How do social media algorithms amplify outrage and loneliness?

Social media algorithms amplify outrage because they rank for engagement, and emotionally charged content gets more engagement. The mechanism is well-documented: a 2025 study by Smitha Milli and colleagues, published in Nature and widely cited since, compared engagement-based ranking algorithms to chronological feeds and found that engagement-optimized algorithms systematically amplified more emotionally charged, out-group-hostile, and politically divisive content. Users in the study consistently preferred the chronological feed when shown both, even though engagement-based algorithms were better at predicting what they would click on. The implication: what gets engagement and what users want to see are not the same thing.

The mechanism is straightforward once you see it:

• Angry content gets more clicks. Outrage triggers physiological arousal, which makes people more likely to click, share, and comment. This is true across the political spectrum and across cultures.
• Comments extend time-on-site. A divisive post generates argument threads. Argument threads are the most engagement-dense content on most platforms. Each comment is a chance to surface another ad.
• Sharing amplifies distribution. People share content that triggers strong emotions — including negative ones. The algorithm rewards shares because shares extend the post’s reach, which extends its ability to generate engagement, which extends its ability to generate ads.

The Milli study’s specific finding: engagement-based ranking amplifies partisan and out-group-hostile content at roughly 1.5–2x the rate of chronological ranking, even when the underlying user preferences would have selected less divisive content. The platform’s optimization metric (engagement) and the user’s actual preference (less outrage) diverge in a way that the platform’s metrics do not capture.

Loneliness is a different mechanism but it shares the same root cause. The National Academies’ 2024 consensus report on social media and adolescent mental health found that frequent, intensive social media use is associated with higher loneliness scores in young adults, with the strongest correlations in the heaviest users. The mechanisms are still being studied, but the most consistent finding is that:

• Comparison-heavy content lowers mood. Curated highlight reels of other people’s lives produce social comparison, which most users experience as downward social comparison (feeling worse about their own life).
• Quantity of interaction is not the same as quality. A thousand passive likes is not the same as one close conversation. Heavy social media use can produce a sense of connection without the relational substance that buffers loneliness.
• Displacement. Time spent scrolling is time not spent on in-person interaction, exercise, sleep, or activities that build durable relationships.

What is not settled:

The causal direction is still being worked out. The strongest evidence suggests a two-way relationship: heavy social media use increases loneliness for some users, and lonely users turn to social media more intensively, which reinforces the cycle. The Germano 2026 paper documents the dynamic feedback loop — the algorithm learns from engagement, the engagement pattern shapes what the algorithm surfaces, the user’s behavior adjusts — and shows that the loop is self-reinforcing without an external correction.

The why-do-i-get-sidetracked answer covers the related attention-side mechanism — how the same phone that holds your reminder also serves a stream of designed interruptions — and the practical interventions that work.

The honest version: the algorithm is not neutral. It is optimizing for the metric it was given, and that metric happens to reward outrage. The loneliness effect is partly an unintended consequence of the same system. None of this requires individual users to be making mistakes — it is structural.