A new study of over a million social posts found 41% of long-form LinkedIn content is fully AI-generated, and it gets 45% less engagement than human-authored posts. Here's what that means for how you should actually be using AI to write.
AI detection company Pangram Labs analyzed more than a million social media posts scraped through a Chrome extension since April 2026 and found that 41 percent of LinkedIn's long-form posts, those over 250 words, are fully AI-generated. That's the highest share of any platform measured, ahead of X at 25 percent fully AI-written and Substack at roughly 10 percent. Pangram's own CEO called the figures a conservative lower bound.
The 41 percent figure gets the attention, but the more useful finding for anyone actually posting on LinkedIn is this: AI-generated posts received an average of 45 percent less engagement than human-authored ones. The platform, and the readers on it, have already started discounting content that reads as machine-written, whether or not they can consciously identify why a specific post feels off.
Pangram also found LinkedIn usage is essentially all-or-nothing. Only 4.3 percent of long-form LinkedIn content was classified as AI-assisted or mixed, the smallest such share of any platform studied. Most LinkedIn users appear to be either writing their own posts entirely or outsourcing the writing completely, with very little middle ground where AI is used as a light editing pass over a genuinely human draft.
LinkedIn's own "Enhance Post" feature makes AI writing a single click away, built directly into the platform's own posting interface. At the same time, LinkedIn has said it deprioritizes low-quality AI content in its feed algorithm, creating a genuinely strange dynamic: the platform promotes the tool that produces the exact content type it says it's trying to suppress. In an detail widely noted in coverage of the study, LinkedIn's own announcement post about fighting low-quality AI content was itself flagged by Pangram's detection model as AI-generated.
The useful reframe isn't whether to use AI when writing LinkedIn content at all. It's understanding which part of the writing process AI should touch and which part it shouldn't. A genuinely useful way to think about this: draw a line through your content process. Above the line is thinking, the actual observation from a client call, the specific number from your own business, the position you'd defend in a room full of peers. Below the line is mechanics, tightening a sentence, restructuring a paragraph that buries its point, cutting a redundant example. AI can reasonably work below that line. It cannot generate what belongs above it, because that requires something specific and true about your business or your actual experience that a model has no access to.
Engagement isn't just a vanity number on LinkedIn specifically, because the platform's algorithm uses early engagement signals to decide how far a post gets distributed. A post generating 45 percent less engagement isn't simply performing worse, it's actively being shown to fewer people as a direct consequence, compounding the visibility gap between AI-generated and human-authored content beyond what the raw engagement numbers alone suggest.
Before publishing anything, check whether the post contains something specifically true about your business or your actual experience, not just a well-organized restatement of an idea that could have come from any company in your category. If AI was used anywhere in the process, confirm it was used for structure and clarity, not for generating the actual substance or point of view. Content that reads as genuinely specific and personally observed is both what performs better on the platform right now and what's increasingly differentiated as more of the feed fills with generic AI output.
Not necessarily. The distinction that matters is between using AI to organize, tighten, or edit content you've already thought through, versus using it to generate the underlying idea or point of view from scratch. The former is unlikely to hurt engagement; the latter is exactly what the 45 percent engagement gap reflects.
The study's own author has called it a conservative lower bound, and the sample comes from Chrome extension users who opted into sharing data, which skews toward people already interested in AI detection rather than a fully random sample of LinkedIn users. The company reports a 0.01 percent false positive rate for its detection model, which supports treating the directional finding as credible even if the precise percentage carries some uncertainty.
LinkedIn showed the highest rate of fully AI-generated long-form content in Pangram's study, but the broader pattern, AI content correlating with lower engagement, showed up across the platforms measured, including X and Medium, suggesting this isn't a LinkedIn-specific phenomenon so much as LinkedIn being the most visible current example of it.
This connects directly to our coverage of why commodity content struggles for visibility and why customers trust AI-generated marketing less than business owners do. If you want a genuine outside read on whether your own content strategy is differentiated enough to stand out, that's exactly what our content development team helps assess.
How We Work
