Content
How to Use AI for LinkedIn Content Without Sounding Like AI
AI writes the average of everything it has read, which is why most AI LinkedIn content sounds identical. Here is where to use it, where it destroys your edge, and the workflow we actually run.
By Niall Ratcliffe, Co-founder, noticed. · 7 min read
Published October 23, 2026
Key takeaways
- AI writes the average of everything it has read. That is why so much LinkedIn content now sounds identical, and why the average is the one thing you cannot afford to sound like.
- Use AI for the parts with a right answer: structure, editing, repurposing, research. Not for the opinion, the story or the number, which are the only parts a reader cannot get elsewhere.
- Your edge is what only exists inside your head and your business. A model has no access to either, so anything it can produce unaided is by definition not differentiated.
- Start from a transcript, not a prompt. Dictate for five minutes and have AI structure it, and the result keeps your thinking while losing the rambling.
- The tell is abstraction. Specific numbers, named clients and real constraints are what AI cannot invent and what readers actually stop for.
Most AI-written LinkedIn content sounds the same, and there is a straightforward reason for it.
AI writes the average of everything it has read. That is not a flaw in the tool, it is the tool working correctly. The problem is that on a platform where you are trying to be remembered, the average is the single worst thing you can sound like.
This is not an argument against using AI. We use it daily. It is an argument about which parts of the job it should touch.
The split that matters
Every piece of content divides into two things: the parts with a right answer, and the parts that are only yours.
| Give it to AI | Keep it human |
|---|---|
| Structure and ordering | The opinion |
| Tightening and editing | The story that happened to you |
| Repurposing one idea into formats | The numbers from your own business |
| Research and summarising | What you think everyone else is wrong about |
| Headline variations to choose between | The client example with a name attached |
The left column has right answers. A model is often better than you at them, and certainly faster.
The right column is the only reason anyone reads your post rather than any other post on the same subject. A model has no access to what happened in your business last Tuesday, what your client said on the call, or what you believe that your industry considers wrong.
Your edge is what only exists inside your head and inside your business. Anything an AI can produce without you is, by definition, not differentiated. So the question is never "can AI write this?" It is "would this be worth reading if anyone could have written it?"
Why generic content costs you reach, not just credibility
There is a mechanical problem as well as an aesthetic one.
LinkedIn tests a post on a small slice of your network and expands distribution based on how that slice responds. Generic content gets weak early engagement, which caps the reach of that post, and teaches the system over time that your posts are not worth pushing.
So publishing more AI-written filler does not just fail to work. It actively reduces the reach of the posts you put real effort into.
The workflow we actually run
1. Start from a transcript, not a prompt
This is the single change that fixes most of the problem.
Open a voice recorder and talk for five minutes about the thing you want to say. Do not script it. Ramble. Then transcribe it and ask AI to structure what is there without adding anything.
What comes back keeps your vocabulary, your examples and your way of arguing, because all of it was in the transcript. What it loses is the rambling, which is exactly what you wanted removed.
Compare that with prompting from a blank page, where the model has nothing of yours to work with and has no option but to produce the average.
2. Put the specifics back by hand
AI will smooth your specifics into categories. "A client" where you said "a fabricator in Widnes". "Significant growth" where you said "two million impressions and 150 site visits from decision-makers".
Go back through and restore every one. The specifics are the content; the connective tissue between them is not.
3. Use it hardest for repurposing
This is where AI genuinely earns its place. One good idea, properly thought through, becomes a post, a carousel outline, a newsletter section and three short video hooks. The thinking happened once and was yours. The reformatting is mechanical, and a model does it faster than you will.
4. Read it out loud before posting
The fastest test available. If it sounds like something you would never say in a meeting, it will read that way too. Most AI tells disappear the moment you edit for how you actually speak.
How readers spot it
Not through any single word. Through abstraction.
AI-written content describes categories: "companies often struggle with alignment between sales and marketing". Human-written content describes instances: "a client last month had four people who each assumed someone else owned the number".
The second one could only have been written by someone who was there. That is the whole difference, and it is also why the advice here collapses into a single rule: AI can shape what you know, but it cannot know it for you.
Why this is an opportunity, not a threat
As more people publish the same generic material, the bar for standing out drops rather than rises.
If most of the feed is the average of everything written about your subject, then one post containing a real number, a named client and an opinion someone might disagree with is unusual enough to stop a reader. That was always true. It is just more true now.
For what to do with that, our content shortlist covers the agencies running this for companies, and the DM playbook covers the outreach side.
Frequently asked questions
- Can you use AI to write LinkedIn posts?
- Yes, for structure, editing and repurposing. Not for the opinion or the specifics. A model writes the average of what it has read, so anything it produces unaided will sound like everything else in the feed.
- Does LinkedIn penalise AI-generated content?
- There is no evidence of a direct penalty, but the effect is the same. Generic content gets weak early engagement, and because the algorithm expands distribution based on how the first readers respond, mediocre content suppresses its own reach.
- How do you make AI writing sound like you?
- Give it your material rather than a prompt. Dictate your thinking for five minutes, transcribe it, and ask AI to structure what is there without adding anything. The voice survives because it was yours to begin with.
- What should you never use AI for in LinkedIn content?
- The opinion, the story and the numbers. Those are the only parts a reader cannot get from any other post on the subject, and a model has no access to what happened inside your business.
- How can readers tell content is AI-written?
- Abstraction. AI-written posts describe categories rather than instances: "companies often struggle with alignment" instead of "a client last month had four people who all thought someone else owned the number". Specifics are the tell in both directions.
- Is AI making LinkedIn content worse?
- It is making the average worse, which makes anything specific stand out more than it used to. The opportunity gets larger as more people publish the same generic material.
- What is the best AI workflow for LinkedIn content?
- Record your thinking, transcribe it, use AI to structure and tighten, then put the specifics back in by hand. The sequence matters: structure after substance, never instead of it.
About the author
Niall Ratcliffe, Co-founder, noticed.
Niall co-founded noticed. in 2022 and has since run LinkedIn content, outreach and strategy programmes for more than 200 B2B companies, from the NHS to Hutchinson Engineering.
AI Content, LinkedIn Content, AI Marketing, B2B Content