AI LinkedIn Ghostwriting: What AI Should Write, and What a Human Must

AI LinkedIn Ghostwriting: What AI Writes, What Humans Must

George Khoshtaria
George KhoshtariaAI LinkedIn Ghostwriting: What AI Writes, What Humans Must
Founder of rethinkGTM
LinkedIn GrowthRead time 8 minPublished November 6, 2026
Key takeaways
81.2% of 5,000 long LinkedIn posts sampled in July 2026 were classified as likely AI-generated, up from about half in late 2024.
AI should handle research, idea mining, first drafts, repurposing, layouts, and analytics; a human must own the positioning, the angle, the real number, the hook, and the final edit.
The hybrid workflow behind a founder profile that grew from 4,000 to 10,000 followers in 4 months produces a finished post in 30 to 40 minutes.
The inputs make the difference: sales calls, client threads, audits, and competitor data give AI something no other account has.
Voice rules belong in a written skill that every AI draft must pass, so the founder's patterns survive at volume.

AI LinkedIn ghostwriting works when AI does the heavy lifting and a human owns the judgment. AI researches, mines ideas from calls and Slack, drafts, and lays out visuals. A human chooses the angle, supplies the real number, writes the hook, and makes the final edit. Skip the human half and you get the generic posts buyers scroll past.

The feed already shows the problem. Originality.ai classified 81.2% of 5,000 LinkedIn posts of 100+ words, sampled in July 2026, as likely AI-generated, up from about half in its late-2024 study (Originality.ai). When most posts come from the same models with the same prompts, they start to sound the same, and readers notice.

I use AI every day. It's how one person can run a daily founder LinkedIn program at 30 to 40 minutes per post. This guide shows exactly where AI sits in that workflow, where it doesn't, and the rules that keep the posts sounding like the founder.

Can AI write your LinkedIn posts?

Yes, AI can write a LinkedIn post that reads fine. It can't decide what the post should say, because it doesn't know your buyers, your deals, or what happened in your business this week. That's the part buyers pay attention to.

The 2025 Edelman-LinkedIn B2B Thought Leadership Impact Report shows what buyers want: 91% of hidden buyers want insights that reveal unseen risks or opportunities, and 86% prefer perspectives that challenge their assumptions (Demand Gen Report). A model trained on the public internet averages what everyone has already said. A challenging, specific insight has to come from your own work.

How I see it

I wrote this in a LinkedIn post earlier this year, and it still holds. The feed is flooded with AI-generated content, and the only thing that survives it is what nobody else has: a real sales call, a real competitor teardown, an industry insight from inside your accounts. AI is a great engine. Your experience is the fuel.

What AI should write, and what a human must

Task

AI

Human

Why

Mining ideas from calls, Slack, and client threads

✅

Reviews the list

AI reads everything; a person can't

Research and fact gathering

✅

Checks sources

Fast, but it can be wrong

Competitor and trend tracking

✅

Picks what matters

Pattern-finding at volume

Choosing the angle and funnel stage


✅

Needs positioning and goals

The real number, client, or story


✅

Only the business has it

The hook (first two lines)

Drafts options

✅ Final pick

The hook decides the reach

First draft of the body

✅

Edits

Saves most of the writing time

Voice check against the founder's patterns

✅ via a skill

✅ Final read

Rules catch drift, a person catches tone

Infographic and carousel layout

✅ from a design system

Approves

Keeps every visual on brand

Performance analysis

✅

Decides next month's plan

Data in, judgment out

Comments on other people's posts

Drafts in voice

✅ Reviews and adapts

Relationships need care

The hybrid workflow I run for founder LinkedIn

This is the workflow behind Olifant Digital's founder program, where the profile grew from 4,000 to 10,000 followers in 4 months at 6 posts a week.

Step 1: Write the founder's voice into a skill

I ran several hour-long calls with the founder and read how he explains things in Slack. Then I wrote the patterns into a Claude skill: lead with the result, name the client and the number, plain words only, short sentences, no em dashes, one call to action. Every AI draft passes through that skill. Without it, every draft drifts back toward the model's default voice.

Step 2: Let an AI agent mine the ideas

Viktor, an AI agent, reads the team's daily Slack and client threads under strict rules and the content skills. It drops a weekly list of post ideas into Asana and tags me. The AI finds what happened. I decide which of those things is worth a post.

Step 3: Build one brief from four inputs

For my own content, I feed four inputs into Claude: the keyword gap from Ahrefs, competitor research from Perplexity, what competitors already shipped (scraped with Apify), and the exact words a prospect used on a recent call (from Granola). Claude, with trained skills, turns the four into a single brief. The brief carries the angle, so the draft starts from something only my business knows.

Step 4: Draft with AI, then rewrite the hook myself

AI writes the first draft and several hook options from the brief. I rewrite the hook almost every time. The hook decides whether anyone reads the rest, and it needs the one specific number or claim that no competitor could post.

Step 5: Check the draft against the hooks library

I keep a library of high-performing hooks and infographics from the niche, turned into a Claude skill. Before a post goes out, the draft gets checked against those patterns: is the first line specific, does the second line promise something, does the body deliver it.

Step 6: Design from a fixed system

Infographics and carousels come from a Claude design system with fixed colors, fonts, and layouts. Complex visuals go to a designer with a written brief. Carousels and PDF audits earned a median of 181 engagements on the founder's profile, against 99 for text and image posts, so the design step earns its time.

Step 7: Let data pick next month's topics

Two in-house apps close the loop. A competitor watcher (Apify's MCP plus Claude) flags which competitor posts take off. An analytics dashboard (Apify, Claude, and Supabase) scores every post by topic and format. Next month's plan leans into what worked.

How to keep AI LinkedIn posts from sounding like AI

  1. Start from a real input. A call transcript, an audit, a Slack thread. A blank prompt produces a generic post.

  2. Put one specific number in the first two lines. "Revenue per visitor rose 10.87%" can't come from a model's imagination.

  3. Write the voice rules down. Banned words, sentence length, punctuation, how the founder opens and closes. Turn them into a skill the draft must pass.

  4. Ban the AI tells. Filler openers about "today's world", hype words like "game-changer", em dashes everywhere, and three-adjective lists.

  5. Read it out loud. If the founder wouldn't say a line on a call, rewrite it.

  6. Never let AI invent proof. Every number, client, and result gets checked against a real source before the post goes out.

  7. Keep one human in the loop for comments. AI can draft a comment in the founder's voice. A person should still read every one, because comments build relationships one at a time.

AI tool or human ghostwriter: which do you need?

Option

Best for

Watch out for

AI writing tool alone

Founders who enjoy writing and want a faster first draft

Generic posts if you skip the inputs and the edit

Human ghostwriter, no AI

Low volume, highly personal posts

Slow and expensive at daily cadence

Hybrid: human strategist with AI workflow

Founders who want daily, specific posts without spending their own time

Ask to see the voice rules and the review process

For what a human ghostwriting program includes, read what LinkedIn ghostwriting is, and for how to vet a provider, how to choose a LinkedIn ghostwriting agency.

How I see it

The question "AI or human?" misses where the value is. The value is in the inputs and the judgment. An AI-only setup with great inputs beats a human writer with none, and a human strategist with an AI workflow beats both. When you hire anyone for LinkedIn, ask them to show you the inputs they'll use and the rules every draft must pass. If the answer is "we'll interview you once a month", expect generic posts.

Frequently asked questions

Can AI write LinkedIn posts for B2B founders?

Yes, for research, ideas, first drafts, and layouts. A human still needs to choose the angle, supply the real numbers and stories, write the final hook, and edit for voice. Without those, AI posts sound like the 81.2% of long posts already flagged as likely AI-generated.

How many LinkedIn posts are AI-generated?

Originality.ai classified 81.2% of 5,000 LinkedIn posts of 100+ words, sampled in July 2026, as likely AI-generated. Its late-2024 study had found about half of long-form posts were likely AI-generated.

What is the best AI tool for LinkedIn ghostwriting?

The tool matters less than the setup. George Khoshtaria uses Claude with trained skills for voice, hooks, and design, plus Viktor for idea mining, Apify for competitor and performance data, and Supabase for the analytics dashboard.

How do you make AI LinkedIn posts sound human?

Start from a real input, put a specific number in the first two lines, write the founder's voice rules into a skill, ban common AI phrases, read every draft out loud, and never let AI invent a result or a client.

Is AI ghostwriting better than a human LinkedIn ghostwriter?

A hybrid beats both. AI is faster at research and drafting; a human is better at positioning, hooks, and judgment. A human strategist running an AI workflow can post daily at 30 to 40 minutes per post without losing the founder's voice.

Should LinkedIn comments be written by AI?

AI can draft comments in the founder's voice from a skill, which saves time when you comment every day. A person should review and adapt each one, because comments build relationships one person at a time.

Will LinkedIn penalize AI-generated posts?

LinkedIn ranks posts on how people respond, and generic posts earn weak responses. The bigger risk is a post nobody stops for. Specific, experience-based posts perform better however the draft was produced.

About the author

George Khoshtaria, founder of rethinkGTM, is a B2B marketing consultant and fractional CMO with 8+ years in B2B marketing. He builds AI content workflows with Claude and runs founder LinkedIn programs, including Olifant Digital's, from his base in Tbilisi. He writes his own LinkedIn profile of 20,000+ followers as a LinkedIn Verified Creator. Read more about George Khoshtaria, or see the B2B LinkedIn marketing service.

Sources

Subscribe to 1% GTM Insights

Abstract wallpaper

Looking for more weekly tips on B2B marketing?

Follow George on LinkedIn (20K+)