A voice profile made of your sentences, not adjectives about your tone
PublishFlow reads the posts you already published and writes down the phrases you reuse, the passages where you sound most like yourself, and the rhythm of your sentences. Every post it writes afterwards is written against those, word for word.
Sentence rhythm
Short declarative statement (3-6 words). Follow with longer explanatory sentence (15-20 words). Another short punch for emphasis. Then medium-length supporting detail. Single-word or fragment emphasis lines scattered throughout.
Signature phrases
- “Here's what's actually happening:”Not used
- “The thing is.”
- “But here's what wasn't working:”
- “This is what we do at”
- “If you run a B2B SaaS”
+ 3 more, all on the page below
Quoted from a real profile, not written for this page.
You already gave a tool your posts. It thought about it and told you that your tone was confident and conversational, which you knew. Then it wrote a post that opened with “Here’s the thing” and closed by asking what you thought. It read like a person. It did not read like you. So you rewrote it, which is the job you were trying to hand over.
An adjective cannot be copied. A sentence can.
“Confident and direct” is not your voice. It is a category that a few million people are in. Hand a model that description and it will write the average of everyone in the category, which is why the output converges no matter whose account produced it.
A voice profile here is not a description. Most of it is quotation. Five to eight phrases you actually reuse, pulled out word for word. Three to five passages from your own posts, copied exactly. Your sentence rhythm written out as a pattern. The model is not asked to interpret how you sound. It is shown how you sound, and it sits at the top of every instruction, above every other rule.
How it works
Give it your profile URL, or two posts.
It reads up to 30 posts you already published. Nothing to fill in, nothing to describe about yourself. If you would rather not connect anything, two pasted posts is the floor.
It writes down what it found, and shows you.
Fifteen observations. The three that matter are quotations, not opinions: your phrases, your passages, your rhythm. Read them. They are recognisably yours or the profile is wrong, and you can tell in ten seconds which it is.
Every post is written against it.
Not as a style preference applied at the end. The voice block is the first thing in the instruction and it outranks everything else, including our own formatting rules.
Reading posts… 20 of 30
Short declarative statement (3-6 words). Follow with longer explanatory sentence (15-20 words). Another short punch for emphasis. Then medium-length supporting detail. Single-word or fragment emphasis lines scattered throughout.
- “The thing is.”
- “But here's what wasn't working:”
We stopped posting three times a week. Here’s what I’ve learned: the third post was never the one people read. The thing is. Nobody noticed it was gone.
One thing it will not do is hand back your worst habit.
If a phrase you genuinely reuse is one of the openers that makes a post read like AI wrote it, the profile keeps the observation, because it is true, and drops it from the instructions. It does not get written into your posts and then quietly deleted again. It never gets asked for.
Here is a real profile
Not an invented one. This is the profile PublishFlow built for its own founder, Shehu AbdulGaniy, from posts already published on LinkedIn, copied out of the product as it stands.
The rhythm field
Real profileShort declarative statement (3-6 words). Follow with longer explanatory sentence (15-20 words). Another short punch for emphasis. Then medium-length supporting detail. Single-word or fragment emphasis lines scattered throughout.
The eight signature phrases it pulled
Real profile- “Here's what's actually happening:”Not used
- “The thing is.”
- “But here's what wasn't working:”
- “This is what we do at”
- “If you run a B2B SaaS”
- “Here's what I've learned:”
- “The shift isn't that content doesn't matter anymore.”
- “Most companies are still optimizing for”
Seven of those eight are handed to the model. The first one is not. “Here’s what’s actually happening:” is a real habit and a bad one, so it stays on the profile as an observation and never becomes an instruction. The profile is a description of how you write, not a promise to reproduce all of it.
What it will not do
The limits, stated plainly, so you can decide before you sign up.
It will not clone your speaking voice.
This is about writing. There is no audio, no text-to-speech, nothing that reads your posts aloud in your accent.
It will not make you sound better than you sound.
It copies habits, and some of your habits are habits. The one category it overrides is the openers that read as machine-written.
It does not keep learning on its own.
Posts you publish are added to the library, but the profile itself only changes when you re-run it. That is a button you press, not a thing that happens behind you.
It cannot read a profile that has nothing on it.
Two real posts is the minimum. If you have never posted, start with a meeting or a voice note instead, and build the profile once there is something of yours to read.
Five voices while your trial is running and on Pro, one after that.
The five exist for people who write as more than one person. If that is not you, one is the whole feature.
It will not tell you who to write for.
Voice is how you sound. Who you are useful to is a separate thing the product asks about separately, and getting the voice right will not fix a post aimed at nobody.
Works with
LinkedIn post generator
Where the voice actually gets used: your own material, turned into posts that sound like the profile above.
The Monday digest
Your voice running on a schedule, so Monday arrives with the week already drafted.
Meeting notetaker
The material side of the same problem. Your calls become something worth writing about.
Voice notes
Say it out loud and get a post back, written in the same voice as everything else.
Questions
How does it learn my voice?+
It reads posts you have already published, up to 30 of them, and writes down fifteen observations about how you write. Three of those are quotations rather than descriptions: the phrases you reuse, passages from your own posts copied word for word, and your sentence rhythm. New posts are written against those.
How many posts does it need?+
Two is the minimum. When it analyses, it reads up to twenty, sampled evenly across everything you have rather than only your most recent posts, so an older post still counts.
Do I have to connect my LinkedIn account?+
No. You can paste your public profile URL and it will read the posts, or paste two posts by hand.
Does it update itself as I post more?+
Posts you publish are added to your sample library automatically. The profile is rebuilt when you ask for it, not continuously, so your voice never shifts under you without you doing something.
Can I have more than one voice?+
Up to five while your trial is running and on Pro, one after the trial ends. One active at a time. People use the others to write as a co-founder or a client.
Is this the same as an AI humanizer?+
No. A humanizer takes text a model wrote and sands off the tells. This starts from your own writing and never generates the generic version in the first place.
Will it use my writing to train anything?+
Your posts are stored on your own profile and used to write your posts. That is the whole of it.
Everyone says it sounds like you. Read the profile and check.
Fourteen days free. No credit card.
Start freeA public profile URL is enough. No password, no export.