How to Build an AI Content System for LinkedIn That Still Sounds Like You

"Your AI-assisted content does not sound like you" is one of the most common complaints I hear from experts who have tried to post consistently. The content is fluent, structured and completely forgettable. The problem is not the AI. The problem is asking a tool to make decisions it cannot make — what you believe, which story is yours, what your audience actually struggles with — and then publishing the first thing it produces.

Here is the system I run for myself and for ghostwriting clients. Human-led, AI-assisted, in that order.

Inputs the system cannot run without

Before any prompt, three documents must exist. Skip them and every downstream stage produces generic output.

• Positioning statement. Who you help, with what, why you, what next. If you do not have one, start with the growth strategy guide.
• Four content pillars. The topics you want to be known for, each with a one-line description of the angle you take.
• Voice file. One page: ten phrases you actually use, five opinions you hold strongly, five stories you can tell, words you never use, your preferred sentence length, how you open and close posts. Paste three of your best real posts at the bottom. This file goes into every drafting prompt.

Stage 1 — Research with AI as an analyst

Use an assistant to cluster, not to invent. Feed it real raw material: comments on your posts, questions clients asked this month, competitor post themes, the language on review sites in your niche. Ask for themes, recurring objections, and the exact phrases people use. Verify anything factual. An answer engine with citations is useful here; a chat model making claims from memory is not.

Stage 2 — The idea bank

An idea bank is a running list of angles, each tagged to a pillar. Aim to keep 40 or more live. Sources: the research above, your own week (a client question, a mistake you saw, a tool that surprised you), and a monthly 45-minute "idea call" — with yourself or, for clients, with me — where stories and opinions are captured verbatim.

Each entry is one line: pillar · angle · the real story or data behind it. The "real story" column is what keeps the content yours.

Stage 3 — The brief

Never prompt "write a LinkedIn post about X". Write a brief:

• Audience and the one problem this post addresses
• The point of view (what you believe that others in your field do not)
• The story or proof, in your own words — even roughly
• Format and length (text post ~180 words; carousel 8 slides; newsletter section 150 words)
• Hook style to try (contrarian, story, list, mistake, data, question)
• The call to action, if any
• Attach the voice file

A good brief takes four minutes and is the single biggest determinant of output quality.

Stage 4 — Draft in angles, then choose

Ask for three to five versions with different hooks and structures, not one. Compare them side by side. Usually the best post is the second half of one draft stitched to the opening of another. This is where AI earns its keep: speed and variety. The choosing is yours.

For carousels, ask for the slide structure first (cover, problem, 5 steps, summary, CTA), approve it, then ask for slide copy. For newsletter sections, draft the headline and the one-sentence takeaway before the body.

Stage 5 — The human edit (three passes)

1. Truth pass. Remove any claim, number or example you cannot stand behind. Replace generic examples with your real ones.
2. Voice pass. Read it aloud. Cut phrases you would never say. Add the words you actually use. Shorten sentences until it sounds like you talking, not writing.
3. Reader pass. First two lines: would a stranger stop? Last line: do they know what to do next? Delete anything the reader does not need.

The final test: would you say this to a client across a table? If not, it is not ready.

Stage 6 — Publish and learn

Publish at consistent times, reply to every comment in the first hour, and once a month review what worked: which pillars, which hook styles, which formats drove profile views and messages — not just impressions. Feed the winners back into the idea bank and the voice file. The system should get more like you over time, not less.

The weekly rhythm

Around three hours a week, and the content sounds like you because the decisions were yours. If even that is more than your week allows, this is the exact system behind my ghostwriting packages — and the tools I use for each stage are on the tools page.

Frequently asked questions

Will LinkedIn penalise AI-generated content?

LinkedIn does not penalise content for being AI-assisted; it penalises content people scroll past. Generic AI text performs badly because it is generic, not because it is detected. A human-led system fixes the generic part.

Which AI tool is best for LinkedIn content?

Any strong assistant works — Claude, ChatGPT or Perplexity for research. The tool matters less than the inputs: your positioning, your stories, your voice samples and a clear brief.

How much time does this system take each week?

About three hours: 45 minutes of idea capture, 90 minutes of drafting and editing in one batch, and 15 minutes a day of engagement.

Can I run this system for clients?

Yes — it is the same system I run for ghostwriting clients. The voice interview and idea call are the parts you cannot skip.