Table of Contents
Do not index
Should I still be using AI to draft my LinkedIn posts? Founders ask me this defensively, the way you ask a question you have already decided the answer to. LinkedIn just gave a clearer answer than I could.
Use AI for the mechanics and never for the judgment. The platform that once offered to write your posts has withdrawn the offer, and that is a product decision worth reading carefully rather than celebrating.
LinkedIn retired its AI writing assistant and replaced it with a proofreading tool. Writing in Inc., Soren Kaplan described the swap plainly. The new tool "corrects your grammar and hands the writing back to you. The message to users is clear: bring your own thinking, and let the AI clean things up." Kaplan also cites Substack's CEO describing the correct use of AI as handling everything "except the hard part," where the hard part is having an idea worth sharing.
Most of the commentary read this as LinkedIn declaring war on AI content. It is not that. LinkedIn shipped an AI feature. It just moved the feature to a different position in the workflow, and the position it chose is exactly where thinking stops and typing begins. That is a much more interesting claim than a ban, because it is a claim about what a platform believes is scarce.
Generation was never the expensive part
Here is the thing operators keep getting backwards. The cost of content was never the drafting. Drafting was always the cheapest hour in the process. The expensive part is knowing which of the 40 things you could say this month is the one that changes how a specific buyer thinks about their problem. Generation tools attacked the cheap hour and left the expensive one untouched, then everyone was surprised when output tripled and results did not move.
Kaplan draws out the difference between two users, and it maps precisely onto something I see in ghostwriting engagements. One person prompts and posts. The other treats the draft as work handed in by an intern and rewrites 70 percent of it, keeping the structure and replacing every judgment. Same tool, opposite outcome. The second person is doing the job. The first person has automated the part that was never the bottleneck and left the bottleneck fully intact.
That gap is also the difference between two kinds of ghostwriter, and it is why the market is about to sort itself out ungently. A writer who extracts your actual thinking and shapes it is doing something a model cannot do, because the raw material lives in your head and has to be pulled out through conversation. A writer who fills a calendar is competing with a free tool that fills calendars faster. One of those keeps a $5k to $30k per month retainer through 2027. The other does not.
Where to draw the line in your own operation
What I would do is run what I call the Thinking Split. Take every step in your content process and sort it into mechanics or judgment, then let AI touch only the first pile.
Mechanics is grammar, formatting, transcription, tightening a paragraph that runs long, generating five headline variants from a claim you already made, and checking whether you have used the same word four times. Judgment is what the claim is, whether it is true from your experience, which of your clients it is aimed at, what you are willing to be wrong about in public, and what you are choosing not to say. The split is not about how much AI you use. Plenty of operators run AI across most of their workflow and still produce work that could only have come from them, because they held the line in the right place.
The test for whether you drew it correctly is simple. Read the post and ask whether anyone else in your category could have published it. If a competing 3 person agency could have posted the identical piece under their own name without changing a word, you outsourced judgment and did not notice. That is the failure mode, and it is quiet. Nothing looks wrong. The post reads fine. It just does not do anything, and neither do the next 200.
This matters most for founders and agency owners between $200k and $2M in revenue whose deal flow depends on being distinguishable from four other firms with the same service page. It matters for anyone whose positioning rests on being a practitioner rather than a commentator, which is the whole argument for leading with practitioner credibility instead of thought leadership language. Skip this if you are producing genuinely commodity content where the goal is coverage rather than distinction. And if you are still selling volume as the deliverable, this article will not change your model, because your model requires generation to be the valuable part.
The trajectory question is about what happens as generation cost keeps falling toward zero. Everything a model can produce becomes worthless as a differentiator, not gradually but on the timeline of a single product release, which is what just happened to anyone whose LinkedIn presence was built on the assistant that no longer exists. What holds value is the thing that had to be extracted from a person who has done the work. Operators who spent the last two years building a process around generation now own a process for producing something free. Operators who built around extraction own something that gets more valuable every time the cost of the alternative drops.
