Plate Nº 58 · recorded November 11, 2026
Neuroscience & MindReported finding
X's Algorithm Pushes Posts That Conflict With Your Values, Study Finds
A PNAS study of 715 X users found the platform's algorithm promotes content that clashes with users' stated values, likely driven by angry replies. Democrats saw the widest gap.
By Elena Vasquez3 min read660 words
In brief
- The study tracked 715 American adult X users via a custom browser extension.
- Researchers found a negative correlation between users' stated values and the values amplified in the For You feed.
- The algorithm boosted posts about order, security and tradition while suppressing posts about caring, dependability and nature.
- Replies, not likes or reposts, drove the misalignment — X weights replies heavily in ranking.
- The gap was wider for Democrats; the study is observational and cannot prove causation.
X's recommendation algorithm systematically promotes posts that conflict with users' explicitly stated values, according to research published in the Proceedings of the National Academy of Sciences. The study of 715 American adults found a negative correlation between what people say they believe and what the platform's "For You" feed serves them.
The likely culprit? Replies. Users who argue with posts they dislike generate engagement, and X's algorithm weights replies heavily, creating a feedback loop that floods feeds with provoking material rather than agreeable content. The effect hits left-leaning users hardest.
How the study worked
Ziv Epstein of MIT and Michael S. Bernstein of Stanford University led a team investigating the gap between people's stated values and the values embedded in algorithmically amplified content. They built the analysis on the Theory of Basic Human Values, a framework developed by social psychologist Shalom H. Schwartz that identifies nineteen foundational motivations recognized across cultures — from societal security and tradition to openness to change and caring for others.
The researchers recruited 715 American adults who actively used X, the platform formerly known as Twitter. Each participant installed a custom browser extension that captured posts from two distinct feeds:
- The Following feed: the raw, unranked inventory of posts from accounts the user chose to follow.
- The For You feed: the algorithm's personalized selection of what the platform thinks the user should see.
An AI model scanned the text of these posts and scored them against the nineteen values in the psychological framework. By comparing the raw inventory to the curated feed, the team could quantify which moral and social themes the platform boosted or suppressed.
What did the algorithm actually do?
The results surprised the researchers. Users generally follow accounts aligned with their personal values — the unranked inventory reflected the beliefs participants reported in surveys. The algorithm's curation flipped that dynamic.
At the highest level of categorization, the algorithm showed a slight preference for individualist values over collectivist ones. Across the platform, it consistently amplified:
- Posts expressing a need for conservation and order — rule conformity, societal security, and tradition.
And it suppressed:
- Posts expressing self-transcendent values — caring for others, dependability, and preservation of nature.
The misalignment appeared on both sides of the political aisle, but the gap ran wider for Democrats.
Why do angry replies matter?
In a follow-up analysis, the team tracked how participants interacted with posts for 24 hours after the initial data collection, sorting interactions into likes, reposts, and replies. The pattern was stark. Likes and reposts generally matched users' personal values — intentional endorsements. Replies did not.
The act of replying aligned far less with users' beliefs, consistent with confrontation. Among Democrats, replies were on average slightly misaligned with stated values. Because X's algorithm treats replies as a strong engagement signal, arguing with a post can effectively request more of it.
A final analysis added political context. Examining platform-wide engagement metrics for posts in participants' feeds, the researchers found globally popular content tended to align more closely with Republican values. Right-leaning users, in other words, encounter naturally aligned content without relying on the confrontational reply loops that drive the feed for left-leaning users.
How solid are the findings?
Preliminary caveats apply. The study is observational — researchers tracked natural behavior without manipulating the algorithm, so it cannot definitively prove that confrontational replies are the sole mechanism behind the misalignment. Future work will need controlled experiments to isolate cause and effect, plus larger samples to confirm the feedback loop across demographics.
The findings still raise hard design questions. If ranking systems learn primarily from friction and argument, they will keep building environments that contradict the preferences of the people using them.
The paper, "Value misalignments in X's feed algorithm is a reflection of value tensions in engagement," was authored by Ziv Epstein, Farnaz Jahanbakhsh, Tiziano Piccardi, Axel Peytavin, Isabel Gallegos, Shardul Sapkota, Dora Zhao, Johan Ugander, and Michael S. Bernstein.
via psypost.org (Source)
More from Elena Vasquez
Show full bio
Correspondent covering money, funding and market shifts around general science news at SciBeat.
384 articles
Nearby plates
- How You Use Social Media Matters More Than How Long, Study Finds
- Short-Video Feeds Hit Users With ADHD Hardest, Study Finds
- Nudge or Boost? Both Cut Student Electricity Use, Study Finds
- 78% of Psychology's Masculinity Scales Measure Only the Negative
- AI Companions May Cause Lasting Psychological Harm, Study Finds