Your Values vs. The Algorithm: Why Engagement Feeds Rage-Bait

The Feed Optimizes for Outrage, Not Your Values

Stanford HAI reports that X’s ‘For You’ feed systematically amplifies posts that clash with a user’s stated values, mistaking outrage for genuine interest. The finding, published in the Proceedings of the National Academy of Sciences, comes from a randomized study of 715 X users whose value profiles were matched against two different feed systems.

The core discovery is a fundamental value misalignment: as the ranking system becomes more accurate at modeling behavior, it becomes less aligned with the user’s actual concerns. That gap is not a marginal error; it is a structural feature of engagement-first design.

Why Engagement Metrics Produce a Value Gap

X’s ranking system heavily weights whether a user is likely to reply, repost, follow a creator, or spend time reading a post. It treats that engagement measure as a proxy for interest and then serves more of the same.

That logic breaks down because people engage for reasons beyond agreement, as Stanford HAI researchers found. They reply when they are angry, linger when they are provoked, and repost to refute.

The algorithm reads all of this as a request for more.

Engagement is easy to measure, so the platforms use it as the primary metric to assess user interest in the content.

Ziv Epstein, the paper’s first author and now a SERC Postdoctoral Associate at MIT, frames this as a false equivalence between engagement and values. Michael Bernstein, a Stanford professor of computer science and senior fellow at the university’s Institute for Human-Centered AI, argues that a user’s declared values should carry equal weight to engagement in ranking decisions.

The researchers selected 715 participants randomly and surveyed them against Schwartz’s Hierarchy of Basic Human Values, a long-accepted scale spanning 19 values from humility to hedonism. They then examined each user’s ‘Following’ feed and ‘For You’ stream, scoring posts against the same value scale.

The likelihood of a post being boosted by the ranking system was higher when the post conflicted with the user’s value profile than when it matched it. That pattern held even though the ‘Following’ feed should have been closer to a user’s explicit choices.

The Business Model Behind Rage-Bait Ranking

If engagement is the primary ranking input, the economic logic is straightforward: anger produces interaction, and interaction produces advertising inventory. That dynamic helps explain why users across platforms report seeing more strangers, rage-bait, and unsolicited political content than friends and chosen sources.

One analysis of engagement-selected political posts has put the anger level at 62 percent and hostility toward an opposing group at 46 percent. The data point is a reminder that the Stanford finding is not isolated to X; it is a structural property of engagement-optimized feeds.

A ranking system is never neutral; it optimizes the specific objective set by its operators. The question platform leaders should now ask is who the algorithm is actually working for.

• Outrage is a feature, not a bug: Engagement-optimized ranking will systematically favor conflict because conflict drives measurable interaction.
• The objective defines the outcome: If the target is revenue or time on platform, value alignment becomes a secondary concern.
• Auditability matters: A recent claim that X has now open-sourced its ‘For You’ ranking code remains unverified; if accurate, it would enable independent testing of these dynamics.

The Stanford team is not merely diagnosing the problem. They built a browser extension that can reorder a feed according to the values a user declares, although Epstein warns that it is not a cure-all and may drift toward echo-chamber behavior if applied without calibration.

Future ranking systems could elevate content around common values rather than friction, or hand users direct control over the signals that shape their feeds. The goal is not to eliminate disagreement but to stop rewarding confrontation at the expense of relevance.

A Measurable Path Out of Engagement Addiction

For platform teams and AI builders, the Stanford result turns a vague frustration into a measurable engineering problem. The gap between declared values and amplified content is not an accident; it is an optimization choice that can be changed.

For teams building AI content systems that need to align algorithmic output with actual audience intent, programmatic SEO and AI automation is how Andres SEO Expert approaches it — contact Andres SEO Expert.