The Election Lies Are Already Starting on X
No one is ready for what Elon Musk’s platform could look like on Tuesday.
Last weekend, Chaya Raichik, the popular right-wing influencer who goes by Libs of TikTok, made an ominous claim to her nearly 5 million followers on X. “In Minnesota, the USPS offices are inside mosques,” she wrote. “The people who handle mail-in ballots.. What are we doing America.” She reshared a 28-minute-long video of a man circling a brick building in Plymouth, Minnesota, and accosting postal workers inside. “It’s the Islamification of the United States,” the man says. “And we have to stop it.”
Almost everything about Raichik’s post is misleading. The post office is not “inside” a mosque; they are housed in different sections of the same building. And there’s no reason to believe that the shared real estate has anything to do with the integrity of elections. In situations like this, X’s crowdsourced fact-checking feature, called “Community Notes,” is supposed to help correct the record. Users propose disclaimers for popular posts they believe are inaccurate, and if other users agree, the disclaimer gets attached to the post. Sure enough, the day after Raichik’s post, someone suggested slapping the post with a Community Note to clarify the facts, citing and linking to articles by The Minnesota Star Tribune and Minnesota Public Radio. But the note was never publicly shown on X. As of yesterday, the video had more than 1.7 million views. (Raichik did not respond to a request for comment.)
Since Elon Musk bought X four years ago, the platform has been a petri dish for rumors and propaganda. Now, having weathered a liberal exodus to other platforms, X could play a central role in the midterm elections—especially if Republicans fare poorly. X has become the go-to place for MAGA supporters to hunt for examples of voter fraud, which could be used to cast doubt on results that don’t go Republicans’ way. President Trump has endorsed this crusade—he recently announced a $1 million bounty for voter-fraud tips that lead to a conviction.
Community Notes, which even some of Musk’s critics have praised as a worthwhile program, would seem to offer a bulwark against bogus claims of vote-rigging. It’s built around a clever idea: Each proposed note gets rated “helpful” or “not helpful” by users who have disagreed about related past notes. X publishes only the notes on which the users have reached a consensus. Yet even though X has expanded the feature, bringing in AI note writers to work alongside the human volunteers, it’s proving stubbornly incapable of addressing most election-related falsehoods. If Tuesday and the days afterward bring widespread claims of rigged results, expect many of them to originate on X.
Since Musk embraced Community Notes on X, crowdsourced fact-checks have gone from a niche experiment to the default on social media. After Trump’s reelection, Meta ditched its partnerships with professional fact-checking organizations in the U.S. in favor of its own version of Community Notes. (Meta still uses professional fact-checks in other countries.) YouTube and TikTok have adopted similar features. But X’s version is the most established, and it has served as the industry’s model. In some respects, it works quite well: The system routinely corrects misleading posts about topics such as sports, pop culture, and advertising.
The problem is that it’s rare for liberals and conservatives to agree on notes that touch on the most polarizing topics—meaning that fact-checks about politics and elections rarely get approved. That has had predictable effects on the motivation of the program’s volunteers, Mike Doris, a longtime contributor to Community Notes, told me. A group chat where active members share information and sources has become a “ghost town,” Doris said. Some have given up; he is close to doing the same.
While humans might get discouraged, bots do not. To scale up the program, X has opened fact-checking to AI, creating a purpose-built agent powered by Grok and inviting outside researchers to build AI fact-checking tools of their own. It might seem ironic that the same technology that can generate falsehoods at scale is now being used to catch them. Yet the bots are more prolific, and X says that notes proposed by AI are also more likely to win approval from the masses, perhaps because they’re trained to use neutral language and cite reputable sources. What started as an experiment last year has quickly become the norm: In May, an analysis by The Indicator’s Alexios Mantzarlis found that half of all Community Notes shown on X are now written by just eight bot accounts. This month, the Georgetown University social-media researcher Renée DiResta analyzed X’s public research and found that six of those eight are controlled by the platform itself.
This has led to a Cambrian explosion in the total number of Community Notes on X. But when it comes to politics, bots haven’t fixed the fundamental problem: Users on the left and the right just can’t find common ground. The proposed note for Raichik’s post about the post office attached to the mosque was, in fact, written by an AI agent, though not one of X’s. NewsGuard, a company that tracks misinformation, found that 93 percent of false claims about the midterms on X failed to receive a Community Note. (DiResta’s own AI fact-checking bot, she told me, has proposed 81 Community Notes; only one has been approved.)
In the weeks before the election, many of these unchecked posts have involved false claims about candidates. Fake videos targeting at least three Democratic Senate candidates have been linked to Russian disinformation campaigns. One starts with a shot of a real Cleveland local-news reporter, then cuts to fabricated footage describing nonexistent claims of sexual misconduct against the Ohio Democrat Sherrod Brown by women who used to work for him. Darren Linvill, a Clemson University professor who researches foreign-interference campaigns, told me that X has taken down some of the accounts behind the posts, but the same group of influencers on the platform keeps amplifying new ones. Of the 37 viral false claims about the midterms identified by NewsGuard’s ongoing tracker, 22 started as X posts, and many of them spread to larger platforms from there. Not all of them target Democratic candidates: One debunked post claims that the Texas Republican Ken Paxton paid for a billboard that read, Pedos ain’t that bad.
Fact-checks were never a cure-all for social-media lies. Many conservatives distrusted mainstream fact-checking organizations, and some research has shown that Community Notes are more broadly believed. A spokesperson for X told me that the addition of AI note writers has roughly doubled the number of Community Notes published in the past year. X, Meta, and other social-media companies also have rules against certain types of election misinformation, which they enforce through other means. For instance, X said earlier this month that it suspended more than 13,000 accounts linked to foreign-influence campaigns, and at least one viral post showing the satirical Paxton billboard now carries a “manipulated media” label.
Still, the limits of crowdsourced fact-checking might become obvious in the days to come. AI bots in particular are likely to struggle to counter fast-flying fraud claims on Election Day, DiResta pointed out. “The AI does not know what is happening on the ground unless a human has written it somewhere retrievable,” she said. In 2020, the last national election when Trump was in the White House, the president’s claims that the election had been stolen gained traction on social media and helped fuel the Capitol insurrection on January 6, 2021. Back then, Twitter applied numerous fact-checks and warning labels to posts by Trump and his allies making unsupported claims that the vote had been rigged. It’s hard to envision the platform doing that this time around.
