liveBook · Manning

Engagement optimization

As explained in Hidden Influences

An algorithm that selects or orders content primarily to maximize user interaction.

A feed can keep people watching by offering whatever is most likely to catch their attention next. Engagement optimization selects or orders content primarily to maximize user interaction. That goal shapes both what users see and what creators have an incentive to produce. section 10.1.2 section 7.5.1

A recommender system responding to previous actions may favor an immediate reaction over information that helps people understand public issues. This is not a new tension: traditional media also used sensational headlines to increase sales. Social media and recommender systems can, however, accelerate existing distrust and polarization. section 7.5.1

Creators adapt to platform incentives

When YouTube changed its algorithm to promote videos based on length rather than raw clicks, creators changed their videos’ structure and length to improve their chances of being recommended. More generally, engagement incentives can push creators toward extreme claims or tactics that capture attention before viewers move on. section 10.1.2

The stakes are especially high when the same incentives apply to entertainment, news, and political discussion. Engagement optimization can bring extreme views into mainstream discussion and enable misinformation or harmful content to spread. section 7.3 section 10.1.2

Hate and genocide in Myanmar

Recommendation systems helped spread hate by bringing divisive content to more people. The platform was not responsible for the full scale of suffering, but its failure to moderate content and heed warning signs contributed to conditions that helped the genocide be perpetrated. section 7.3

Choosing engagement as a target is therefore also a choice about values. One alternative under investigation is to rank content that people who usually disagree both consider valuable. Whether such bridging content can reduce polarization over time remains an open question. section 10.1.2

How Hidden Influences covers it

How deep it goes
Explains social mechanisms and consequences through platform cases, without detailing ranking implementations.
Examples
Uses YouTube creator responses and the spread of hate in Myanmar to illustrate incentives and consequences.
What you can do after
Evaluate engagement-driven recommendations in light of their incentives, social risks, and possible ranking alternatives.
How it connects
Connects recommender systems with polarization, misinformation, and content moderation.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Hidden Influences. The section links point to the parts of the book it draws on.