LinkedIn AI Assistant Case Study: 8 to 35 Posts in 90 Days

Mark Smith, a consultant who writes about AI, CRM, and consulting on LinkedIn, went from 8 LinkedIn posts in 90 days to 35 in the next 90, about 4.4 times as many. The numbers come from his own activity brief, dated October 28, 2026.

He did it by connecting his AI assistant to FeedBoss, a LinkedIn content platform that learns your voice from your posts, keeps your content pillars in an Authority Map, designs images and carousels, and schedules your posts. Twice a week, the assistant finds sourced ideas, and the drafts follow his FeedBoss voice profile. He reviews, edits, and schedules every post himself.

This case study pulls out the seven practices behind that result, so you can borrow them for your own setup, along with the prompt he started with.

Results: 8 to 35 LinkedIn Posts in 90 Days

His goal is three posts a week. In the latest 90 days, he averaged 2.7.

"My goal of posting 3 times a week is just so much easier with FeedBoss!"

The Problem: AI Kept Repeating the Same Ideas

Before FeedBoss, he tried several AI writing tools that promised to write in his voice. They kept suggesting the same angles:

"They kept serving up the same angle on repeat. My interests have more than one facet. Those tools never did them justice."

He writes about six content pillars: AI in consulting and delivery, the consultant and business analyst role, CRM and business transformation, AI governance, requirements and discovery, and building in public. He needed a workflow that keeps all six covered, finds real sources, and turns good ideas into posts he only has to review.

What FeedBoss Adds to ChatGPT or Claude

ChatGPT and Claude can research and draft. On their own, they don't see your LinkedIn posts, your drafts, or your content pillars, and they can't schedule anything to LinkedIn. FeedBoss supplies that part, which is why every post in this workflow runs through FeedBoss. (For a side-by-side, see FeedBoss vs ChatGPT.)

• Authority Map: his content pillars and goals live here, so he can see which topics he has neglected and avoid posting "five times in a row on the same hobbyhorse."
• Voice profile: FeedBoss learns your voice from your own posts, so drafts sound like you, not like a generic corporate announcement.
• Image Studio and Video Studio: visuals and carousels are made inside FeedBoss. His October 29 carousel, "10 AI teammates. Then a Master Control.", came from Image Studio, and he called it "absolutely STUNNING!"
• First comment: written together with the post, usually a link to the original source or to an earlier post on the same topic. The feature started as his request, and the FeedBoss team shipped it within a few days.
• Analytics: see which posts and formats get the most engagement. A connected assistant can read your analytics and top posts too.
• Content calendar: every post is scheduled from FeedBoss, after he has reviewed it.

In his words, FeedBoss became:

"The calm place where my posts get finished."

How the AI-Powered LinkedIn Workflow Works

Mark's AI assistant, Grokbot, is a set of specialized bots. Its LinkedIn bot, Network, connects to FeedBoss through the FeedBoss MCP connector, which lets it work with his posts, drafts, and voice profile. The workflow has two stages:

1. Find ideas twice a week. Every Sunday and Wednesday morning, Network checks his recent posts against his pillars, researches fresh ideas, and tags each one with its pillar and a source link.
2. Turn picked ideas into drafts. The ideas land in a Network Brief, a Google Doc with a checkbox for every idea, next to a briefing page with charts. A tick tells the assistant to prepare that idea as a FeedBoss draft, which he then reviews and schedules.

At first, the ideas sat in a chat and got lost. The briefing page and the brief fixed that, and their charts are what let the strategy correct itself.

7 Best Practices for Posting on LinkedIn With an AI Assistant

These are the habits behind the jump from 8 posts to 35. The first four make the strategy self-correcting: every brief measures what happened and steers the next round of ideas. The last three keep the posts accurate and keep you in control.

1. Check Your Pillar Balance Before You Ask for Ideas

Before it suggests anything, the assistant counts recent posts per pillar, including scheduled posts and drafts. Then the prompt asks for 4 to 5 ideas for each least-covered pillar and fewer for the pillars already covered, so the thin ones get attention first.

The October 28 brief flagged two pillars as thin: AI in consulting and delivery, and building in public. All three ideas he picked from that brief came from those two pillars: AC5 and AC6 from AI in consulting, and BP2 from building in public.

2. Keep a Pool of Ideas That Carries Over

Ideas he doesn't use aren't thrown away. They carry over to the next brief, and every run adds fresh ones, with the prompt asking for more on the pillars that need to catch up. Before a new brief is created, the old one is saved as a dated copy.

More than half of the ideas in that brief came from the brief before. A good idea doesn't expire just because it wasn't the right week.

3. Let Your Results Choose the Format

The brief tracks average interactions per post for each format, and the assistant uses that to suggest post types. In his posts, carousels win by a wide margin, with almost four times the interactions of a text post.

His carousels are designed in FeedBoss Image Studio, and FeedBoss Analytics breaks your engagement down by post type, so you can spot your best format without building a chart yourself.

4. Learn From Your Top Posts

The brief also lists his five best posts by interactions, so new ideas build on what has already landed, not on guesswork. On October 28, a post on Copilot product naming led the list, and a carousel on the hidden cost of good meetings was among the five. A connected assistant can pull your own top posts straight from FeedBoss.

5. Use Real Sources and Write Down Your Guardrails

Every idea needs a source. Have the assistant search your own saved reading first (he uses Readwise and Fabric), then FeedBoss popular posts and the web, link each source, and flag any statistic it can't trace. When a figure is one person's impression rather than a measurement, the draft adds a "To be precise:" caveat.

The prompt also spells out what never goes into a post: confidential client details, private topics, and figures, quotes, or sources that weren't actually found. When in doubt, the assistant leaves it out and flags it.

6. Pick Ideas With a Tick, Not a Chat Command

Typing "draft number 3" into a chat works, but it's clunky. A checkbox in a doc is faster, works from a phone, and leaves a record. Every idea in the brief has a short code for its pillar, so AC5 is the fifth idea for AI in consulting and delivery.

Here is what happens after a tick:

1. Tick an idea: tick a checkbox in the Network Brief and tell Network to check the brief.
2. Add a review task: the idea becomes a Todoist task due today, after a check that it isn't there already.
3. Choose a format: carousel, video, article, image, or text. Text can be English only, or English with German below.
4. Choose the first comment: a link to the source or to an earlier post.
5. Draft in his voice: the draft follows his FeedBoss voice profile, with a strong hook, short paragraphs, a closing question, and up to three hashtags.
6. Make the visual and save the draft: carousels come in his navy and neon style, and images come from FeedBoss Image Studio. Then it saves the post to FeedBoss as a draft.
7. Review and schedule: he checks the draft, makes any edits, and schedules it himself.

7. Let the Assistant Draft, but Schedule Every Post Yourself

This is the one rule in the workflow that never changes:

"The bot never schedules. The bot never publishes. I do. Every time. The bot does the legwork. I keep my name."

It is written into the prompt. The assistant researches, suggests ideas, prepares posts, and saves drafts. He reviews and schedules everything. FeedBoss backs the rule: through the connector, an assistant can only schedule a post after it shows you a preview with the LinkedIn account and the scheduled time, and you confirm.

The Prompt Behind the Workflow

Here is a version of the prompt Mark started with, ready to adapt for your own AI assistant:

Check content coverage: My content pillars are [YOUR CONTENT PILLARS]. Review my FeedBoss published posts from the last four weeks, scheduled posts, and drafts. Count scheduled posts as covered. Start with a one-line coverage summary.

Suggest ideas: Generate 4–5 ideas for each least-covered pillar, 2–3 for well-covered pillars, and 1 for each supporting theme. Prioritize the topics I have neglected.

Research ideas: Search [SOURCE 1, e.g. Readwise], then [SOURCE 2, e.g. Fabric, Notion, or bookmarks], followed by FeedBoss popular posts and the web. Include real source links and flag any unsourced statistics. Mention [YOUR PRODUCT/PROJECT] only when it fits naturally.

Write in my voice: Follow my FeedBoss voice profile: [TONE], [SIGNATURE HABITS, e.g. strong hooks and closing questions], [0–3] hashtags, and [LANGUAGE]. Never repeat anything already published, scheduled, or saved as a draft.

Protect my privacy: Never include confidential details about [CLIENTS/EMPLOYER], private topics such as [PRIVATE TOPICS], or figures, quotes, and sources that you cannot verify. When in doubt, leave something out and flag it for me.

Never schedule or publish: Only after I select an idea, prepare it as a post and save it to FeedBoss as a draft. Create a review task in [TASK APP] with today's due date.

Replace the placeholders with your own pillars, sources, voice preferences, and privacy rules, and adjust how many ideas you want and how often it runs.

How to Set Up Your Own AI Assistant for LinkedIn

You don't need the full setup to start. Five steps get you the core of it:

1. Connect your AI assistant to FeedBoss. In ChatGPT, add FeedBoss from the plugin directory. In Claude or another assistant that supports MCP connectors, add the FeedBoss MCP server, https://server.feedboss.ai/api/v1/mcp. Step-by-step guides: Claude and ChatGPT.
2. Set your pillars. Add your pillars and goals to the FeedBoss Authority Map, then paste the same pillars into your assistant's prompt.
3. Customize the prompt. Fill in your research sources, writing preferences, and privacy rules.
4. Create a routine. Run your assistant twice a week, by hand or through scheduled runs if it supports them.
5. Review your drafts. Check every post in FeedBoss, make your edits, and schedule the ones you approve.

For how the connector works, see the FeedBoss LinkedIn MCP server guide and the Connected Assistants docs.

You don't need several bots or a Google Doc to begin. This workflow started as a single prompt in a chat. The brief came later, once the ideas got hard to keep track of.

Frequently Asked Questions

Is ChatGPT or Claude enough to run your LinkedIn content?

ChatGPT and Claude can research and draft. On their own, they don't see your LinkedIn posts, your drafts, or your content pillars, and they can't schedule anything to LinkedIn. FeedBoss supplies those: a voice profile that learns from your posts, an Authority Map for your pillars and goals, Image Studio and Video Studio for visuals, first comments, and a content calendar. Connected to FeedBoss, Claude, ChatGPT, or another assistant can draft in your voice and schedule through FeedBoss. That is how consultant Mark Smith runs his LinkedIn.

Can an AI assistant write LinkedIn posts in your voice?

Yes, when it has your voice to work from. FeedBoss builds a voice profile from your own LinkedIn posts, and an assistant connected to FeedBoss can draft a post in that voice and check it for AI slop. In this case study, the drafts follow the consultant's FeedBoss voice profile, and he still reviews and edits each one before scheduling it.

How did a consultant go from 8 to 35 LinkedIn posts in 90 days?

Consultant Mark Smith connected his AI assistant to FeedBoss. Twice a week, the assistant checks his recent posts against his content pillars, finds sourced ideas for the pillars he has covered least, and saves the ones he picks as FeedBoss drafts in his voice. His own activity brief, dated October 28, 2026, shows 35 posts in the latest 90 days, against 8 in the 90 days before that.

Can an AI assistant connected to FeedBoss publish posts without you?

Not without your confirmation. Through the FeedBoss connector, an assistant can only schedule a post after it shows you a preview with the LinkedIn account and the scheduled time, and you confirm. In this case study, the assistant goes no further than saving drafts, and the rule is "The bot never schedules. The bot never publishes. I do."

How do you connect an AI assistant to FeedBoss?

In ChatGPT, add FeedBoss from the plugin directory. In Claude or another assistant that supports MCP connectors, add the FeedBoss MCP server, https://server.feedboss.ai/api/v1/mcp. Either way, you sign in to FeedBoss and approve access, with no API key to paste. Once connected, the assistant can draft posts in your voice, check them for AI slop, and schedule them after you confirm a preview.

How do you keep your LinkedIn content pillars balanced with AI?

Have your AI assistant count your recent posts per pillar, including scheduled posts and drafts, before it suggests anything, and ask it for more ideas on the pillars you have covered least. FeedBoss keeps your pillars and goals in its Authority Map, and an assistant connected to FeedBoss can read your posts and drafts to run that check.

Do you need several AI bots to automate LinkedIn content?

No. The workflow in this case study started with one prompt in one chat, connected to FeedBoss. The extra bots, briefing page, and Google Docs checklist came later, once ideas started getting lost in the chat.

Post More on LinkedIn Without Losing Control

The assistant finds the ideas. FeedBoss holds the voice, the pillars, the visuals, and the schedule, and every post still gets his approval. That combination took him from 8 posts in 90 days to 35. His verdict:

"Credit to the FeedBoss team for building a great tool that also plays nicely with agents."