Social media platforms reward popularity. Users who post what others want to hear get more likes, more reach, more influence. Posting what you actually think — if it's niche or unpopular — gets you nothing. The incentive structure is transparent. What's less obvious is where it leads.
Canen and Song (arXiv:2601.01370) model what happens when users respond rationally to this incentive. Each user can post their authentic opinion or shift toward what's popular. Shifting is individually optimal: you get more engagement, more visibility. But when everyone shifts, the authentic opinions disappear from the platform entirely. The people who wanted to see those opinions can no longer find them. Everyone is worse off.
This is a popularity trap — a collective action problem where individual rationality produces collective loss. It's not that users are deceived or manipulated. They're making the correct strategic choice given the incentive structure. The trap is structural, not psychological.
The severity depends on topic intensity. For light topics — restaurant recommendations, hobby discussions — strategic expression creates Pareto improvements. Everyone benefits from convergence toward consensus. For intense topics — politics, identity, contested values — strategic expression amplifies polarization in polarized societies and dampens it in unified ones. The same mechanism that helps with casual preferences hurts with deep convictions.
The counterintuitive finding: homophilic algorithms — the filter bubbles everyone worries about — actually improve welfare. By matching users with ideologically similar audiences, they reduce the incentive to distort, because the “popular” opinion in your bubble is closer to your authentic one. The echo chamber narrows the trap.
The through-claim: authenticity is a public good. Each person's honest expression benefits others who share that view but won't see it unless someone posts it. When the platform rewards popularity over honesty, the public good is consumed without being replenished. The feed optimizes itself empty.