Skip to content

User preferences

User preferences are central to social media design but present a fundamental tension: what users say they want often differs from what they actually interact with. This gap between stated preferences (survey-based, explicitly stated wants) and revealed preferences (inferred from behavioral engagement) has profound implications for algorithmic design and content visibility.

Stated vs. revealed preferences

Stated preferences — What users report wanting when explicitly asked through surveys, user controls ("Do Not Recommend"), or preference settings. Generally more reflective and deliberate.

Revealed preferences — What users actually engage with through likes, shares, comments, watch time, and click-through. The basis for most current engagement-based ranking algorithms. May reflect habitual, reactive, or passive behavior rather than deliberate choices.

Research shows these frequently diverge:

  • Engagement, User Satisfaction, and the Amplification of Divisive Content on Social Media — Twitter users rated political tweets selected by the engagement-based algorithm as less valuable than tweets in the reverse-chronological baseline (−0.18 SD, p = 0.005), despite the algorithm prioritizing these tweets
  • Users were less likely to want to see tweets like those selected by the engagement algorithm, especially political tweets, despite spending more time on the engagement-based timeline

The preference-engagement gap

This gap arises from several mechanisms:

  • Algorithmic optimization — Engagement metrics (clicks, dwell time, shares) don't directly measure user welfare or satisfaction; they measure platform metrics
  • Emotional capture — Engagement-based algorithms amplify emotionally arousing content (anger, moral outrage), which users may not deliberately prefer but which captures attention
  • Revealed preference bias — Habitual or impulsive engagement doesn't necessarily reflect considered preferences
  • Network effects — Users engage with content that their friends find important or engaging, not necessarily what they personally prefer

Design implications

Recognizing the stated/revealed preference gap enables alternative algorithmic approaches:

Engagement, User Satisfaction, and the Amplification of Divisive Content on Social Media demonstrates that a ranking system based on users' stated preferences (obtained through survey): - Reduces exposure to partisan and hostile content - Increases happiness and reduces anger, sadness, and anxiety - Maintains overall engagement and user satisfaction - Does not require explicit user controls or separate preference elicitation (in principle; the experiment used surveys to establish the counterfactual)