---
title: X Algorithm Shifted Users’ Political Views and the Effect Did Not Reverse
description: European researchers ran a randomised experiment on 4,965 X users, finding the algorithm promoted conservative content and demoted traditional news sources.
author: Darie Nani (Editor-in-Chief)
date: 2026-02-18T23:04:12.000Z
updated: 2026-02-26T17:55:07.246Z
canonical: https://www.sovereignmagazine.com/article/x-algorithm-shifted-users-political-views-and-the-effect-did-not-reverse
image: https://cdn.nanimediahouse.com/d4_acs3jsq0.jpg
categories: Business
content_type: News
region: United States
publication: Sovereign Magazine
---

X’s algorithmic ‘For You’ feed pushed users toward more conservative political positions over seven weeks, according to a randomised experiment published in [Nature](https://www.nature.com/articles/s41586-026-10098-2) on 18 February 2026. Switching the algorithm off did not undo the shift.

## European researchers ran a controlled field experiment

Researchers from Bocconi University in Italy, the University of St. Gallen in Switzerland and the Paris School of Economics randomly assigned 4,965 active US-based X users to one of two groups during the summer of 2023. One group used X’s default algorithmic ‘For You’ feed. The other used a chronological feed showing only posts from accounts they followed, displayed in the order they were published.

Lead author Germain Gauthier and his team measured political attitudes before and after the seven-week period using surveys on policy priorities, attitudes toward public figures and views on the Russia-Ukraine conflict.

## The algorithm promoted conservative content and demoted news organisations

Users on the algorithmic feed were 4.7 percentage points more likely to prioritise policy issues favoured by US Republicans, including crime, inflation and immigration. They were also 7.4 percentage points less likely to view Ukrainian President Volodymyr Zelenskyy positively and scored higher on a pro-Russian attitude index.

The mechanism is visible in the content itself. Conservative posts were 20% more likely to appear in the algorithmic feed than the chronological one. Liberal posts were 3.1% more likely to appear. Traditional [news organisations](https://www.sovereignmagazine.com/article/why-europe-is-caught-between-american-and-chinese-tech-giants) appeared 58% less often, while posts from political activists appeared 27.4% more frequently and entertainment accounts 21.5% more.

## Turning off the algorithm did not reverse the effect

The study’s most significant finding is the asymmetry. Switching from a chronological feed to the algorithmic feed changed political attitudes. Switching from the algorithmic feed back to chronological did not produce a comparable reversal. Users who had been exposed to the algorithm followed more right-leaning accounts during the experiment, and those following patterns persisted after the algorithmic feed was removed.

Previous experiments on Meta’s platforms found no measurable political effects from algorithmic feeds. The difference suggests that X’s recommender system operates in a structurally distinct way, actively reshaping the information environment rather than simply reordering existing preferences.

## The findings arrive amid EU enforcement action against X

The European Commission fined X EUR 120 million on 16 January 2026 for violating the Digital Services Act, the first penalty issued under the regulation. The Commission found that X’s blue checkmark system deceived users about identity verification and that the platform obstructed researcher access to public data.

Separately, French prosecutors raided X’s Paris offices on 3 February 2026 as part of an investigation into alleged [algorithm manipulation](https://www.sovereignmagazine.com/article/what-does-sovereign-tech-actually-mean-for-europe) and the spread of illegal content. The probe, which now involves Europol, expanded to include allegations related to child sexual abuse material, sexually explicit deepfakes generated by X’s AI chatbot Grok and Holocaust denial.

Under the DSA, platforms with more than 45 million EU users must assess how their algorithms amplify [systemic risks](https://www.sovereignmagazine.com/article/european-digital-stack-can-europe-build-its-own-eurostack-for-digital-sovereignty) and offer users the option of non-personalised feeds. The Nature study provides the first large-scale experimental evidence that X’s algorithmic feed does not merely filter content but actively shifts political attitudes in one direction, with effects that outlast the algorithm itself.

## Further Context

**Q: How do social media algorithms affect political opinions?**
Algorithms select and rank content based on predicted engagement rather than editorial judgement. On platforms where engagement correlates with emotional intensity, this creates a feedback loop that amplifies divisive or partisan material. The Nature study on X is the first randomised experiment to demonstrate a measurable, directional political shift caused by a specific platform’s algorithm, distinguishing it from earlier studies on Meta that found no comparable effect.

**Q: What are the risks of algorithmic content curation for democracy?**
Algorithmic feeds concentrate attention on a narrower range of sources and viewpoints than chronological feeds. When the algorithm simultaneously demotes professional journalism and promotes activist accounts, users receive a less diverse information diet. The persistence of changed following patterns after the algorithm is removed suggests these systems can permanently alter how users access information, even after the algorithmic intervention ends.

**Q: How is the EU regulating social media algorithms?**
The Digital Services Act, which took full effect in February 2024, requires very large online platforms to conduct systemic risk assessments of their algorithms and provide users with non-algorithmic feed options. The European Commission has opened formal investigations into X, Meta, TikTok and other platforms. The European Centre for Algorithmic Transparency supports enforcement by analysing how recommender systems function and what content they amplify.