AI posts sound generic because voice matching only solves style. Writing that sounds like you still fails if it contains nothing only you know, and if it isn't arguing for anything in particular. There are three layers — voice, substance and strategy — and models trained on your previous posts can imitate the first while leaving the other two empty. The fix is not a better prompt; it is feeding the system your actual raw material.
We ran our AI on a CEO's last fifty posts.
It matched his voice perfectly. Cadence, vocabulary, sentence structure, the way he opened with a short line and landed on a question — identical.
Then he read the output and said:
This sounds like every CEO on LinkedIn.
He was right, and that sentence sent us back to first principles. Because the voice match was not the failure. The voice match was excellent. The failure was everything the voice was wrapped around.
The three layers
| Layer | The question it answers | Can a model do it from your old posts? |
|---|---|---|
| Voice | How do you sound? | Yes — reliably, and quickly |
| Substance | What do you know that nobody else does? | No — it isn't in the text |
| Strategy | Why this post, now, for this audience? | No — it requires intent |
Voice is the layer everyone sells, because it is the layer that demos well. Paste in fifty posts, get back something that sounds convincingly like the author. It is genuinely impressive and it is genuinely the easy part.
Substance is the layer that makes a post worth reading. It is the number that surprised you last week, the decision you reversed, the thing a customer said on a call that changed your mind. None of that exists in your published archive — by definition, it is the material you have not written about yet. A model trained on your old posts can only recombine positions you have already taken.
Strategy is the layer that makes a post worth publishing. Why this argument, this week, to these people, laddering to what you want your market to believe. Without it you get a stream of individually reasonable posts that accumulate into nothing.
Miss either of the last two and you get exactly what that CEO got: flawless imitation of a man saying nothing in particular.
Why "every CEO on LinkedIn" is the specific failure
There is a reason the generic output converges on a recognisable house style.
Take any executive's public posts. Strip out the proprietary detail — because most executives are careful, and most of their posts are careful. What remains is the shared vocabulary of the category: alignment, resilience, lessons learned, grateful to the team. Train on that and you reproduce the average of it.
The uniqueness of a leader is not in how they write. It is in the intersection of this company × this leader × this industry — a combination nobody else occupies. That intersection lives in the work, not in the archive.
Which is why the fix is not prompt engineering. You cannot prompt your way to information the system does not have.
The evidence that generic doesn't just feel bad — it performs worse
This matters commercially, not just aesthetically.
In controlled testing of whether AI-generated content could lift search performance, the AI-generated material produced worse results than original content. And there is now more AI-generated content online than original content, which means the discount is being applied at scale.
The pattern is familiar to anyone who watched early SEO. People duplicated millions of pages stuffed with keywords and backlinks. It worked for a while. Then it was cut out in a single update.
What holds up is AI-assisted rather than AI-generated: your ideas, your evidence, your unusual angle, with the machine handling drafting and refinement. The authenticity has to originate with a person, because that is the only part that cannot be reproduced at zero marginal cost.
What actually fixes it
The practical shift is from imitating your output to capturing your input.
- Feed it raw material, not finished posts. Call transcripts, voice memos on the drive home, the message you sent a co-founder at midnight, the commit you just pushed. That is where substance lives.
- Start from a specific artefact, not a topic. "Write about hiring" produces the average of all hiring content. "Here's what happened when I made an offer to someone who'd already turned us down once" produces one post that only you can write.
- Give it the strategy explicitly. What you want to be known for, who you are talking to, what you have already argued. If the system does not remember your positioning between sessions, it will re-derive a generic one each time.
- Keep an editorial gate. The final judgement about what is true, what is yours to say, and what is too early to say has to stay with you. That is not a compliance step — it is the part that makes the post worth anything.
- Test the sentence. If a competitor could publish this post with their name on it and nobody would notice, it is not finished.
That last test is the whole thing in one line. Generic writing is not a style problem. It is the absence of a claim only you can make.