You scroll LinkedIn. You see a 12-part career journey post with numbered lists, bullet points, and the kind of polished insight that feels like it was pulled from a template generator rather than lived experience. You don’t need to be paranoid to suspect it’s AI-generated — you just need to have noticed the pattern by now.
A new study from Pangram, an AI detection platform, confirms what many of us already feel in our bones: one out of every four long-form social media posts — over 250 words — is fully written by an AI model. That’s not a rough estimate. That’s data from over one million posts scanned across LinkedIn, X (formerly Twitter), Medium, Substack, and Reddit.
The study, released earlier this week by The Register, cuts through the noise with numbers that are surprisingly stark. On LinkedIn alone, nearly half (41%) of long-form posts were generated entirely by AI. This isn’t just about quantity, though. It’s about the kind of writing that gets buried in feeds and makes people crave the simplicity of a text message or even a straight-up image post.
For security & compliance analysts, this isn’t just an observation about content quality. It’s another signal that the environment where professionals share threat intelligence, compliance updates, and incident reports has become polluted with AI-generated noise that looks authentic but offers zero actionable insight. If your cloud security incident response playbook assumes users are reading human-written analysis, you may need to recheck your sources.
LinkedIn: Epicenter of AI-Generated Long-Form Content
Let’s be real: if you spend any time on LinkedIn, you’ve probably noticed the rise of a certain kind of post. It’s long, it’s polished, and it follows an eerily consistent structure — introduction, three bullet points that sound profound but don’t hold up to inspection, a generic conclusion about growth or resilience, and usually, at least one hashtag that’s supposed to convey vulnerability but ends up feeling like a corporate slogan.
The Pangram study found that 41% of long-form posts on LinkedIn — any post over 250 words — are fully AI-generated. Another way to look at that: for every ten posts you save in your feed as potentially useful, four of them were written by a model that has never experienced the frustration of a failed deployment or the elation of fixing a critical production bug. And here’s where it gets weird: people don’t just write these posts with AI assistance; they often use AI to generate the whole thing and then post it as if they wrote it themselves. The study found only 4.3% of long-form LinkedIn posts use AI to assist writing, meaning the vast majority are either fully human or fully AI — there’s very little in between.
Only 55.2% of long-form LinkedIn posts are actually written by humans, according to the study. That’s a stunning stat, and it’s one that hits differently when you realize LinkedIn is the de facto home for security professionals to share threat intelligence, compliance updates, and incident write-ups. If your 365 security posture depends on parsing these posts for clues, you’re walking into a minefield of plausible-sounding but entirely fake information.
X and Medium: Close Behind, But Different Patterns
If LinkedIn is the epicenter of AI-generated long-form content, X and Medium are close behind. The study found that 25% of posts on X (formerly Twitter) are fully AI-authored, with an additional 23.2% written with AI assistance. That leaves only 52.7% of posts on X actually attributed to humans.
The difference between LinkedIn and X? On X, people are more likely to use AI as a writing assistant rather than a replacement. You’ll see folks draft a tweet, run it through an AI to polish the phrasing or expand on a point, and then post it. It’s less about full-scale substitution and more about augmentation — but the sheer volume of AI-assisted content still makes it hard to separate signal from noise.
Medium, as you might expect, follows a similar pattern. Roughly one in three posts on Medium are likely written by or with AI aid, according to Pangram. The platform’s long-form format makes it particularly attractive for marketers and content farms that want SEO juice without the human effort. That’s a problem when Medium posts include security recommendations or compliance advice — the kind of information where mistakes can have real consequences for readers.
Substack and Reddit: The Oddball Data Points
Substack is an outlier in the study’s findings. Even though it’s a long-form platform, only 21.9% of posts are written by or with AI help. That’s the lowest among all platforms studied, even though it’s still almost one in five posts.
Why is Substack relatively cleaner? Probably because the platform’s paywall model creates a natural filter. If you’re going to charge for your newsletter, you had better deliver something that only a human could produce — otherwise, why would anyone subscribe? That doesn’t mean Substack is free of AI-generated content. It just means the incentives are different.
Reddit, meanwhile, is a world unto itself. Only 11.6% of top-level posts are AI-authored or assisted, and a staggering 98.1% of comments are human-written. That’s the highest percentage of human-authored content across all platforms studied.
The study explains this gap nicely: Reddit users post far fewer top-level comments than they do replies. Most of the volume on Reddit comes from discussion, and humans are still far more likely to engage in threaded discussions than AI models. The upshot? If you want human-written security insights, Reddit might be your best bet — but only if you’re willing to dig through the noise of tens of millions of comments to find them.
How Pangram Detects AI Sloppiness
Pangram’s methodology is worth a closer look, not just because the numbers are compelling, but because it shows how far AI detection technology has come. The company launched its Chrome extension in April 2026, and since then, more than a million posts have been scanned by users who opted in to share their data.
The extension scans feeds across LinkedIn, Medium, Substack, X, and Reddit in real time. It looks for structural patterns — things like sentence length consistency, punctuation overuse, and paragraph structure — that AI models tend to produce. It also uses semantic analysis to detect unnatural phrasing, repetition of ideas, and other tells that distinguish human writing from AI-generated text.
Pangram claims 99.98% accuracy in detecting AI content, with a false positive rate of just 1 in 10,000. That’s been verified by researchers at the University of Chicago and the University of Maryland, which gives it more credibility than many AI detection tools that have been debunked in peer-reviewed studies.
The company’s CEO, Max Spero, put it this way: “An internet that is completely flooded with undisclosed AI content is bleak, but we don’t believe it’s inevitable.” Pangram’s vision is simple: give users the tools to identify AI-generated content so they can make informed decisions about what they read and share. The Chrome extension is free for basic use, with a $20-per-month pro version for higher daily limits and additional features like API access and LMS integrations.
Why This Matters for Security & Compliance Analysts
You’re reading this because you care about security and compliance. You’re not here for the AI hype cycle — you’re here because a misconfigured bucket or a forgotten 365 security setting can snowball into a headline-grabbing breach faster than you’d like to admit.
So here’s the real question: when your cloud security incident response playbook tells you to scan social media for threat indicators, how do you know which posts are worth following up on? How do you distinguish between a human analyst who spotted something real and an AI-generated post designed to generate engagement (or worse, mislead)?
The Pangram study is a wake-up call. If one in four long-form posts are fully AI-generated, and nearly half of LinkedIn posts fall into that category, then the noise-to-signal ratio has tipped toward noise. That’s not just annoying — it’s dangerous when people act on AI-generated security advice that looks good but isn’t actionable.
The fix isn’t to avoid social media entirely. It’s to develop a habit of skepticism. When you see a long-form post on LinkedIn about cloud security, ask: who wrote this? What’s their track record? Is the advice specific and testable, or is it vague and generalized? And if you’re sharing it — do you know for sure that it wasn’t written by a model trained on the last three months of LinkedIn slop?
The short answer for security & compliance analysts is this: your 365 environment might be secure, but your information hygiene isn’t. And that’s a problem no amount of automation can fix.