In late 2024, my LinkedIn presence was inconsistent. I had ideas but no system. Posts happened when time allowed, which meant they often didn’t happen. I knew the problem wasn’t motivation. It was operations. So I treated it like one.
The challenge
Creating LinkedIn content that actually sounds like you is harder than it looks. The obvious bottleneck is time. But the deeper one is structure. Without a repeatable system, every post starts from scratch: the brainstorm, the framing, the draft, the edit. That’s not a content problem. It’s a workflow problem.
I also had a quality issue. AI-assisted content was fast but impersonal. The posts that performed best were the ones where I’d added something specific like a personal observation, a real example, a genuine opinion. The challenge was building a system that preserved that human input without making the process slow.
The approach
I started with a framework from Creator School’s AI LinkedIn Content Factory course, which introduced the concept of a structured AI-assisted content workflow. The initial version used a custom GPT and a Google Sheet to track ideas and manage the production pipeline. It worked, and I learned a lot from building it.
Over the following year I rebuilt the system from the ground up. In early 2026 I migrated from the custom GPT to Claude, which gave me more flexible project-based context and better writing quality. The bigger change was structural: I separated the work into four dedicated conversations: personal posts, LinkedIn posts for Tuesday and Thursday, LinkedIn comments, and content planning. Keeping these separate means each conversation builds context over time without one type of work polluting the others.
The personal input problem, the one I’d struggled with since the beginning, got solved with voice notes. Instead of typing out what I wanted to add to a post, I record a short voice note directly into Claude. It’s faster, more natural, and the output actually sounds like me rather than a polished version of a brief I wrote. That single change made the biggest difference in quality.
How the system works now
The content planning conversation is where the quarter starts. I export the data from the previous quarter, feed it into Claude, and generate a full content plan for the next quarter. For example, for Q2 2026 that produced 35 ideas across my content pillars, enough to run for the full quarter without brainstorming from scratch each week.
When I’m ready to draft a post, I open the relevant conversation, paste the idea, and add a voice note with my actual perspective. Claude drafts the post. I refine. For images, I take Claude’s description of the post’s message and generate the visual in Gemini, either by describing what I want or by pasting the post text directly and letting Gemini interpret it. The output has been consistently on-brand.
For comments, the workflow is lighter. I paste the post I’m responding to, give a short note on my angle, and Claude drafts the comment. I give feedback over time to train the voice. The goal isn’t automation. It’s reducing the friction between having something worth saying and actually saying it.
Results
Content creation time dropped from roughly an hour per post to about 15 minutes. That’s across the full cycle: ideation, drafting, and refinement. Running at two strategic posts per week plus regular commenting, the system handles a posting cadence that would have been unsustainable without it.
Profile views tripled over the period I ran the system consistently. The quality improvement was measurable too: posts with my personal voice input consistently outperformed posts where I skipped that step. The data reinforced the design decision. The human input layer isn’t optional. It’s what makes the system work.
What this is really about
The LinkedIn Content Factory isn’t a content hack. It’s an operations problem approached as one. The question I started with wasn’t “how do I use AI to write LinkedIn posts.” It was “what does a sustainable content workflow actually look like, and what breaks if I don’t build it properly.”
The answer involved separating concerns, reducing friction at the human input stage, building in measurement, and iterating on the design. That’s the same process I apply to any operational system. The fact that it’s about LinkedIn content is almost beside the point.
Alex Gita 