Table of Contents
Do not index
Should you stop using AI to write your LinkedIn posts? Every founder and agency owner I talk to has asked some version of that question since Pangram Labs published its study of more than a million social posts. My answer is no. You should stop using AI to think. The writing was never the real problem, and the operators who understand that distinction are about to get the easiest competitive window LinkedIn has offered in years.
The numbers set the scene. "More than 40% of LinkedIn's long-form posts were classified as fully AI-generated," according to Tech Times' coverage of the Pangram study, the highest share of any platform measured. The market is already punishing it. "AI-generated posts on LinkedIn received an average of 45% less engagement than human-authored posts," per the same report. LinkedIn is now the most machine-written professional feed on the internet, and both readers and the algorithm have learned to discount it.
If you are an agency owner between $200k and $2M in revenue, a ghostwriter charging $5k to $30k per month, or a founder using personal-brand content to drive deal flow, this is your operating environment now. Your prospects scroll past feeds that are two-fifths machine-written to find someone worth trusting. The bar for standing out just dropped, and your competitors are the ones who lowered it.
This is not for everyone. If you run a volume shop selling posts by the batch, if your model depends on shipping 200 posts a month to $500 clients, nothing here changes your math. Detection and discounting are built into that model. Skip this if you are optimizing for output instead of outcomes, because the study describes your ceiling, not your opportunity.
What the LinkedIn AI content study actually found
The most useful number in the study is not the 40%. It is the 4.3%. Only 4.3% of LinkedIn's long-form content was classified as AI-assisted, meaning a mix of human and machine. Almost everyone treats AI as all-or-nothing. They either hand the entire post to a machine, thinking included, or they refuse to touch the tools at all. The middle position, where a human supplies the ideas and the judgment while the machine handles mechanics, sits nearly empty. That empty middle is where the advantage lives, and almost nobody on the platform occupies it.
A second finding deserves more attention than it got. AI posts underperformed by 45% on average, except in the leadership and inspiration category, where AI posts outperformed human ones by 75%. Sit with that for a second. The one category where machine writing wins is the category that was already interchangeable before AI arrived. Generic inspiration was commodity content five years ago. AI just produces the commodity faster. If your content lives in that category, the study is not indicting your tools. It is indicting your category.
And in a detail that should humble every content team on the platform, LinkedIn's own announcement post about fighting low-quality AI content was itself flagged by Pangram's model as AI-generated.
Where AI belongs in a founder content system
Here is what I would actually do, and it starts with what I call the Outsourcing Line. Draw one line through your content process. Everything above the line is thinking. The observation from yesterday's client call. The take you would defend in a room full of peers. The number from your own operation, like the retainer that renewed at month 14 or the post that produced three qualified sales conversations. Everything below the line is mechanics. Tightening sentences. Restructuring a draft that buried its point. Cutting the second example that repeats the first. AI never crosses the line upward. The moment it supplies the take instead of sharpening yours, you have joined the 41% and inherited the 45% discount that comes with it.
In practice the workflow is unglamorous. The founder records a three-minute voice note or types an ugly draft with the real story in it. The machine compresses, structures, and cleans. Then the human makes the final pass and puts back the phrasing only they would use. The test at the end is simple. If a reader who knows you cannot point to one sentence only you could have written, the post fails, no matter how clean it reads.
This also changes how you measure the work. Engagement on a feed this saturated tells you less than it ever has, which is why I keep arguing that the real signal lives in pipeline conversations rather than impressions. I broke this down in a piece on measuring LinkedIn success beyond your analytics dashboard, and a 45% engagement discount on machine content makes that argument stronger, not weaker. When nearly half the posts around yours are discounted on sight, raw engagement comparisons stop meaning much. What a post opens in your pipeline is the number that survives.
The strategic read is this. LinkedIn is splitting into two markets. One trades in commodity attention, where 41% of the supply is machine-made and the price of a reader's trust keeps falling. The other trades in demonstrated thinking, where supply is shrinking because most operators quit doing the hard part. Every month you spend on the right side of the Outsourcing Line, your content compounds against a field that is actively diluting itself. The operators who win the next two years on LinkedIn will not be the ones who used AI least. They will be the ones who never let it do their thinking.
