LinkedIn post generator that writes in your voice
Most AI LinkedIn posts are recognisable within a line and a half. The rhythm is wrong, the hook is borrowed, and everyone on your feed has read the same three openers this week.
Why it matters here
LinkedIn is the format where sounding generic costs most, because your name is on it and your network already knows how you talk. A post that reads like everyone else's does not just fail to land; it quietly tells people you outsourced it.
How ScriptGrain does it
- Give ScriptGrain writing you have already published: posts, newsletters, anything with your fingerprints on it.
- It measures 45 style attributes: sentence-length variance, how often you use dashes, whether you front-load or build to the point, your actual opening habits.
- Every LinkedIn post it drafts is constrained to that profile, so the rhythm and the openers are yours rather than the model's defaults.
Posts are drafted around 220 words by default, which is the length that tends to survive the feed's truncation, and you approve an outline before the draft is written.
What a LinkedIn post in your voice looks like
Start with the boring stuff. A post in your voice is mostly rhythm, and rhythm is measurable.
Look at your last few posts. How long are your sentences on average? More to the point, how much do they vary? Some people write a long, winding sentence and then stop dead. Five words. Done. Others keep a steady middle-length pace all the way down the page. Neither is right. Yours is yours, and readers who know you pick up the pattern without ever naming it.
Openers matter just as much. Some writers start with the conclusion and back it up. Others open on a small scene, a number, a question, or a flat statement of opinion. You probably have two or three habits you fall into without noticing. The same goes for paragraph openers: do you lean on "So", "Here's", "The", "You"? That's a fingerprint.
Then the punctuation. Contractions or none. Dashes or none. Parentheses for asides, semicolons for the occasional longer thought, a question mark mid-post to pull the reader in. Commas sparse or generous. These habits are small, and together they're most of what "sounds like you" actually means.
Length is the last piece. A post that runs 220 words reads differently from one that runs 380, and your own pattern sits somewhere in that range. The format spec here allows 80 to 400 words, with 220 as the default. Pick the length your real posts tend to land at, not the one an assistant reaches for when left alone.
Where AI LinkedIn posts give themselves away
Generic AI posts fail in predictable places. Once you've spotted them, you can't stop.
The first is the stacked hook. A bold claim, then a one-line paragraph, then another one-line paragraph, each one a little more breathless than the last. Real writers do use short lines. They just don't use them on every line, and they don't use them to manufacture suspense about a point that turns out to be modest.
The second is uniform length. Read an AI draft aloud and notice that nearly every sentence is about the same size. Twenty words, twenty-two, eighteen, twenty-four. Human writing is lumpier. It has a long sentence that sprawls a bit, then a fragment. Then another long one. That unevenness is hard to fake on purpose, and models smooth it out by default.
The third is the borrowed structure. Problem, lesson, three takeaways, a question for the comments. It's a template, and your network has seen it a hundred times this month. Add the humble-brag close ("I was nervous to share this, but...") and the post announces its own genre before it says anything.
The fourth is the vocabulary. Words like "journey", "unlock", "leverage" and "landscape" show up because they're statistically safe, not because you'd ever say them. If you want a fast read on this, the AI cliché checker flags the usual suspects. Fair enough, it's a blunt instrument. But it catches the obvious ones before a colleague does.
A worked example
Take an illustrative brief: a post about why your team stopped running a weekly status meeting and what replaced it. The writer has a dry, conclusion-first habit, short fragments after a longer sentence, and a fondness for "so" as a bridge.
A default draft from a general assistant opens with a hook line about meetings being broken, drops into three or four single-sentence paragraphs, lists three lessons in matching grammar, and finishes by asking the audience what they think. Every sentence is about the same length. Nothing in it is wrong. Nothing in it is yours either.
Now constrain the same brief to the profile. Here's what changes.
The opening goes direct: the conclusion, stated flat, in the writer's own register. The paragraphs stop being one line each and go back to varied lengths, because the profile says this person writes some paragraphs long and some short. The fragments land after a longer build-up sentence instead of being sprinkled everywhere. The contractions come back. The dashes the writer never uses stay out. The three matching lessons become two, written differently, because the profile's habits don't run to triplets. And the close turns into a wry, low-pressure sign-off rather than a call for engagement.
The facts in the post are identical. The argument is identical. The draft just stops sounding like a template and starts sounding like a person who had the meeting and got tired of it.
Most of what's left between a generated draft and a finished one is editing. The model hands you flat, evenly spaced sentences, and you keep them because nothing in them is wrong. We cover that gap in an editing checklist, which is about getting back to your own baseline, how you write when nobody's smoothing you out.
What to feed the profile
Give it writing that has your fingerprints on it. Published posts are the obvious choice, but they shouldn't be the only thing. A profile needs at least one sample of 50 or more words, and the recommendation is three or more pieces and around 3,000 words. More variety helps. Newsletters, long emails you'd stand behind, an article you wrote without an editor rewriting it.
Why mix formats? Because a LinkedIn post is short and compressed, and short writing hides habits. Your paragraph structure, your argument order and your use of parentheses only show up reliably in longer pieces. Feed the profile posts alone and it learns your hook style but not how you actually build a case. Feed it posts plus longer work and it learns both.
One caution. Don't include anything a comms team or an editor heavily rewrote, unless you want that voice in the profile. The profile measures what's on the page. If the page isn't you, neither is the output.
Extraction usually takes under a minute; the median measured was 39 seconds. You can also run the samples through the LinkedIn writing style analyser first to see what the measurements look like before you commit. The full list of what gets measured is in the writing voice attributes reference.
Checking a post before it goes out
Every draft comes back with a voice-match score from 0 to 1. On-voice drafts typically land between 0.85 and 0.95. Partial matches sit between 0.5 and 0.8. A clearly different voice falls under 0.45. The method behind it is published in the voice measurement framework, so you can see what the number is made of rather than taking it on trust.
The headline score is a starting point. The useful part is the per-feature notes, which show the draft's value against your profile's value for each attribute. Say the draft's average sentence length is well above yours and its sentence-length variance is well below. That tells you exactly what's wrong: too long, too even. You fix that by breaking up a couple of sentences and adding a fragment, not by rewriting the lot.
Look at contraction rate, dash use and opening style first. They're the quickest to correct and the most noticeable to readers. If the score is under your target, polish will revise the draft toward it; the default target is 0.9. And when you edit a draft and save it, the app learns from those edits, which is how a profile keeps up with you. If your style moves over a year, voice drift is the term for it, and the score is how you notice.
So before anything goes live: read the score, read the deltas, fix the two or three biggest, and publish. That's the whole check.
Drafting from the assistant you already use
You don't have to leave your usual tool. The profile reaches ChatGPT, Claude, Gemini, Cursor, Meta Muse and any MCP client through the MCP server and the REST API, and it's read live each time. It is never a pasted prompt, so there's no block of style instructions to maintain or forget to update. Both the MCP server and the API are available on every plan, including Free. The details are on the API and MCP server page.
In practice, you open ChatGPT or Claude, ask for a LinkedIn post on your topic, and the assistant pulls the profile as a constraint while it drafts. You get the assistant's reasoning and your rhythm. To be fair to both, they're good at structure and idea generation; the profile just stops them defaulting to the feed's house style.
The same loop works for anything automated. Generate, score the voice match, revise if it falls short, then publish behind a measurement gate. The post only goes out when the number says it sounds like you.
Start free: post ideas from what you have already written
Partly. The free part is the content ideas tool: it suggests what to post, from what you have already written. It doesn't write the post.
Here's how it works. You paste or link 2 to 5 of your own pieces, each at least 60 words. A LinkedIn post is long enough. It reads them with the same extractor Voice Brain uses, which keeps an idea, claim, story or framework only if it can point to the exact words it came from. So nothing gets made up.
What comes back is useful. Five angles to write next, each built on one or two things you've already said, and each naming what your posts haven't said yet. Up to three threads you keep coming back to. And up to two pairs of claims from different pieces that pull against each other. Every item shows your own quoted words, so you can check the reasoning yourself. It never invents facts, numbers, clients or events about you.
No signup, nothing stored. The limits are 5 runs an hour and 10 a day per visitor. Fair enough for a tool you're not paying for.
The post itself gets drafted in the app. "LinkedIn Post" is one of 16 content types there, and it follows a plain brief: a strong first line, short paragraphs, one clear point, a light close. The Free plan gives you one free draft for the life of the account, up to 400 words. After that, drafting is paid, and the cheapest plan is Writer at £12 a month. Scoring a draft against your profile stays free either way.
So the honest route is this. Take the ideas from the free tool. Pick the angle that feels most like something you'd say. Then draft it in your voice, and see whether the post sounds like you before it goes anywhere near your feed.
When a generator is the wrong tool
Sometimes you shouldn't use a LinkedIn post generator at all. Here's the plainest case: a post about something only you lived.
A generator works from what you've given it. It has your measured voice, and it has your earlier posts. It doesn't have the afternoon a client walked out, or the thing you got wrong in your first year, or the detail that made you change your mind. It cannot add facts you didn't supply, and it shouldn't. If the post depends on that detail, you need to write it. Maybe rough, maybe in fragments, and then use the tool to tidy the rhythm rather than to invent the substance.
Second, a practical point. ScriptGrain doesn't schedule or publish to LinkedIn. You copy the post across and put it up yourself. If you want a tool that queues posts for the week, this isn't it, and a dedicated scheduler will do that job better.
Third, a ghostwriter. To be fair, a good one brings something a generator can't. Judgement. Which story to leave out. When to wait a week because the timing is off. When a post that's perfectly on voice is still the wrong post for your feed right now. A generator can tell you whether a draft sounds like you. It can't tell you whether you should say it.
That judgement costs money, and the cost varies a lot. The ghostwriter cost post sets out what you'd actually be paying for, so you can weigh it against a tool at £12 a month.
The two aren't rivals. Some people use a generator for the steady weekly posts and keep a human for the ones that matter. Others write the hard posts themselves and let the tool handle the rest. Pick whichever matches how much of the post is really yours to tell.