On Tuesday, in this most normal year of 2020, President Donald Trump woke up, hopped on Twitter, and fired off a few tweets amplifying a conspiracy theory that a cable TV host he dislikes had murdered someone decades ago. Later that day, for the first time, Twitter made the momentous decision to flag Trump for tweeting false information. Amazingly, it was in response to a completely different set of tweets.
It was a bold move, long awaited by Trump’s critics, and Twitter’s. It was also a clumsy one. With less than six months until what could be the most contentious American election since the Civil War, social media platforms have clearly still not cracked the code of political fact-checking.
What prompted Twitter to finally break the seal was a two-tweet tirade about the supposed dangers of expanding vote-by-mail during the coronavirus pandemic:
Twitter content
This content can also be viewed on the site it originates from.
Later that evening, Twitter added a note to the bottom of the tweet: a big exclamation mark and a message reading “Get the facts about mail-in ballots,” which linked to a Twitter Moment fact-check. “Trump makes unsubstantiated claim that mail-in ballots will lead to voter fraud,” read the bold heading. Beneath it came some bullet points on vote-by-mail, followed by a curated feed of tweets picking apart Trump’s various recent claims about it.
Why was this the moment when Twitter finally took action? The company says that neither of Trump’s incendiary Tuesday threads violated the platform’s rules. But the mail-in ballot tweet fell into a gray area of misleading content that the company believes is so important that it must be corrected, a Twitter spokesperson said. The platform rolled out its new fact-checking approach on May 11, and until this week had only applied it to misleading tweets about Covid-19. (For tweets that the company considers not just misleading but dangerously so, the response is deletion. Most notably, Twitter recently took down posts by the presidents of Brazil and Venezuela that promoted quack medicine.)
The Twitter spokesperson said that flagging Trump’s vote-by-mail tweet was the debut of a similar policy to protect “civic integrity” by correcting false information about voting or the census. Twitter didn’t say so, but it surely helped that this particular tweet contained one nugget of rock-solid, certifiably false information: the claim that California was sending ballots to “anyone living in the state, no matter who they are or how they got there.” In fact, ballots are going only to registered voters.
Give Twitter credit for making a gutsy decision with the public interest in mind. It’s not easy to go up against the most powerful person in the world, especially when that person is the biggest draw on your platform and spends an incredible share of his free time using it to nurse petty grievances. But the execution left much to be desired. After years of debate about whether and how to push back on misleading claims from elected officials, social media platforms are still struggling to form a coherent response to the problem.
Fact-checking requires subtle judgments about what’s true, how to tell, and whom to trust. It’s a challenge even in areas like science and medicine, where respected authorities enforce established standards of knowledge. It can be damn near impossible in politics. At Twitter, the daunting task falls to the Moments team. According to the spokesperson, these are experts on compiling social media posts and arranging them into a narrative.
No doubt that’s true, but it hardly qualifies them as master arbiters of truth. Take one of the three bullet points in the top “What you need to know” section: “Though Trump targeted California, mail-in ballots are already used in some states, including Oregon, Utah and Nebraska.” This statement is itself misleading. First, all 50 states allow some level of mail-in absentee voting. Second, Oregon and Utah are two of the five states that have transitioned to universal vote-by-mail elections. California was already on track to become the sixth before the pandemic hit. Nebraska was not. Yet the implication of the bullet point—which is as far as most users are likely to scroll, if they even bothered to click through—is that California is preparing to introduce a system already in use in Nebraska.
It’s remarkable that, after having years to prepare for the portentous moment of calling out a false Trump tweet, Twitter couldn’t even bulletproof its own fact-check. (On Wednesday, Twitter updated the page to fix that bullet point, which is now accurate.)
Which is maybe why the company outsources the rest. Social media platforms—not just Twitter but also Facebook and YouTube—have a tendency to hide behind the judgment of third-party fact-checkers. In Twitter’s case, this takes the highly stylized form of constructing a fact-check almost entirely out of other people’s tweets: shackling the platform to the 280-character format and leaving readers on their own to figure out whether the linked articles are legit. Many of the sources Twitter cites are themselves based on other articles or studies that more directly answer the question.
Tuesday’s fact-check highlights the pitfalls of this approach. One of the top authorities cited is CNN’s Chris Cillizza, a pundit notorious for his focus on horse-race political optics, not public policy. A bit farther down is a tweet from Jennifer Bendery, a reporter at HuffPost, an unabashedly liberal publication. Her source? Oregon governor Kate Brown, a Democrat, whom Bendery quotes as saying, “He doesn’t know what he’s talking about.” That may be true, but the fact that another politician disagrees with Trump hardly proves that Trump is wrong. The president was bound to accuse Twitter of bias no matter what, but this sort of sourcing makes the complaint far too easy to lodge.
The debate around fact-checking on social media platforms tends to dwell on tricky and sometimes philosophical questions of free speech and internet governance. But the problem with farming out the job is more prosaic: third-party fact-checkers are often just as full of shit as the politicians they’re debunking. Facebook has had a program since 2016 that flags posts deemed misleading by organizations like Politifact and Factcheck.org. (The policy notably exempts speech by politicians, including Trump’s election fraud post, “even if the substance of that claim has been debunked elsewhere.”) This can produce absurd outcomes. Earlier this month, the platform slapped a “Partly False Information” label on an anti-Trump ad because Politifact disputed its claim that “Trump bailed out Wall Street, but not Main Street.” Considering that each part of that sentence—bailed out, Wall Street, Main Street—is a vague metaphor, it’s not even possible to say whether it’s true or false. Yet Politifact executive director Aaron Sharockman told WIRED’s Steven Levy, “I feel real good about the rating as calling it false.”
Some of the most prominent names in the fact-checking business repeatedly make a hash of it. In 2018, Glenn Kessler of The Washington Post gave Bernie Sanders three Pinocchios (out of four) for touting a study that found that Medicare for All would lower overall health care spending by $2 trillion. The author complained to Kessler that Democrats were ignoring the fact that government spending would go up, which he considered the more important point. But the $2 trillion result was right there in his paper; Sanders’ claim was objectively true. More recently, after the first Democratic primary debate last summer, Kessler and his team took issue with Sanders’ statement that “three people in this country own more wealth than the bottom half of America.” When you add in debts, they argued, the bottom half actually has negative wealth. Somehow, this made Sanders’ claim misleading.
The point here isn’t to defend Sanders’ arguments. The point is that political truth is a difficult thing to pin down, because politics is fundamentally about convincing other people to accept your interpretation of reality. Professional fact-checkers pretend to be floating above that process, but in fact they are often participants—which is the exact position social media platforms are trying to avoid.
On that score, the early returns for Twitter are not great. A few hours after the inaugural fact-check, Trump surprised precisely no one by accusing the platform of “interfering in the 2020 Presidential Election” and “completely stifling FREE SPEECH.” The next morning, he continued the tirade: “Republicans feel that Social Media Platforms totally silence conservatives voices. We will strongly regulate, or close them down, before we can ever allow this to happen.”
We hardly need to dwell on the irony of Trump accusing Twitter of censoring him—on Twitter. Suffice it to say that the platform’s well-intentioned efforts may have stumbled into the ultimate please-nobody compromise. Trump’s soapbox remains undiminished; howls of anti-conservative discrimination draw new sustenance; and many of Trump’s critics say Twitter didn’t go far enough.
Still, let’s not write Twitter off just yet. Trump may have continued his tirade against mail-in voting, tweeting Wednesday morning that it “would be a free for all on cheating,” but he has so far stopped short of repeating the demonstrably false claim about California ballots. The president may be bombastic, but he’s also sensitive; as his morning rant makes clear, he really hates being corrected. Perhaps the threat of the little blue exclamation mark will make him, and other politicians, think twice before promoting the most blatant electoral disinformation. That’s a big if, and a small victory, but it would be a victory nonetheless.
Updated 5-27-20, 12:50 pm ET: This story has been updated to indicate Twitter updated its fact-check.