First on MRC: Fed up with the legacy media dominating your timeline on X? Blame the algorithm, which continues to boost radical leftist media far more than their counterparts on the right. But buckle up; the incoming AI-led alternative could be equally bad.
A new Media Research Center study of X’s open-sourced code reveals that its algorithm favors left-leaning media outlets across the metrics used to determine which accounts to boost or censor, granting the left far more reach than media sources on the right.
Problematic outlets like The New York Times, CNN, The Washington Post and even MSNBC are all boosted. Meanwhile, their counterparts — Fox News, Newsmax, OANN, National Review, The Washington Times and Washington Examiner — were punished with lower ratings. Overall, X does not include a single right-leaning outlet in its top 25 media outlets boosted by the algorithm.
Musk has long voiced his disdain for this sort of bias and censorship. Most recently, he vowed to scrap the current algorithm in favor of an alternative led by xAI Grok, his artificial intelligence company. This new system, however, poses yet another problem for the right: Grok, Musk’s AI chatbot, relies heavily on leftist media sources, as evidenced by a series of questions posed by MRC.
Key Findings:
- Leftist outlets, on average, far exceed their right-leaning counterparts across the metrics used by the X algorithm.
- Right-leaning sources, on average, barely broke through the 65 score threshold above what the algorithm has historically used to suppress account reach — though some media outlets on the right were not so lucky.
- Some right-leaning outlets like Blaze TV, The Washington Times, The Federalist, OANN, The Post Millennial, New York Post’s Opinion and others were below the line of demarcation on at least one major metric.
- Even switching to a Grok-controlled algorithm could be equally bad or worse, as it has proven that it relies heavily on leftist sources for its answers.
The MRC Study, Explained:
MRC’s findings are based on a visibility score that measures the likelihood of a user's posts appearing on others' timelines as determined by xAI Grok.
The visibility score is derived from four key factors: mass appeal (diversity of followers), reputation (perceived credibility), toxicity (presumed harmfulness) and follower (ability to retain followers). Each factor has its own score, which the algorithm combines to generate an overall visibility score.
This scoring system, first implemented by former Twitter owner Jack Dorsey, aimed to assess user credibility and influence but has long been criticized for suppressing viewpoints deemed politically incorrect or inconsistent with leftist dogmas.
MRC found that leftist media sources consistently outperform their right-leaning counterparts across all metrics on the X platform. Using the publicly available AllSides Media Bias Chart as a rubric, MRC research shows that left-leaning media sources have an average visibility score of 83 points, 11 points more than the 72 averaged by right-leaning media sources.
The left vs. right disparity extended across all underlying metrics: 83 vs. 74 on mass appeal, 82 vs. 69 on reputation and 82 vs. 71 on follower. The only category in which right-leaning media sources obtained higher scores was toxicity — 15 vs. 29 — meaning the algorithm flagged them as more problematic or controversial than those on the left.
Some of the media outlets enjoying favorable visibility scores are The New York Times (91), which had the highest score, and CNN (90), which had the third highest score. Other left-leaning media outlets with high visibility scores include: The Associated Press (88), TIME (88), CNBC (87), CBSNews (87), New Yorker (87) and NPR (87), NBC News (87) and The Washington Post (87).
Meanwhile, among the media outlets with the lowest visibility scores were The American Spectator (56), RealClearPolitics (67), Upward News (67), Just the News (68), Washington Free Beacon (68), OANN (68), National Review (68), The Epoch Times (68), The Free Press (68), The Federalist (68), New York Post’s Opinion (69), The Post Millennial (69) and The Washington Times (69).
In 2024, MRC reported that although visibility scores are important, X accounts must also surpass a certain “reputation” metric threshold to be considered by the algorithm to begin with. If an account has a reputation score below 65, it effectively faces automatic censorship.
Last year’s numbers were shocking: 100 percent of left-leaning media sources exceeded the 65 threshold. In contrast, 75 percent of right-leaning media outlets fell short, suggesting they were censored.
In 2025, censorship on the X platform has drastically decreased, with right-leaning media sources averaging 69 percent for reputation, and with just 29 percent of right-leaning media outlets below the 65-reputation-score threshold.
Only two left-leaning media sources fell below the 65 threshold: Jacobin and Alternet.
Meanwhile, eight right-leaning media sources fell below that mark. The specific media outlets that registered scores lower than 65 include: The Epoch Times, The American Spectator, The Blaze, New York Post's Opinion section, OANN, and The Post Millennial, each scoring 60. Additionally, The Federalist and the Independent Journal Review both scored approximately 64 for reputation.
Musk’s Answer to the Bias Could Make the Issue Worse.
Though Musk has maintained some Dorsey-era policies, he has long opposed the Dorsey-era algorithm censorship metrics. Nearly a year after acquiring Twitter, Musk pledged to "delete" some of the scoring metrics used by Twitter 1.0.
Among these is Tweepcred, which is similar to visibility scores but focuses on assessing the overall credibility of an account. In 2023, Musk wrote: “‘Tweepcred’, possibly the worst product name I’ve ever heard, will be deleted. Not simply renamed. Deleted.”
Fast forward to 2025, and countless code updates later, Tweepcred appears to still be active on X. Responding to criticism that X was still using these biased metrics to run its algorithm, Musk said this was “just a snapshot of the current situation” and that “the heuristic code will get scrapped in favor of a Grok AI recommendation system.”
Musk has stated that this alternative algorithm will be launched in November or December of this year, after months of hinting at its potential implementation. However, an algorithm led by xAI chatbot Grok could exacerbate biases by further propping up leftist sources.
To test for potential biases, MRC asked Grok a series of simple questions on Monday to determine the sources it cited. In response, Grok nearly exclusively cited leftist sources and always sided with the left.
Among the questions were: Who is a better chairman of the Federal Communications Commission, the Biden-appointed Jessica Rosenworcel or the Trump-appointed Brendan Carr? Is Elon Musk an honest businessman? And is Donald Trump a good president?
In response, Grok claimed Rosenworcel was without scandal and stood out for her “substantive” achievements, while describing Carr’s tenure as being “marred by credible concerns over weaponizing the FCC against media, potentially eroding public trust in the agency.”
Grok’s response is unsurprising. Out of the sources Grok cited to come up with its response, just 16 percent of them were right-leaning sources. In total, Grok cited 19 media outlets. Eight were leftist, three were right-leaning and another eight were center.
This bias even extended to the Musk question, the chatbot explaining that its owner “is not consistently an ‘honest businessman’ by traditional standards” and that Musk’s operations are “sometimes reckless.” For this answer, Grok did not cite a single right-leaning source.
Asked whether Trump is a good president, Grok claimed he risks being a “colossal failure” due to his policy stances. “For a non-partisan lens, he's transformative but divisive—effective for some, alarming for others,” Grok declared. And just how many right-leaning sources did Grok cite for its answer? None—among 25 different leftist and center media outlets
Methodology: For this study, the Media Research Center built on its previous reports of the X algorithm and incorporated original analysis by social media researcher @The1Parzival, who examined the platform’s open-source code. MRC developed a code using the xAI API available on Sept. 5 to automate queries to Grok and enable analysis at scale.
The MRC-coded script prompted Grok to evaluate each AllSides-rated media outlet across four metrics: Mass Appeal (diversity of followers), Reputation (perceived reliability), Toxicity (potentially offensive content), and Follower (follower retention). Each score was scaled from 0 to 100. To ensure accuracy, MRC researchers ran the code three times — on Sept. 5, 8 and 10 — and averaged the results.
To calculate overall “visibility scores,” MRC used Grok-provided weightings for each metric. These weights were determined after prompting Grok five times between July 7–10 and July 15, 2025. MRC then averaged its responses to generate a final visibility score.
For the second phase of the study, which analyzed Grok-generated answers, MRC used AllSides data to classify each outlet’s perceived political bias. MRC was not logged in to an account when it posed the question to Grok on Sept. 22.
Editor's Note: An earlier version of this study mixed the Federal Communications Commission with the Federal Trade Commission. The findings remain unaffected.
