Нейросеть для текста генерирует 41% контента в LinkedIn

Pangram published its first large-scale study on the prevalence of AI-generated content on social media on July 9, 2026. Based on analysis of over a million posts, users are seeing far more AI-generated text than they realize.

Figures by platform

Pangram installed a Chrome extension on volunteers' computers and tracked content during real browsing over a two-month period. The results:

• LinkedIn: 41% of long-form content (250+ words) is entirely AI-generated
• X (formerly Twitter): 33% of long posts (25% entirely written by AI, 23% AI-assisted)
• Reddit: 10% of long-form content is AI-generated
• Substack: 10% of long-form content is AI-generated

Long-form posts (over 250 words) across all platforms are more often generated by neural networks than short posts. Top-level posts contain more AI than comments.

The first honest study

"This is not what has been studied before — how much AI-generated content people actually see during everyday use of these platforms," said

Max Spero, CEO of Pangram.

Until now, platforms and analysts could only guess at the scale of the problem. Pangram conducted the first full-scale study based on real user behavior, not artificial samples.

Platform responses

LinkedIn officially acknowledged the problem and stated it is working to reduce low-quality automated content. The company emphasized: "Professionals come to LinkedIn to hear the voices of real people."

X and Substack did not respond to requests for comment.

What this means

The problem of AI spam is moving from theory to fact with concrete figures. If neural networks generate nearly half of the content on the largest professional network, this indicates ecosystem degradation and the need for strict filters against low-quality synthetic text.