Why Platform-Side Fixes Can't Work

LinkedIn, X, and Medium have each deployed aggressive interventions to combat AI-generated content. These fixes are structurally doomed because every major platform simultaneously builds AI creation tools and fights the content degradation those tools produce. The contradiction is not a bug. It is the business model.

LinkedIn: Selling the Disease and the Cure

LinkedIn shifted from social graph to interest graph, deploying 360Brew — a 150-billion-parameter model that scores content against months of individual user behavior. The impact on creators has been dramatic: visibility dropped 47% in 2025, engagement fell 39%, follower growth declined 42%.

But LinkedIn still sells Premium subscribers AI writing tools for drafting posts, rewriting profiles, and composing messages. It labels AI-generated images via C2PA metadata tags but has no mechanism to label AI-generated text. The Collaborative Articles experiment — AI questions receiving AI answers — was retired in October 2025. The platform is fighting the symptoms of a problem its own products create.

X Has a Deeper Problem

If LinkedIn's issue is content quality, X's is existential. Bot prevalence estimates range from 9% to 64% of accounts. Average engagement dropped to 0.12% in 2025, down 48% year-over-year. Ad revenue collapsed from $4.46 billion in 2021 to roughly $2.5 billion in 2024. The EU fined X $140 million, citing that its paid verification system made authenticity harder to judge. Meanwhile, Grok's AI is built into the post composer, encouraging the very content X's head of product calls "the scourge of X."

The Substack Comparison

Medium and Substack provide a natural experiment. Medium's open-publishing model makes it vulnerable: an estimated 47% of posts are AI-generated. Substack's subscription model creates natural defense — only 10% of top newsletters show significant AI use. Readers paying for specific writers disincentivize anonymous mills.

This serves as evidence for a broader point. Platforms most resilient to AI contamination share a common feature: economics that align incentives with authenticity rather than volume. Building channels where the incentive is verified depth — not viral reach — extends this principle to the agent era. See the parent position on the AI content crisis for the full argument.