tone" is not.

The anti-invention rule

The most damaging thing an AI can do on your behalf is invent a statistic or a customer anecdote. It reads well and it is false, and on LinkedIn, where your name is attached, one fabricated number found out costs more than a year of good posts. Any system you use should be able to trace every claim in a post back to something you said or shared, and should refuse to write the ones it cannot. If it needs a detail it does not have, it should ask you, not guess.

This is the design principle behind the interview in ShipMyPosts: the model plans the topics, then asks you for the specifics. "you said pricing went from $9 to $24; what broke at $9?", and writes only from your answer and your sources. Nothing in the post exists that you did not put there.

A workflow that keeps the voice

• Feed it your last 30 posts (the ones you wrote yourself) as the voice sample, and mark the two or three you like most.
• Feed it your sources: podcast transcripts, articles, talks. See turning a podcast into 10 posts.
• Let it plan topics, then interview you for the specifics, in your language, by voice if that is how you think. Ten minutes of talking beats an hour of prompting.
• Have it draft in your rhythm with explicit rules: line length, no emojis if you never use them, no em dashes, no triplets, your vocabulary.
• Generate several hooks and pick one: the hook is where AI defaults show most. See 25 hook formulas.
• Edit. Read each post aloud. If a line is not yours, rewrite it or tell the model exactly what to change ("shorter, drop the second paragraph, keep the €10K number").

The tells that expose AI writing

Use this list on every draft; delete or rewrite anything that matches.

• The triplet. "Not a feature. Not a tool. A mindset." Humans rarely talk in threes.
• The false contrast. "It's not about X, it's about Y" when nobody said it was about X.
• The inspirational close. "And that, my friends, is the real lesson." Just end.
• Adjective stacks. "Powerful, transformative, game-changing." Pick zero.
• The universal opener. "In today's fast-paced world…", "Let's be honest…", "Here's the thing:".
• Emoji bullets. 🚀 in front of every line.
• Suspiciously round numbers and anecdotes with no names, dates or places.
• Em dashes everywhere: one is fine; five is a signature.
• Perfect balance. Every paragraph the same length, every point with a counterpoint. Real writing is lumpier.

How to edit a draft back into your voice (five minutes)

• Read it aloud. Anywhere you stumble, the model wrote it, not you.
• Swap the vocabulary. Replace every word you would not say with the one you would.
• Break the symmetry. Cut one paragraph short. Let one line stand alone.
• Add the detail only you have. A name, the exact sentence someone said, the date.
• Cut the close. Delete the last inspirational line; end on the specific or the question.

A draft that started at 80% you becomes 100% you in five minutes. A draft that started at 0%, generic prompt, no material, cannot be edited into your voice at all; you would be rewriting it.

When AI helps most (and least)

Most: extracting angles from long material, drafting twenty bodies from decided hooks and specifics, rewriting to a constraint ("shorter", "punchier", "focus on the churn number"), producing hook variants, generating a carousel from a text post.

Least: deciding what you think. Inventing stories. Writing the post you did not have material for. Replying to comments, that is a conversation, and readers can tell.

The honest position

Readers do not resent AI-assisted writing; they resent being served the average. A post that says something specific, in a recognisable voice, with a detail only you could know, is yours regardless of what tool arranged the words, and a post with none of those things is nobody's regardless of who typed it. The goal is not posts that sound like AI wrote them well. It is posts that sound like you, on a day you had time.

A before/after

Here is what the difference looks like in practice. Same angle. A pricing change, two drafts.

Generic prompt, no material: "Pricing is one of the most important decisions for any SaaS founder. It's not just about revenue, it's about positioning. When we raised our prices, we learned valuable lessons about customer perception and value. Here are three takeaways every founder should know: 1) Know your worth. 2) Communicate clearly. 3) Don't be afraid to lose customers. 🚀"

From the founder's material, in their voice: "We raised MailTracker from $9 to $24 in one email. I rewrote that email eleven times. Signups dropped about 30% the first month. Revenue doubled. Support tickets fell by more than half, because the people who stayed were the people who actually needed it. The number I was scared of was the wrong number."

The second draft has a name, two figures, a scene (eleven rewrites), a consequence nobody expected (support fell) and a last line the reader can argue with. Nothing in it could have been written by anyone else, and a model wrote most of the sentences, from an interview answer that took ninety seconds.

A short voice rulebook you can copy

Write yours in ten minutes and hand it to whatever tool you use. Ours, as an example of the shape:

• Sentences: mostly under 15 words. Fragments allowed. One idea per line.
• Never: emojis, hashtags in the body, em dashes more than once per post, the words "leverage", "journey", "unlock", "game-changer", "delve".
• Always: at least one number or name per post. Say "customers", not "users". Say "I", not "we", when it was me.
• Openers: a scene, a number, or a claim. Never "Excited to share", never a question I can answer myself.
• Endings: a question the reader has an opinion on, or the specific, or the callback. No inspirational summary.
• Stance: direct, no hedging, admit the mistake before the lesson.

Keep it next to your drafts. When a line breaks a rule, the rule tells you why it does not sound like you.

Your voice is not something AI takes from you. It is something you have to give it, and then refuse to let it overwrite.