The Ubiquity of AI Content Across Social Networks: A Focus on LinkedIn

By Max Spero | Published July 9, 2026

Two months ago, we introduced our Chrome extension designed to mitigate the proliferation of "AI slop" on social media. This tool empowers users to analyze posts in real-time as they browse, identifying AI-generated text so they can consciously decide where to allocate their attention.

At its core, Pangram is a research-driven organization. We don't just build top-tier detection algorithms; we actively monitor the prevalence and associated risks of synthetic content. While analyzing social media is significantly more complex than auditing research papers, Amazon reviews, or news articles, it is a vital endeavor because these platforms represent the highest-volume conduits for AI-generated text.

To facilitate this research, we implemented an opt-in feature allowing users to anonymously share their scanning statistics. The following report synthesizes the data collected over the initial months.

---

📌 Executive Summary: Key Findings

• Longform Vulnerability: AI is most prevalent in longer posts.
• LinkedIn Dominance: The platform is a primary hub for AI-generated "slop."
• Structural Variance: Primary posts are more likely to be AI-authored than comment threads.

---

📈 The Impact on Longform Content

AI-generated text is pervasive across all studied platforms, with an overall average AI detection rate of $13.8\%$. However, the distribution is not uniform; the length of the content plays a critical role.

On the majority of platforms (four out of five), there is a positive correlation between content length and the likelihood of AI generation. Specifically, one in four longform items (defined as those exceeding 250 words) were flagged as entirely AI-generated, or $\approx 25.72\%$.

Platform Comparison Table

The Substack Exception: Unlike other platforms, Substack showed a flat or slightly inverse relationship between length and AI usage. Even so, it isn't immune; $21.9\%$ of its content was flagged as either AI-assisted or fully AI-generated.

This trend mirrors the broader shift toward synthetic writing seen in other professional spheres, such as newspaper op-eds.

---

💼 LinkedIn: The Epicenter of AI Slop

LinkedIn stands out as the most AI-saturated environment in our dataset. Despite representing only one-third of all scanned items, it accounted for a staggering $62\%$ of all flagged AI content.

Why LinkedIn?

It is paradoxical that users are more comfortable using AI to represent their professional identity—where their real name and career are at stake—than they are on anonymous or casual platforms. This is likely bolstered by platform-native tools:

• The built-in Write with AI Enhance post button explicitly encourages synthetic writing.

The community is starting to push back against this "slop." In a twist of irony, a LinkedIn executive recently announced that the platform would use an internal algorithm to detect and downrank AI content—yet the announcement itself was generated by AI.

---

🤖 Reddit and the "Composition Effect"

Reddit presented a unique case. While it had the highest volume of scans ($36.7\%$), it had one of the lowest AI rates ($4.4\%$). This is explained by the composition effect:

Analysis of Reddit vs. LinkedIn

• Reddit: Top-level posts are $5.25\times$ more likely to be AI-generated than replies, regardless of length.
• LinkedIn: Top-level posts are $1.35\times$ more likely to be AI-generated than comments. (Interestingly, when controlling for length, LinkedIn comments are slightly more likely to be AI).

The Moderation Gap: Reddit's anti-spam measures are effective at killing high-volume AI bot replies. However, they struggle with top-level posts. Because these posts are lower in volume, they bypass rate-limiting filters while maintaining a much higher impact on the audience.

---

🛠 Method and Data Collection

The data was aggregated via the Pangram Chrome extension. Users who opted into the research program provided anonymous telemetry on the posts they scanned.

{
  "data_collection": {
    "source": "Pangram Chrome Extension",
    "method": "Opt-in anonymous telemetry",
    "metrics": ["AI-generated", "AI-assisted", "Human-authored"],
    "timeframe": "First few months post-launch"
  }
}