← Back to Blog

AI LinkedIn Ghostwriting: Build a Voice System for a Client Instead of Selling Posts

A former executive ghostwriter on this site lost his clients to AI, then rebuilt by selling the knowledge base behind the writing. Here is how to build that system for one client this week, and what it costs.

Strategy9 min read•

Your client can already get a LinkedIn post from ChatGPT

That is the problem with selling LinkedIn posts in 2026. A founder who pays a ghostwriter every month for eight posts can open a chatbot and get eight posts in a minute. They will be bland, but they will be free, and plenty of clients decide bland and free is good enough.

David Johnson-Igra, documented on this site, lived through exactly this. A tech ghostwriter whose clients included Amazon, a16z, Meta, GitHub and OpenAI, he lost his client base in April 2025, according to Fortune's May 2026 coverage. He rebuilt by selling something a chatbot cannot produce on its own: an organized record of everything an executive has said, an analysis of what has worked for them, and a drafting setup that writes from that record. In his words, as quoted in the coverage, the outputs are now a means to an end, and the end is the system. He has not disclosed revenue or client counts, so this guide copies his method, not a promised result.

The voice system: what you build once, what runs weekly 1. Archive Interviews, posts Talks, memos Tagged in Obsidian 2. Evidence Post analytics Peer content audit Gaps to own 3. Draft Short brief Claude reads archive Cites past quotes 4. Approve Client edits Client posts No bots Steps 1 and 2 are the product you sell. Step 3 is cheap and anyone can do it. Step 4 keeps the client's name, and their LinkedIn account, safe.

What the voice system is

The case study on this site, David Johnson-Igra Lost All His Ghostwriting Clients to AI in 2025, describes four parts.

The archive is every interview, article, LinkedIn post and public statement the executive has made, stored as notes in Obsidian and tagged by topic, tone and the context it came from. Obsidian is plain files on disk, so the archive belongs to the client and works with whichever AI model you use next year.

The performance analysis is a Python script that reads the executive's full post history, sometimes about 4,000 rows, and measures which topics, formats and angles earned engagement from their followers.

The competitor audit pulls peer executives' content, up to 1,000 pieces across 10 executives in the documented project, to find angles that are crowded and gaps the client can own.

The drafting setup connects Claude to the archive through MCP, so when you give it a brief (topic, platform, goal, tone notes) it drafts using the executive's own past positions and phrases instead of starting cold. The executive then reviews and approves each draft.

Why this is worth more than posts

A post is worth something for a week. The archive and the analysis are worth something as long as the client keeps talking in public, and they get better with every interview you add. They also answer the question every client eventually asks, which is why should I pay you when I have ChatGPT: because the drafts come from their own words, their best-performing topics and a gap their peers are not covering. That is also the defense against the worst failure of AI ghostwriting, which is a post that confidently states an opinion the client does not hold.

A worked example: a 40-person software founder

Say you take on the founder of a 40-person software company who has done 14 podcast interviews, given six conference talks and written about 300 LinkedIn posts over four years. In week one you collect the transcripts and posts, about 60 documents once you drop duplicates, and turn each into an Obsidian note with a topic tag (hiring, pricing, product), a tone tag (contrarian, story, how-to) and the source and date.

In week two you export a year of post analytics from LinkedIn and ask Claude to group posts by topic and format and rank them by engagement rate. Suppose it shows that hiring stories outperform product announcements several times over. You then read 10 posts from each of five peer founders and note what they all say about hiring. Your positioning note for the client says: talk about hiring mistakes with numbers, which peers avoid, and stop announcing features.

From week three on, the routine is a 15-minute call each Monday, three briefs, three drafts by Tuesday that quote or paraphrase the founder's own past statements, and edits back by Wednesday. The founder posts them. This example is an illustration of the method, not a documented client.

Step by step: build it for one client

Step 1: Agree on who posts and who approves

Before collecting anything, agree in writing that the client approves every post and posts it from their own account, or schedules it with LinkedIn's own scheduler. LinkedIn's help center says it does not permit third-party software, including bots and browser extensions, that automates activity, and its User Agreement bars using bots or unauthorized automated methods to create, comment on, like or share posts. Do not hand a client's login to a tool that posts on its own.

Step 2: Build the archive in Obsidian

Install Obsidian, which its pricing page lists as free without limits, with an optional $50 per user per year commercial license. Create one folder per client. Make one note per source with three tags at the top: topic, tone and context. Paste the transcript or post text in full. Ask the client for unpublished material too: internal memos, old decks, a voice memo of them ranting about their industry. That material is often where the best positions are.

Step 3: Pull a year of post analytics

Have the client open their LinkedIn creator analytics and use the Export button, which LinkedIn's help says produces an XLSX file. Check the date picker for how far back the export goes, since sources disagree on the window. Upload the file to Claude with this prompt: group these posts by topic and by format (story, list, opinion, announcement), give the median engagement rate for each group, and list the top ten posts with one line on what they have in common. Check two or three numbers against the spreadsheet yourself, because a model can miscount rows.

Step 4: Audit five to ten peers by hand first

Pick five to ten people your client's buyers also follow. Read their last ten posts each and note the topic and the position. Ask Claude to summarize your notes into crowded angles and open angles. Do this from what you read, not from scraping: automated collection of LinkedIn content runs into the same rules as automated posting.

Step 5: Connect Claude to the archive

The simple route is a Claude project per client with the archive notes and the positioning note uploaded as project knowledge. The documented route is Claude's desktop app with a filesystem connector pointed at the client's Obsidian folder, so Claude reads notes directly through MCP and the archive stays current without re-uploading. Either way, add project instructions: write in the client's voice, use only positions found in the archive, name the note each claim comes from, and say so when the archive has nothing on a topic.

Step 6: Draft from a four-line brief

For each post write the topic, the goal (hire, sell, be known for something), the format and one tone note. Ask for two drafts and the source notes behind each claim. Edit for rhythm, check every claim against its source note, and send one version to the client. If Claude cites a note that does not say what it claims, rewrite that line yourself.

Step 7: Update the archive every month

Add every new interview and approved post as a note, and rerun the analytics review each quarter. The system only stays better than a blank chatbot if it keeps getting fed.

What this costs

As of November 2026: Obsidian is free, with Sync at $4 a month per user billed annually if you want the vault on several devices. Claude Pro is $20 a month or $17 billed annually, and Max is $100 a month for heavy use across many clients. Claude Team starts at $20 per seat a month billed annually. LinkedIn's creator analytics export is free. The listing on this site puts startup cost at $50 to $200 a month, which matches a Claude plan plus Obsidian Sync and a commercial license for a small agency.

Time is the real cost. Expect 10 to 15 hours to build the first client's archive and analysis, then about two hours a week per client for briefs, drafts and edits. Those are planning estimates, not documented figures.

When not to do this

Skip it for a client with almost no public record. With five old posts and no interviews, there is nothing for the system to draw on, and you are back to writing from scratch with extra steps. Skip it if the client will not review drafts: an executive who wants posts to appear without reading them will eventually publish something they disagree with under their own name. And do not sell it as fully automated. The whole value is that a person who knows the client checks every line.

Frequently Asked Questions

Is AI LinkedIn ghostwriting allowed?▾

Writing posts with AI help is not the issue. LinkedIn prohibits bots and unauthorized automated methods that create, comment on, like or share posts, so the client should post or schedule each one themselves through LinkedIn rather than through a tool that posts on its own.

How do I make AI-written LinkedIn posts sound like my client?▾

Give the model the client's own words. An archive of their interviews, talks and past posts, tagged by topic and tone, lets Claude draft from positions they actually hold, and an instruction to cite the source note for each claim makes it easy to catch invented opinions.

What should a LinkedIn ghostwriter charge in the AI era?▾

There is no reliable public benchmark, and this site has not verified any. The ghostwriter documented here moved from charging for writing to selling the knowledge system behind it, which is easier to price as a setup fee plus a monthly retainer than per post.

Can I export LinkedIn post analytics?▾

Yes. LinkedIn's help says creator analytics have an Export button that produces an XLSX file. Check the date range in your account, since third-party guides disagree on how far back it goes.

Do I need Obsidian for this?▾

No. Any folder of text files works, and a Claude project with uploaded notes is the simplest start. Obsidian helps because it is free, stores plain files the client owns and makes tagging and linking notes quick.

Should I use automation tools to post on LinkedIn for clients?▾

No. LinkedIn says accounts using third-party automation risk restriction or shutdown. Use LinkedIn's own scheduler or have the client post, and keep AI on the drafting side.

Keep Reading

Explore Related Niches

Related Strategies

Dev ToolsMarketing AgenciesInternal Tool

Seven AI agents ran a company on a single VPS for a week. They pivoted the business, shipped an MVP with 158 passing tests, then forgot basic housekeeping.

AdvancedUnder $50/moLow (under $2K/mo)
View Strategy
Customer ServiceMarketing AgenciesFreelance/Agency

A full live build of a customer service chatbot with knowledge base retrieval, lead capture, and booking integration, sold for $3,500.

Intermediate$50-200/moMedium ($2K-10K/mo)
View Strategy
Workflow AutomationMarketing AgenciesInternal Tool

Twelve n8n automations eliminated $3,000 per month in manual tasks covering lead management, onboarding, newsletters, and server monitoring.

IntermediateUnder $50/moMedium ($2K-10K/mo)
View Strategy

Keep Exploring the Directory

Browse more case studies, compare tools, or dig into the full tool stack behind these workflows.

Want more strategies like this?

Get weekly AI agent case studies in your inbox.