Hathaway Field Notes
Signal

The engagement trap: why social platforms can't self-correct

Social media platforms are locked in a structural feedback loop that makes meaningful reform nearly impossible without external forcing functions.

Concepts feedback loops·reinforcing loops·local vs global optimization·incentive alignment

There’s a common narrative that social media companies could “just fix” their algorithms if they wanted to. This misunderstands the system they operate in.

The core loop

The engagement-advertising feedback loop works like this:

  1. Engagement drives revenue — More time on platform = more ad impressions
  2. Emotional content drives engagement — Outrage, fear, and tribal identity generate 3-5x more interaction than neutral content
  3. Algorithms optimize for engagement — ML systems surface what works, not what’s healthy
  4. Content creators adapt — Producers learn what the algorithm rewards and optimize accordingly
  5. The content environment shifts — The median post becomes more emotionally charged over time
  6. Users acclimate — Baseline expectations shift; moderate content feels boring
  7. Return to step 1 — With a higher emotional baseline

This is a classic reinforcing loop with no natural balancing mechanism.

Why internal reform fails

Any team that reduces engagement to improve content quality is directly reducing revenue. In a public company, this creates an immune response:

  • Q1: “Responsible AI” team launches content quality initiative
  • Q2: Engagement metrics dip 2-3%
  • Q3: Revenue forecast adjusted downward
  • Q4: Team reorganized, initiative “evolved”

The system protects itself.

The only viable interventions

Interventions must come from outside the feedback loop:

  • Regulation — Changes the cost function (fines > lost engagement revenue)
  • User behavior shifts — Structural migration to different platforms
  • Advertiser pressure — Brand safety concerns that make toxic engagement unprofitable

The platform cannot fix itself for the same reason a thermostat can’t decide to change its own set point. The optimization target is the problem, and the optimization target is set by the business model.

The human benefit test

If we score social media against the question “does this technology improve the human system?”, the answer is increasingly complex. The connection utility is real. The information access is real. But the engagement optimization layer sits on top of those genuine benefits like a tax — extracting attention and emotional energy to serve the advertising loop.

The technology works. The business model doesn’t — at least not for the humans in the system.