Over the past 24 hours, both Twitter and Facebook have slapped a relentless stream of labels on misleading posts about voting and election results — most prominently from the president himself.
Misinformation researchers have praised some of their efforts. But a much larger question hangs over each of these decisions: Do these labels even work?
Some preliminary studies have found that warning labels on fake news stories can have the unintended effect of making readers more willing to share unlabeled stories — even if those turn out to be untrue as well. Still, other surveys have suggested that in search, when websites are rated as containing unreliable information, the majority of people are less likely to share news from those sites. The impact Facebook, Twitter and YouTube's labels have had over these last few days, however, remains a mystery.
"There's been no research on the effectiveness of this," said Aimee Rinehart, U.S. deputy director of the misinformation project First Draft News.
As election results came in Tuesday and Wednesday, researchers applauded Twitter's strategy — particularly the most heavy-handed labels from the social media platform, which require users to click through an interstitial and prevent them from sharing or engaging with the post.
"Twitter's been the fastest to actually append effective labels and to actually hide objectionable content," said Emerson Brooking, a resident fellow at the Atlantic Council's Digital Forensic Research Lab.
The company took forceful action against tweets from President Trump falsely claiming Democrats were trying to "STEAL the election," hiding it behind a label that said some of the content in the post was "disputed and might be misleading about an election or other civic process."
Facebook's approach has, meanwhile, received more middling reviews. It also labeled President Trump's posts about election stealing, but those labels appeared beneath the post and did not limit users from sharing or engaging with the underlying message. "Final results may be different from initial vote counts," Facebook's label reads. Facebook has also begun applying the same label to all posts from both Biden and Trump, despite the fact that only the Trump campaign has prematurely declared victory, leading some to wonder whether the labels might confuse Facebook users about who's telling the truth.
Researchers have expressed similar concerns about YouTube, which has affixed small, subtle "information panels" with factual information directly under videos and search results related to the election, an approach that experts said could confuse people trying to differentiate between misinformation and reputable news sources. "Too often, YouTube has tried to get away with doing the minimum, and this is another instance of that," said Paul Barrett, deputy director of the New York University Stern Center for Business and Human Rights.
YouTube spokesperson Ivy Choi said in a statement that YouTube "remains vigilant with regards to election-related content in this post-election period."
"In this post-election period, our teams are continuing to work around the clock to quickly remove content misleading people about voting or encouraging interference in the democratic process, raise up authoritative news publishers in search results and 'watch next' panels, and reduce the spread of harmful election-related misinformation," Choi said. "On Election Day, we removed several livestreams for violating our spam policies, and our election results information panel is prominently surfaced above search results and under videos about the election."
Nina Jankowicz, a disinformation fellow at the Wilson Center, said Twitter's labels are likely the most effective because they provide "friction," requiring users to click through warnings with relevant context about why the posts are inaccurate. That process can slow people down and force them to reconsider what they're interacting with.
But Twitter has mainly reserved such muscular actions for the president's tweets. When Trump aides, including White House press secretary Kayleigh McEnany and Eric Trump, prematurely declared Trump had won Pennsylvania on Wednesday afternoon, Twitter affixed a less aggressive label underneath their posts. "Official sources may not have called the race when this was tweeted," the Twitter label reads.
"As votes are still being counted across the country, our teams continue to take enforcement action on tweets that prematurely declare victory or contain misleading information about the election broadly," said a Twitter spokesperson. "Our teams continue to monitor tweets that attempt to spread misleading information about voting, accounts engaged in spammy behavior, and tweets that make premature or inaccurate claims about election results."
One key question is whether any of this is slowing the spread of misinformation. Researchers are doubtful. "The information we have so far regarding the effectiveness of labeling generally is it doesn't really reduce the spread of content," Brooking said.
Some of Trump's most devoted followers have started to copy and paste the tweets that Twitter hides, according to the Election Integrity Partnership, which created an even bigger mess for Twitter to deal with. As of last night, "some of them were cleaned up [and] some of them weren't," said Kate Starbird, a researcher with the EIP.
In some cases, groups like the Real Facebook Oversight Board, a collective of academics and activists focused on accountability at Facebook, reported that misinformation continued to go viral on multiple platforms, even after it got labeled. "#StopTheSteal went from Twitter and transferred over to Facebook with millions of views," said Shireen Mitchell, a member of the group and founder of Stop Online Violence Against Women, in a statement. "It was labeled as inaccurate but it was still spread. It's the perfect example of digital voter suppression."
Beyond the impact information labels have on the spread of those posts, there are even trickier questions to answer, like do the labels actually convince people not to believe the underlying message? Do the labels unintentionally create a sort of Streisand effect, driving people to the original posts purely because they have labels? How much can a single misinformation label really accomplish now when, for four years, the president has been using social media to seed the idea that voter fraud is rampant in America, entirely without objection from Facebook or Twitter? By Election Day, was it already too late?
Getting to those answers would require more sophisticated polling of social media users, which so far, doesn't exist.
On Facebook, at least, that could change. Earlier this year, Facebook announced it would be working with a 17-person independent team of researchers to study the platform's impact on the 2020 election. Among the areas of study was the role Facebook plays in the spread of political misinformation. But it's unclear if the researchers will specifically look at whether people are actually processing the labels in ways that limit the misinformation's spread or, at the very least, helps deter people from believing in the misinformation themselves. Neither Facebook nor the lead researchers on the project responded to Protocol's request for comment.
Of course, it's noteworthy that tech companies are making an effort on this front at all. It's more than they could say they did in 2016 when misinformation went entirely unchecked by every social platform. And there are limits to what these companies alone can do. On Wednesday, even as Facebook and Twitter tried to correct the record on the president's claims about election stealing, his campaign was sending the same message to voters by email — where no one could say he's wrong.
Update: This story was updated at 4:52 p.m. PT to include statements from Twitter and YouTube.
