New Stanford Algorithm Derank Divisive Political Posts on X

Stanford University researchers have found that a web-based algorithm that reorganises what people see on X (formerly known as Twitter) can reduce political hostility online. The study, conducted during the 2024 US election, tested whether pushing extreme or highly partisan posts lower in the feed would influence how users felt about people they politically disagreed with. According to the report, the tool did not block or hide posts; it only moved inflammatory content further down. Participants reportedly showed warmer attitudes toward the other political side after using the tool for just ten days.

Third-Party Tool for X Can Change People's Cross-Party Attitudes

In a newsroom post, Stanford University detailed the findings from the study conducted on a group of participants who agreed to use this tool on their X feed. The methodology and other details were shared in the study itself, which has been published in the Science journal. Notably, the researchers highlighted that the algorithm was developed without any cooperation from X or by deleting or hiding posts.

Before the start of the study, the team developed a browser-based tool that operates as a layer above X's existing algorithm. The experiment involved roughly 1,200 volunteers representing a mix of political views across the US. Throughout the 10-day study period, these participants used X normally while the tool quietly reordered their feeds. By the end of the trial, both liberal and conservative users reportedly expressed more positive feelings toward their political opponents than those in the control group.

When participants opened X, the tool scanned the posts in their timeline and identified ones containing partisan hostility, extreme rhetoric or antidemocratic statements. Instead of removing them, which could raise censorship concerns, the system simply moved those posts lower so they were less prominently visible. The study also highlights the impact of information captured by immediate attention, since the control group, which saw the divisive political posts unrestricted, was impacted more negatively.

Stanford's study provides empirical evidence that even minor adjustments to the ranking, without changing what content is available, may shift how people perceive one another politically. The researchers highlight that the goal is not to suppress views but to reduce how frequently users encounter the most inflammatory material at the top of their feeds.

However, the experiment offers insight into how ranking changes on X could affect online behaviour, but the findings are limited to the specific period and user group involved. As per the Stanford team, the trial ran only for ten days during a highly charged political moment and with a relatively small participant pool.

The tool also does not evaluate whether similar results would appear in less polarised contexts, on other platforms, or among users who consume significantly more political content. It additionally does not attempt to reduce misinformation; it only reorders posts containing hostility or extreme partisan framing.