Editor’s Note (Aug. 13, 2024): This piece has been updated to feature the X data sooner in response to reader feedback and to enhance clarity. The data itself remains unchanged.
The Media Research Center has unveiled evidence showing that the X algorithm is biased against right-leaning media outlets while actively boosting left-leaning media outlets.
An MRC study of the X algorithm found that 100 percent of left-leaning media outlets were assigned favorable ratings from X employees. This allows these outlets to reach a broader audience, gain more followers and, as a result, attract more readers. In contrast, the algorithm did not offer the same advantage to media outlets on the right.
This bias reflects the views of a radical remnant from the Jack Dorsey era, with Twitter 1.0 employees working tirelessly to undermine X owner Elon Musk’s pledge to protect free speech—a promise that prompted his purchase of the platform.
MRC researchers selected media outlets to review based on the media bias chart from AllSides. The chart featured 62 media outlets (30 left-leaning, 23 right-leaning and nine centrist).
As detailed in an earlier MRC report, the X algorithm considers “reputation scores” crucial in determining which accounts to boost. Anything below 65 out of 100 is considered too detrimental and is less likely to be promoted by the X algorithm.
By applying the same criteria that X’s algorithm uses to decide which accounts to promote or suppress, the study uncovered the following key takeaways:
- All Left-leaning Media Outlets Enjoy Favorable “Reputation Scores.” MRC’s analysis found that all outlets listed as “left” and “lean left” by AllSides received scores above 65, with an average score of 82.64.
- Meanwhile, Nearly Two-Thirds of Right-Leaning Media Outlets Are Deamplified. In stark contrast, 74 percent of right-leaning media outlets received scores lower than 65, with an average of 56.52. This gives left-leaning media outlets a staggering 19.08-point advantage on average over right-leaning media outlets.
- The Left Is Favored Across All Metrics. Beyond “reputation,” X uses three other metrics—mass, toxicity and follow—to determine a general “visibility” score. In all four metrics, left-leaning media outlets consistently received more favorable treatment, even when they ran afoul of the X algorithm’s rules of conduct.
- An Ongoing Bias Problem. These findings are consistent with MRC’s previous studies, which showed how X favors Democrats in Congress over their Republican counterparts, and how the platform also favors Vice President Kamala Harris over Sen. J.D. Vance (R-OH), former President Donald Trump’s running mate.
Below is a more detailed explanation of the data.
The “reputation” scores run from 0 to 100 and weigh heavily in the algorithm’s calculation for which accounts to recommend to others, according to Grok’s answers to MRC. This score is also called “Tweepcred.”
Grok told MRC that a score of 65 out of 100 is the “minimum” score required for a user to be recommended. “Accounts with higher Tweepcred scores are more likely to have their posts seen by a larger audience, while those with lower scores may have their posts deprioritized in the platform's algorithm,” the chatbot claimed.
Socialist magazine Jacobin and infamous leftist outlets HuffPost, Mother Jones, MSNBC and DemocracyNow! are among those deemed reputable by the staff at X who seem to be rebelling against Musk’s vision of a free-speech platform devoid of election interference.
In contrast to these left-leaning media outlets, X skewered the right, deboosting nearly 74 percent of the right-leaning media outlets reviewed by MRC. That is 17 out of 23 or three-fourths of the right-leaning media outlets.
Responding to these findings, MRC Free Speech America Vice President Dan Schneider said, “We have seen over and over that there remains a radical remnant within X fighting against Elon Musk. This latest report is shocking. It shows that X’s radicals treat right-leaning news outlets even worse than NewsGuard does.”
In sum, of the 62 media outlets reviewed by the MRC, the algorithm deemed that those on the left — including the self-described socialist magazine Jacobin and MSNBC — are more appealing, more reputable, less toxic and retain followers at a higher rate on average when compared to their counterparts on the right. New York Post, The Daily Wire and Blaze Media are among those media outlets on the right.
The Data, Explained: To determine each account’s score, MRC prompted Grok with tailored questions so that the AI chatbot reviewed the platform’s backend to unveil the scores. This methodology was first disclosed by an anonymous researcher on X known as “@The1Parzival.”
According to data reviewed by the MRC, X uses four core metrics to calculate each user's “visibility” score. These metrics are “Mass Appeal” (diversity of followers), “Reputation” (purported reliability), “Toxicity” (potentially offensive content or perceived harmfulness) and “Follower” (follower retention).
According to @The1Parzival, the algorithm uses this score to determine whether to amplify or suppress posts from accounts using these metrics. Posts will receive greater reach depending on the specific score awarded by X. For a more in-depth explanation of the methodology, revisit our first study of the X algorithm.
Let’s Run Some Numbers: MRC used the media bias chart by ratings firm AllSides that classifies media outlets on their perceived bias on a left-to-right scale. The evidence showed that X anointed left-leaning media outlets with an average and favorable, visibility score of 82.64/100. In contrast, X punished right-leaning outlets with an average score of 63.56, giving the left a 19.08-point advantage on average.
An explanation for how X granted these outlets such favorable ratings is found in how the platform determined the individual score for each metric. X’s favoritism for leftist outlets is even more blatant in these metrics.
With an average score of 56.52, the right is far more likely to be censored than the left. Some of these media outlets include well-respected media outlets The Washington Times, The Federalist, Fox News, The Daily Wire and The Daily Caller.
The outlets described by AllSides as centrist are also safe from the X algorithm. Outlets like the BBC, Forbes, NewsNation and Reuters, enjoyed an average reputation score of 74.44 out of 100, which is a 10-point advantage over outlets on the right.
X Turns a Blind Eye to Leftist’s Toxic Posts: The X algorithm grants high “toxicity” scores to media outlets known for their subjectively divisive and hostile posts. Left-leaning media outlets received an average toxicity score of 26.33, a grossly low number when compared to the 47.60 average score granted to right-leaning media outlets. That is a 21-point disparity.
Accounts with conveniently low toxicity scores include that of The New York Times (10/100), which on Oct. 17 incorrectly blamed Israel for an airstrike against a Gaza-based hospital.
Jacobin, the socialist magazine, enjoyed a relatively low toxicity score of 40, despite sharing posts hailing radical philosopher Karl Max and praising socialism. On June 18, Jacobin was community-noted after falsely linking Noam Chomsky to Marx. Moreover, on July 19, Jacobin suggested Republicans were to blame for the failed assassination attempt on Trump.
MSNBC, unmatched by Jacobin and The Times, and with a toxicity score of 20, still features posts about the demonstrably false claims about the Russian collusion hoax. “Watch @chrislhayes ask authors Michael Isikoff and David Corn about the 'pee tape,’” MSNBC tweeted on March 15, 2018.
Musk Is Not to Blame: This MRC study follows two earlier studies suggesting that X’s algorithms favored the left. In one study, X promoted Harris over Vance (R-OH), Trump’s running mate—likely helping Harris reach a broader audience months ahead of her elevation as the Democratic nominee for president. Another earlier study showed how X boosted congressional Democrats’ posts while suppressing those of Republicans ahead of the 2024 election.
X Employees Leading Anti-Musk Revolt: Upon purchasing X (then Twitter), Musk pledged to elevate free speech as a cornerstone of the platform. However, several MRC studies, most recently on its algorithms, suggest that there is a remnant of Jack Dorsey-era employees at X working tooth and nail to undermine free speech and maintain the censorship apparatus.
In May, Musk responded to Sen. Mike Lee’s (R-UT) concerns that X’s algorithms cunningly reduced the visibility of specific accounts. “How long will it take to get rid of the stage-five clingers at X—those who still periodically throttle conservatives?” Lee asked.
The tech mogul pledged to investigate the issue, saying, “Well, neither conservative [sic] nor progressives should be throttled. The point is to have an even playing field. I will investigate.”
It isn’t immediately clear whether Musk followed up with his pledge. However, the MRC studies suggest that the algorithms still disproportionately target Republican politicians, right-leaning media outlets and Biden-Harris opponents.
MRC’s Schneider hailed Musk’s past free speech pledge but cautioned him to watch his back, as the ghost of the Jack Dorsey-owned Twitter 1.0 is still haunting the algorithms.
Former MRC Free Speech America Intern Christian Baldwin contributed to this report by calculating the X data.
Former MRC Free Speech America Intern Christian Baldwin contributed to this report by calculating the X data.
METHODOLOGY: Social media researcher @The1Parzival first unveiled how to get Grok to scrounge X’s backend to determine how the algorithm determines which accounts to censor and which ones to amplify.
Between June 7 and June 25, MRC researchers used Parzival’s prompts to ask Grok to ascertain “visibility” scores for the media outlets listed in AllSides’s Media Bias Chart.
The questions were specifically tailored so Grok would only report the essential criteria that would determine whether or not a specific account on X is pushed/recommended to other users on the platform (“Mass Appeal,” “Reputation,” “Toxicity,” and “Follow”).
Grok calculated numerical value scores between 1 and 100 for each account. MRC researchers then calculated the average scores for media outlets across all four criteria used by the algorithm to determine X’s “visibility” scoring process.
To determine how X weighted each of the four categories in determining a “visibility” score, MRC researchers prompted Grok with eight tests between July 1 and July 2 to assess how each of the metrics were weighted to uncover which was considered most important when determining whether or not an account would be boosted or deboosted.
Grok repeatedly provided information with minor variations on how each category was weighted to determine a final “visibility” score.
MRC researchers then took the average weighted percentages across the eight tests to deduce a mathematical formula. When the averages were computed, the categories were weighted as follows:
Mass Appeal (38 percent), Reputation (30 percent), Toxicity (21 percent), and Follow (11 percent). “Mass Appeal” and “Reputation” scores were consistently in the top two spots in measures of importance across the eight tests. As explained by Grok, X uses “Tweepcred” to determine which accounts to recommend to others.
Conservatives are under attack. Contact your representatives and demand that Big Tech be held to account to mirror the First Amendment while providing transparency, clarity on hate speech and equal footing for conservatives. If you have been censored, contact us using CensorTrack’s contact form, and help us hold Big Tech accountable.
